diff --git a/backend/api/criteria/context.py b/backend/api/criteria/context.py
new file mode 100644
index 00000000..0ae7235d
--- /dev/null
+++ b/backend/api/criteria/context.py
@@ -0,0 +1,104 @@
+"""Canonical citation and imported-answer context helpers.
+
+This module deliberately has no database or web dependencies so prompt builders,
+CSV/RIS importers, and tests all use the same matching and formatting rules.
+"""
+from __future__ import annotations
+
+import re
+from typing import Any
+
+
+def resolve_source_value(row: dict[str, Any], header: str | None) -> Any:
+    """Resolve a configured source header without silently guessing."""
+    if not header:
+        return None
+    if header in row:
+        return row[header]
+    normalized = _normalize_key(header)
+    matches = [
+        value for key, value in row.items(
+        ) if _normalize_key(str(key)) == normalized
+    ]
+    return matches[0] if len(matches) == 1 else None
+
+
+def _normalize_key(value: str) -> str:
+    return re.sub(r'[^a-z0-9]+', '_', value.strip().casefold()).strip('_')
+
+
+def normalize_answer_text(value: Any) -> str:
+    """Normalize formatting only; semantic synonyms are intentionally excluded."""
+    return re.sub(r'\s+', ' ', str(value or '').strip()).casefold()
+
+
+def match_answer_label(raw_value: Any, options: list[dict[str, Any]]) -> dict[str, Any]:
+    """Match a raw value to an option label and retain the raw audit value."""
+    raw = '' if raw_value is None else str(raw_value)
+    if not raw.strip():
+        return {'status': 'blank', 'raw': raw, 'label': None, 'option_id': None}
+    normalized = normalize_answer_text(raw)
+    for option in options:
+        label = str(option.get('label') or '')
+        if normalized == normalize_answer_text(label):
+            return {'status': 'matched', 'raw': raw, 'label': label, 'option_id': option.get('id')}
+    return {'status': 'unmatched', 'raw': raw, 'label': None, 'option_id': None}
+
+
+def resolve_existing_human_value(
+    row: dict[str, Any], item: dict[str, Any],
+) -> dict[str, Any]:
+    """Resolve and classify an explicitly mapped human answer/parameter value."""
+    header = item.get('answer_column')
+    raw = resolve_source_value(row, header)
+    options = [*(item.get('answers') or []), *(item.get('options') or [])]
+    if options:
+        return match_answer_label(raw, options)
+    if raw is None or not str(raw).strip():
+        return {'status': 'blank', 'raw': '' if raw is None else str(raw), 'value': None}
+    return {'status': 'matched', 'raw': str(raw), 'value': raw}
+
+
+def format_title_abstract_context(
+    row: dict[str, Any], citation_fields: dict[str, Any],
+) -> str:
+    title = resolve_source_value(row, citation_fields.get('title'))
+    abstract = resolve_source_value(row, citation_fields.get('abstract'))
+    excluded = {citation_fields.get('title'), citation_fields.get('abstract')}
+    pairs = []
+    for header in citation_fields.get('l1_include') or []:
+        if header in excluded:
+            continue
+        pairs.append(f"{header}: {resolve_source_value(row, header) or ''}")
+    other = '; '.join(pairs) or '(none configured)'
+    return f"Title: {title or ''}\nAbstract: {abstract or ''}\nOther fields: {other}"
+
+
+def format_item_context(item: dict[str, Any]) -> str:
+    parts = [str(item.get('question') or item.get('name') or '').strip()]
+    if item.get('description'):
+        parts.append(f"Description: {item['description']}")
+    if item.get('context'):
+        parts.append(f"Context: {item['context']}")
+    if item.get('critical_prompt_additions'):
+        parts.append(
+            f"Critical prompt additions: {item['critical_prompt_additions']}",
+        )
+    if item.get('unit_instructions'):
+        parts.append(f"Units: {item['unit_instructions']}")
+    if item.get('calculation'):
+        parts.append(f"Calculation: {item['calculation']}")
+    if item.get('selection_mode'):
+        parts.append(f"Answer cardinality: {item['selection_mode']}")
+    options = [*(item.get('answers') or []), *(item.get('options') or [])]
+    if options:
+        parts.append(
+            'Answer-specific guidance:\n' + '\n'.join(
+                f"- {option.get('label')}: {option.get('context') or '(none)'}" for option in options
+            ),
+        )
+    if item.get('trigger', {}).get('all'):
+        parts.append(
+            f"Visible only when trigger conditions are met: {item['trigger']['all']}",
+        )
+    return '\n'.join(part for part in parts if part)
diff --git a/backend/api/criteria/models.py b/backend/api/criteria/models.py
index 390c4eaa..905d25b0 100644
--- a/backend/api/criteria/models.py
+++ b/backend/api/criteria/models.py
@@ -52,6 +52,9 @@ def conditions_are_unique(self) -> Trigger:
 
 
 class CitationFields(StrictModel):
+    # Nullable while loading legacy criteria; save/use preflight requires both.
+    title: NonEmptyText | None = None
+    abstract: NonEmptyText | None = None
     l1_include: list[NonEmptyText] = Field(
         default_factory=list, max_length=100,
     )
@@ -77,7 +80,9 @@ class ScreeningQuestion(StrictModel):
     id: CriteriaId
     question: NonEmptyText
     answers: list[ScreeningAnswer] = Field(min_length=2, max_length=50)
+    context: str | None = Field(default=None, max_length=10000)
     trigger: Trigger = Field(default_factory=Trigger)
+    answer_column: NonEmptyText | None = None
 
     @model_validator(mode='after')
     def answer_ids_are_unique(self) -> ScreeningQuestion:
@@ -97,17 +102,20 @@ class TextParameter(StrictModel):
     id: CriteriaId
     name: NonEmptyText
     description: NonEmptyText
+    context: str | None = Field(default=None, max_length=10000)
     type: Literal['text']
     unit_instructions: str | None = Field(default=None, max_length=10000)
     calculation: str | None = Field(default=None, max_length=10000)
     trigger: Trigger = Field(default_factory=Trigger)
     legacy_category: str | None = Field(default=None, max_length=1000)
+    answer_column: NonEmptyText | None = None
 
 
 class SelectionParameter(StrictModel):
     id: CriteriaId
     name: NonEmptyText
     description: NonEmptyText
+    context: str | None = Field(default=None, max_length=10000)
     type: Literal['selection']
     selection_mode: Literal['single', 'multiple']
     options: list[ParameterOption] = Field(min_length=1, max_length=100)
@@ -115,6 +123,7 @@ class SelectionParameter(StrictModel):
     calculation: str | None = Field(default=None, max_length=10000)
     trigger: Trigger = Field(default_factory=Trigger)
     legacy_category: str | None = Field(default=None, max_length=1000)
+    answer_column: NonEmptyText | None = None
 
     @model_validator(mode='after')
     def option_ids_are_unique(self) -> SelectionParameter:
diff --git a/backend/api/criteria/service.py b/backend/api/criteria/service.py
index 4c565021..91f82d06 100644
--- a/backend/api/criteria/service.py
+++ b/backend/api/criteria/service.py
@@ -19,6 +19,12 @@
 LEGACY_KEYS = {'include', 'criteria', 'l2_criteria', 'parameters'}
 LEGACY_ONLY_KEYS = {'include', 'criteria', 'l2_criteria'}
 MAX_YAML_BYTES = 1_000_000
+DEFAULT_CITATION_FIELDS = ('Title', 'Abstract')
+
+
+def _with_default_citation_fields(values: list[str]) -> list[str]:
+    """Keep the fixed Title and Abstract fields in every criteria document."""
+    return [value for value in dict.fromkeys(values) if value not in DEFAULT_CITATION_FIELDS]
 
 
 class Diagnostic(BaseModel):
@@ -94,6 +100,19 @@ def normalize(self, raw: dict[str, Any], *, source_kind: str = 'backend_load') -
                     f'v2 criteria cannot contain legacy keys: {sorted(mixed)}',
                 )
             normalized = deepcopy(raw)
+            citation_fields = normalized.get('citation_fields')
+            if citation_fields is None:
+                normalized['citation_fields'] = {
+                    'title': 'Title', 'abstract': 'Abstract', 'l1_include': [],
+                }
+            elif isinstance(citation_fields, dict):
+                configured = citation_fields.get('l1_include', [])
+                if isinstance(configured, list) and all(isinstance(value, str) for value in configured):
+                    citation_fields['l1_include'] = _with_default_citation_fields(
+                        configured,
+                    )
+                citation_fields.setdefault('title', 'Title')
+                citation_fields.setdefault('abstract', 'Abstract')
             diagnostics = self._move_question_context_to_answers(normalized)
             return CriteriaLoadResult(
                 criteria=CriteriaConfigV2.model_validate(normalized),
@@ -110,39 +129,14 @@ def normalize(self, raw: dict[str, Any], *, source_kind: str = 'backend_load') -
 
     @staticmethod
     def _move_question_context_to_answers(raw: dict[str, Any]) -> list[Diagnostic]:
-        """Preserve obsolete v2 question guidance by applying it to every answer."""
-        diagnostics: list[Diagnostic] = []
-        for stage in ('l1', 'l2'):
-            items = raw.get(stage)
-            if not isinstance(items, list):
-                continue
-            for question_index, item in enumerate(items):
-                if not isinstance(item, dict) or 'context' not in item:
-                    continue
-                context = item.pop('context')
-                if not isinstance(context, str) or not context.strip():
-                    continue
-                answers = item.get('answers')
-                if not isinstance(answers, list):
-                    continue
-                for answer in answers:
-                    if not isinstance(answer, dict):
-                        continue
-                    answer_context = answer.get('context')
-                    parts = [context.strip()]
-                    if isinstance(answer_context, str) and answer_context.strip():
-                        parts.append(answer_context.strip())
-                    answer['context'] = '\n\n'.join(parts)
-                diagnostics.append(
-                    Diagnostic(
-                        severity='warning',
-                        code='question_context_moved_to_answers',
-                        source_path=[stage, question_index, 'context'],
-                        target_path=[stage, question_index, 'answers'],
-                        message='Question-level context was copied to each answer because screening guidance is answer-specific.',
-                    ),
-                )
-        return diagnostics
+        """Keep item context separate from answer-specific guidance.
+
+        Older code copied question context into every answer. That loses the
+        distinction required by the current prompt contract and duplicates
+        guidance. Reading is intentionally non-mutating and therefore safe for
+        both legacy and current v2 configurations.
+        """
+        return []
 
     def export_yaml(self, criteria: CriteriaConfigV2 | dict[str, Any]) -> str:
         model = criteria if isinstance(
@@ -151,6 +145,30 @@ def export_yaml(self, criteria: CriteriaConfigV2 | dict[str, Any]) -> str:
         payload = model.model_dump(mode='json', exclude_none=True)
         return yaml.safe_dump(payload, sort_keys=False, allow_unicode=True, width=100)
 
+    @staticmethod
+    def required_citation_field_errors(
+        criteria: CriteriaConfigV2,
+        available_fields: set[str] | None = None,
+    ) -> list[dict[str, Any]]:
+        """Return save/use errors while allowing legacy criteria to be loaded."""
+        errors: list[dict[str, Any]] = []
+        fields = criteria.citation_fields
+        for key, label in (('title', 'Title'), ('abstract', 'Abstract')):
+            value = getattr(fields, key)
+            if not value:
+                errors.append({
+                    'code': f'missing_required_{key}_field', 'path': [
+                        'citation_fields', key,
+                    ], 'message': f'{label} source column is required.',
+                })
+            elif available_fields is not None and value not in available_fields:
+                errors.append({
+                    'code': 'unavailable_required_citation_field', 'path': [
+                        'citation_fields', key,
+                    ], 'message': f'{label} source column {value!r} is not available.',
+                })
+        return errors
+
     def build_compatibility_projection(
         self, criteria: CriteriaConfigV2 | dict[str, Any],
     ) -> dict[str, Any]:
@@ -188,6 +206,10 @@ def stage(items):
             grouped[category].append(item)
         return {
             'schema_version': 2,
+            # Keep the canonical mappings/items alongside the legacy-shaped
+            # projection so batch pipelines can use the same source columns
+            # and guidance as the interactive endpoints.
+            'citation_fields': model.citation_fields.model_dump(mode='json', exclude_none=True),
             'l1': {'include': model.citation_fields.l1_include, **stage(model.l1)},
             'l2': stage(model.l2),
             'parameters': {
@@ -217,6 +239,7 @@ def _migrate_legacy(self, legacy: dict[str, Any], *, source_kind: str) -> Criter
                 value.strip() for value in include
             ),
         )
+        normalized_include = _with_default_citation_fields(normalized_include)
         if len(normalized_include) != len(include):
             diagnostics.append(
                 Diagnostic(
@@ -323,7 +346,7 @@ def questions(key: str, stage: str) -> list[dict[str, Any]]:
         l2 = questions('l2_criteria', 'l2')
         criteria = CriteriaConfigV2.model_validate({
             'schema_version': 2,
-            'citation_fields': {'l1_include': normalized_include, 'doi': None},
+            'citation_fields': {'title': 'Title', 'abstract': 'Abstract', 'l1_include': normalized_include, 'doi': None},
             'l1': l1, 'l2': l2, 'parameters': converted_parameters,
         })
         canonical_source = json.dumps(
diff --git a/backend/api/extract/prompts.py b/backend/api/extract/prompts.py
index b9166967..6c3ba399 100644
--- a/backend/api/extract/prompts.py
+++ b/backend/api/extract/prompts.py
@@ -12,17 +12,18 @@
 - "found": a boolean (true/false) indicating whether the parameter was located or could be confidently derived.
 - "value": the extracted value as a string (or null if not found).
 - "explanation": a concise explanation (1-4 sentences) describing why this value was chosen or how it was derived.
-- "evidence_sentences": an array of integers indicating the sentence indices you used as evidence (e.g. [2, 5]). If there are no supporting sentences, return an empty array.
-- "evidence_tables": an array of integers indicating table numbers used (e.g. [1, 2]) or [].
-- "evidence_figures": an array of integers indicating figure numbers used (e.g. [3]) or [].
+- "evidence_sentences": an array of integers indicating the sentence indices you used as evidence (e.g. [2, 5]). Use this for both positive and negative conclusions.
+- "evidence_tables": an array of integers indicating table numbers used (e.g. [1, 2]) or []. Use this for both positive and negative conclusions.
+- "evidence_figures": an array of integers indicating figure numbers used (e.g. [3]) or []. Use this for both positive and negative conclusions.
 
 Requirements:
 - If the parameter is explicitly present, return the value exactly as found (preserve units/format) and list the sentence indices.
 - If the parameter must be computed or approximated, include the computed value and explain the computation in "explanation", and list the sentences used for calculation.
-- If the parameter is not present and cannot be deduced, set "found": false, "value": null, "explanation": briefly state why not found, and "evidence_sentences": [].
+- If the parameter is not present and cannot be deduced, set "found": false and "value": null, explain what the article does and does not report, and list the sentence indices that support that conclusion. Do not return an empty evidence_sentences array when relevant sentences support the explanation. An empty array is appropriate only when the full text contains no relevant supporting sentence.
+- Every factual claim in "explanation" must be grounded in the supplied full text, tables, or figures. In particular, a negative result still requires evidence for the reported quantities, definitions, or limitations that justify why the parameter cannot be extracted.
 - If a calculation is defined for the parameter, with a description of variables to be computed, find those variables and walk through the computation in the explanation.
 - Do NOT include any extra keys, XML, or human commentary. The output must be parseable by json.loads.
-- If a table or figure is referenced, ensure the explanation references the table/figure number and what was extracted from it.
+- If a sentence, table, or figure is referenced, ensure the explanation references the sentence/table/figure number and what was extracted from it.
 
 Inputs available for formatting:
 - {parameter_name}  (a short name for the parameter)
diff --git a/backend/api/extract/router.py b/backend/api/extract/router.py
index f36cb258..864ca2d3 100644
--- a/backend/api/extract/router.py
+++ b/backend/api/extract/router.py
@@ -460,26 +460,6 @@ def _norm_int_list(v: Any) -> list[int]:
             detail='Citation not found to update',
         )
 
-    # Auto-fill human_param_* from llm_param_* if missing (never overwrite)
-    try:
-        col_human = col_name.replace('llm_param_', 'human_param_', 1)
-        human_payload = {
-            **stored,
-            'autofilled': True,
-            'source': 'llm',
-            'timestamp': datetime.utcnow().isoformat() + 'Z',
-        }
-        await run_in_threadpool(
-            cits_dp_service.copy_jsonb_if_empty,
-            citation_id,
-            col_name,
-            col_human,
-            human_payload,
-            table_name,
-        )
-    except Exception:
-        pass
-
     return {'status': 'success', 'sr_id': sr_id, 'citation_id': citation_id, 'column': col_name, 'extraction': stored}
 
 
diff --git a/backend/api/jobs/pipelines/screening_executor.py b/backend/api/jobs/pipelines/screening_executor.py
index 42bab348..02d7b18b 100644
--- a/backend/api/jobs/pipelines/screening_executor.py
+++ b/backend/api/jobs/pipelines/screening_executor.py
@@ -5,6 +5,9 @@
 from fastapi.concurrency import run_in_threadpool
 
 from ...citations import router as citations_router
+from ...criteria.context import format_item_context
+from ...criteria.context import format_title_abstract_context
+from ...criteria.context import resolve_existing_human_value
 from ...extract.prompts import PARAMETER_PROMPT_JSON
 from ...extract.router import extract_fulltext_from_storage
 from ...screen.agentic_utils import build_critical_options
@@ -27,9 +30,11 @@
 from .control import wait_if_paused
 
 
-def _should_skip_ai_output(existing_value: Any, *, force: bool) -> bool:
+def _should_skip_ai_output(
+    existing_value: Any, *, force: bool, skip_existing_ai: bool,
+) -> bool:
     """Return True if we should skip doing work because output already exists."""
-    if force:
+    if force or not skip_existing_ai:
         return False
     # Missing output is NULL/empty per planning doc.
     if existing_value is None:
@@ -39,6 +44,13 @@ def _should_skip_ai_output(existing_value: Any, *, force: bool) -> bool:
     return True
 
 
+def _should_skip_human_answer(
+    human_status: Any, *, force: bool, skip_existing_human: bool,
+) -> bool:
+    """Return True only when the explicit human-answer skip option is enabled."""
+    return not force and skip_existing_human and human_status == 'matched'
+
+
 def _eligible_ids(*, sr_id: str, table_name: str, step: str) -> list[int]:
     """Return eligible citation ids for server-computed kickoff mode.
 
@@ -88,21 +100,27 @@ async def _run_l1_for_citation(
     citation_id: int,
     model: str | None,
     force: bool,
+    skip_existing_ai: bool = False,
+    skip_existing_human: bool = False,
 ) -> tuple[int, int, int]:
     """Returns (done, skipped, failed) increments for this citation."""
     row = await run_in_threadpool(cits_dp_service.get_citation_by_id, citation_id, table_name)
     if not row:
         return (0, 0, 1)
 
-    include_cols = cits_dp_service.load_include_columns_from_criteria(sr)
-    if not include_cols:
-        include_cols = ['title', 'abstract']
-
-    citation_text = citations_router._build_combined_citation_from_row(
-        row, include_cols,
-    )
-
     cp = sr.get('criteria_parsed') or sr.get('criteria') or {}
+    citation_fields = cp.get(
+        'citation_fields',
+    ) if isinstance(cp, dict) else None
+    if isinstance(citation_fields, dict) and (citation_fields.get('title') or citation_fields.get('abstract')):
+        citation_text = format_title_abstract_context(row, citation_fields)
+    else:
+        include_cols = cits_dp_service.load_include_columns_from_criteria(sr) or [
+            'title', 'abstract',
+        ]
+        citation_text = citations_router._build_combined_citation_from_row(
+            row, include_cols,
+        )
     l1 = cp.get('l1') if isinstance(cp, dict) else None
     questions = (l1 or {}).get('questions') if isinstance(l1, dict) else []
     possible = (l1 or {}).get(
@@ -114,6 +132,8 @@ async def _run_l1_for_citation(
     questions = questions if isinstance(questions, list) else []
     possible = possible if isinstance(possible, list) else []
     addinfos = addinfos if isinstance(addinfos, list) else []
+    l1_items = (l1 or {}).get('items') if isinstance(l1, dict) else []
+    l1_items = l1_items if isinstance(l1_items, list) else []
 
     if not questions:
         return (0, 1, 0)
@@ -130,9 +150,22 @@ async def _run_l1_for_citation(
         xtra = addinfos[i] if i < len(
             addinfos,
         ) and isinstance(addinfos[i], str) else ''
+        item = l1_items[i] if i < len(l1_items) and isinstance(l1_items[i], dict) else {
+            'question': q, 'answers': [{'label': opt} for opt in opts],
+        }
+        xtra = format_item_context(item) or xtra
+        human = resolve_existing_human_value(row, item)
+        if _should_skip_human_answer(
+            human.get('status'),
+            force=force,
+            skip_existing_human=skip_existing_human,
+        ):
+            continue
         col = snake_case_column(q)
         existing = row.get(col)
-        if _should_skip_ai_output(existing, force=force):
+        if _should_skip_ai_output(
+            existing, force=force, skip_existing_ai=skip_existing_ai,
+        ):
             continue
 
         if not azure_openai_client.is_configured():
@@ -246,25 +279,6 @@ async def _call_agent(prompt: str):
 
         await run_in_threadpool(cits_dp_service.update_jsonb_column, citation_id, col, classification_json, table_name)
 
-        # Best-effort autofill human_ if empty
-        try:
-            core = snake_case(q, max_len=56)
-            human_col = f"human_{core}" if core else 'human_col'
-            human_payload = {
-                **classification_json,
-                'autofilled': True, 'source': 'llm',
-            }
-            await run_in_threadpool(
-                cits_dp_service.copy_jsonb_if_empty,
-                citation_id,
-                col,
-                human_col,
-                human_payload,
-                table_name,
-            )
-        except Exception:
-            pass
-
         any_ran = True
 
     # Update derived decisions (uses fresh row inside)
@@ -316,6 +330,8 @@ async def _run_l2_for_citation(
     citation_id: int,
     model: str | None,
     force: bool,
+    skip_existing_ai: bool = False,
+    skip_existing_human: bool = False,
 ) -> tuple[int, int, int]:
     row = await run_in_threadpool(cits_dp_service.get_citation_by_id, citation_id, table_name)
     if not row or not row.get('fulltext_url'):
@@ -358,6 +374,10 @@ async def _run_l2_for_citation(
     l2_possible = l2_possible if isinstance(l2_possible, list) else []
     l1_addinfos = l1_addinfos if isinstance(l1_addinfos, list) else []
     l2_addinfos = l2_addinfos if isinstance(l2_addinfos, list) else []
+    l1_items = (l1 or {}).get('items') if isinstance(l1, dict) else []
+    l2_items = (l2 or {}).get('items') if isinstance(l2, dict) else []
+    l1_items = l1_items if isinstance(l1_items, list) else []
+    l2_items = l2_items if isinstance(l2_items, list) else []
 
     merged: list[tuple[str, str, int]] = []  # (question, source_step, idx)
     seen_q: set[str] = set()
@@ -375,12 +395,18 @@ async def _run_l2_for_citation(
     if not merged:
         return (0, 1, 0)
 
-    include_cols = cits_dp_service.load_include_columns_from_criteria(sr) or [
-        'title', 'abstract',
-    ]
-    citation_text = citations_router._build_combined_citation_from_row(
-        row, include_cols,
-    )
+    citation_fields = cp.get(
+        'citation_fields',
+    ) if isinstance(cp, dict) else None
+    if isinstance(citation_fields, dict) and (citation_fields.get('title') or citation_fields.get('abstract')):
+        citation_text = format_title_abstract_context(row, citation_fields)
+    else:
+        include_cols = cits_dp_service.load_include_columns_from_criteria(sr) or [
+            'title', 'abstract',
+        ]
+        citation_text = citations_router._build_combined_citation_from_row(
+            row, include_cols,
+        )
     fulltext = row.get('fulltext') or citation_text
 
     # Tables/Figures context from row
@@ -452,6 +478,9 @@ async def _run_l2_for_citation(
             xtra = l1_addinfos[idx] if idx < len(
                 l1_addinfos,
             ) and isinstance(l1_addinfos[idx], str) else ''
+            item = l1_items[idx] if idx < len(l1_items) and isinstance(l1_items[idx], dict) else {
+                'question': q, 'answers': [{'label': opt} for opt in opts],
+            }
         else:
             opts = l2_possible[idx] if idx < len(
                 l2_possible,
@@ -459,9 +488,22 @@ async def _run_l2_for_citation(
             xtra = l2_addinfos[idx] if idx < len(
                 l2_addinfos,
             ) and isinstance(l2_addinfos[idx], str) else ''
+            item = l2_items[idx] if idx < len(l2_items) and isinstance(l2_items[idx], dict) else {
+                'question': q, 'answers': [{'label': opt} for opt in opts],
+            }
+        xtra = format_item_context(item) or xtra
+        human = resolve_existing_human_value(row, item)
+        if _should_skip_human_answer(
+            human.get('status'),
+            force=force,
+            skip_existing_human=skip_existing_human,
+        ):
+            continue
         col = snake_case_column(q)
         existing = row.get(col)
-        if _should_skip_ai_output(existing, force=force):
+        if _should_skip_ai_output(
+            existing, force=force, skip_existing_ai=skip_existing_ai,
+        ):
             continue
 
         # --- Agentic (screening + critical) ---
@@ -583,25 +625,6 @@ async def _call_agent(prompt: str):
 
         await run_in_threadpool(cits_dp_service.update_jsonb_column, citation_id, col, classification_json, table_name)
 
-        # Best-effort autofill human
-        try:
-            core = snake_case(q, max_len=56)
-            human_col = f"human_{core}" if core else 'human_col'
-            human_payload = {
-                **classification_json,
-                'autofilled': True, 'source': 'llm',
-            }
-            await run_in_threadpool(
-                cits_dp_service.copy_jsonb_if_empty,
-                citation_id,
-                col,
-                human_col,
-                human_payload,
-                table_name,
-            )
-        except Exception:
-            pass
-
         any_ran = True
 
     try:
@@ -623,6 +646,8 @@ async def _run_extract_for_citation(
     citation_id: int,
     model: str | None,
     force: bool,
+    skip_existing_ai: bool = False,
+    skip_existing_human: bool = False,
 ) -> tuple[int, int, int]:
     row = await run_in_threadpool(cits_dp_service.get_citation_by_id, citation_id, table_name)
     if not row or not row.get('fulltext_url'):
@@ -652,8 +677,14 @@ async def _run_extract_for_citation(
     categories = categories if isinstance(categories, list) else []
     possible = possible if isinstance(possible, list) else []
     descs = descs if isinstance(descs, list) else []
+    parameter_items = (params or {}).get(
+        'items',
+    ) if isinstance(params, dict) else []
+    parameter_items = parameter_items if isinstance(
+        parameter_items, list,
+    ) else []
 
-    params_flat: list[tuple[str, str]] = []
+    params_flat: list[tuple[str, str, dict[str, Any]]] = []
     for i, _cat in enumerate(categories):
         arr = possible[i] if i < len(possible) and isinstance(
             possible[i], list,
@@ -669,7 +700,14 @@ async def _run_extract_for_citation(
             if j < len(darr):
                 d = str(darr[j])
                 d = d.replace('<desc>', '').replace('</desc>', '')
-            params_flat.append((name, d or name))
+            item = next(
+                (
+                    candidate for candidate in parameter_items if isinstance(
+                        candidate, dict,
+                    ) and candidate.get('name') == name
+                ), {'name': name, 'description': d or name},
+            )
+            params_flat.append((name, d or name, item))
 
     if not params_flat:
         return (0, 1, 0)
@@ -761,19 +799,28 @@ def _extract_json_object(text: str) -> str | None:
                     return t[start: i + 1]
         return None
 
-    for name, desc in params_flat:
+    for name, desc, item in params_flat:
         # More responsive pause/cancel: check between parameters
         if await run_in_threadpool(run_all_repo.is_canceled, job_id):
             raise PipelineCanceled()
         await wait_if_paused(job_id)
         col = snake_case_param(name)
         existing = row.get(col)
-        if _should_skip_ai_output(existing, force=force):
+        if _should_skip_ai_output(
+            existing, force=force, skip_existing_ai=skip_existing_ai,
+        ):
+            continue
+        human = resolve_existing_human_value(row, item)
+        if _should_skip_human_answer(
+            human.get('status'),
+            force=force,
+            skip_existing_human=skip_existing_human,
+        ):
             continue
 
         prompt = PARAMETER_PROMPT_JSON.format(
             parameter_name=name,
-            parameter_description=desc,
+            parameter_description=format_item_context(item) or desc,
             fulltext=fulltext,
             tables=tables_text,
             figures=figures_text,
@@ -820,21 +867,6 @@ def _extract_json_object(text: str) -> str | None:
 
         await run_in_threadpool(cits_dp_service.update_jsonb_column, citation_id, col, stored, table_name)
 
-        # best-effort autofill human_param
-        try:
-            human_col = col.replace('llm_param_', 'human_param_', 1)
-            human_payload = {**stored, 'autofilled': True, 'source': 'llm'}
-            await run_in_threadpool(
-                cits_dp_service.copy_jsonb_if_empty,
-                citation_id,
-                col,
-                human_col,
-                human_payload,
-                table_name,
-            )
-        except Exception:
-            pass
-
         any_ran = True
 
     if any_ran:
diff --git a/backend/api/jobs/pipelines/screening_pipeline.py b/backend/api/jobs/pipelines/screening_pipeline.py
index d2899329..b77bdd4c 100644
--- a/backend/api/jobs/pipelines/screening_pipeline.py
+++ b/backend/api/jobs/pipelines/screening_pipeline.py
@@ -50,7 +50,9 @@ async def execute_item(
         try:
             executor = self._executors[context.step]
         except KeyError as exc:
-            raise ValueError(f"Unsupported screening step: '{context.step}'") from exc
+            raise ValueError(
+                f"Unsupported screening step: '{context.step}'",
+            ) from exc
 
         common = {
             'job_id': context.job_id,
@@ -59,6 +61,8 @@ async def execute_item(
             'citation_id': int(work_item),
             'model': context.model,
             'force': bool(context.config.get('force')),
+            'skip_existing_ai': bool(context.config.get('skip_existing_ai')),
+            'skip_existing_human': bool(context.config.get('skip_existing_human')),
         }
         if context.step in {'l2', 'extract'}:
             common['sr_id'] = context.sr_id
@@ -75,4 +79,4 @@ def format_phase(self, work_item: int) -> str:
         return f'citation {work_item}'
 
     def error_stage(self, context: JobContext) -> str:
-        return context.step
\ No newline at end of file
+        return context.step
diff --git a/backend/api/jobs/router.py b/backend/api/jobs/router.py
index c9abc96c..8b356bd7 100644
--- a/backend/api/jobs/router.py
+++ b/backend/api/jobs/router.py
@@ -23,9 +23,9 @@
 from .procrastinate_app import jobs_enabled
 from .procrastinate_app import worker_concurrency
 from .run_all_repo import run_all_repo
+from .run_all_tasks import enqueue_available_chunks
 from .run_all_tasks import run_all_chunk
 from .run_all_tasks import run_all_start
-from .run_all_tasks import enqueue_available_chunks
 from .scheduler_service import InvalidJobTransition
 from .scheduler_service import SchedulerService
 # Import task objects so we can enqueue via Task.defer_async (Procrastinate 3.2.x)
@@ -63,7 +63,13 @@ class RunAllStartRequest(BaseModel):
     step: str = Field(..., description='l1 | l2 | extract')
     model: str | None = Field(None, description='LLM model name')
     force: bool = Field(
-        False, description='If true, overwrite existing outputs',
+        False, description='Legacy option: overwrite existing AI outputs and ignore human answers',
+    )
+    skip_existing_ai: bool = Field(
+        False, description='Skip criteria that already have an AI answer',
+    )
+    skip_existing_human: bool = Field(
+        False, description='Skip criteria that already have a human answer',
     )
     # NOTE: backend currently forces chunk_size=1 to ensure fair interleaving
     # of multiple active run-all jobs.
@@ -143,6 +149,14 @@ async def start_run_all(
     # If legacy llm_* outputs exist but normalized agent runs are missing, we must
     # regenerate results to populate screening_agent_runs.
     # We enforce this by auto-enabling force overwrite.
+    # The explicit skip options are preferred. Keep `force` as a backwards-
+    # compatible alias for the old behavior, which reruns everything.
+    skip_existing_ai = bool(
+        payload.skip_existing_ai,
+    ) and not bool(payload.force)
+    skip_existing_human = bool(
+        payload.skip_existing_human,
+    ) and not bool(payload.force)
     force = bool(payload.force)
     try:
         table_name = (screening or {}).get('table_name') or 'citations'
@@ -197,6 +211,8 @@ async def start_run_all(
             model=normalized_model,
             meta={
                 'force': force,
+                'skip_existing_ai': skip_existing_ai,
+                'skip_existing_human': skip_existing_human,
                 'chunk_size': int(payload.chunk_size),
                 'explicit_ids': bool(sanitized_ids is not None),
                 'legacy_auto_force': (force and (not bool(payload.force))),
@@ -274,7 +290,9 @@ async def start_job(
     if payload.pipeline_key != 'pdf_linkage':
         raise HTTPException(status_code=400, detail='Unsupported pipeline_key')
     if not jobs_enabled():
-        raise HTTPException(status_code=503, detail='Background jobs are disabled')
+        raise HTTPException(
+            status_code=503, detail='Background jobs are disabled',
+        )
     await load_sr_and_check(sr_id, current_user, srdb_service)
     await run_in_threadpool(run_all_repo.ensure_tables)
     existing = await run_in_threadpool(run_all_repo.get_active_job_for_sr, sr_id)
diff --git a/backend/api/screen/prompts.py b/backend/api/screen/prompts.py
index 1ed53cb0..9214d7a8 100644
--- a/backend/api/screen/prompts.py
+++ b/backend/api/screen/prompts.py
@@ -98,6 +98,11 @@
 Citation:
 {cit}
 
+For title/abstract screening, interpret the citation using this structured metadata format:
+Title: ...
+Abstract: ...
+Other fields: ...
+
 Choose EXACTLY ONE of these options (exact text):
 {options}
 
@@ -129,6 +134,11 @@
 Citation:
 {cit}
 
+Structured citation metadata:
+Title: ...
+Abstract: ...
+Other fields: ...
+
 The first model answered:
 "{screening_answer}"
 
diff --git a/backend/api/screen/router.py b/backend/api/screen/router.py
index 7c6b4dce..6a0b2e6b 100644
--- a/backend/api/screen/router.py
+++ b/backend/api/screen/router.py
@@ -24,6 +24,8 @@
 from ..core.cit_utils import load_sr_and_check
 from ..core.config import settings
 from ..core.security import get_current_active_user
+from ..criteria.context import format_item_context
+from ..criteria.context import format_title_abstract_context
 from ..criteria.runtime import item_is_visible
 from ..services.azure_openai_client import azure_openai_client
 from ..services.cit_db_service import cits_dp_service
@@ -693,12 +695,16 @@ def _parse_selected_from_human_payload(v: Any) -> str | None:
         try:
             obj = json.loads(s)
             if isinstance(obj, dict):
+                if obj.get('source') == 'llm':
+                    return None
                 sel = obj.get('selected')
                 return str(sel).strip() if isinstance(sel, str) else None
         except Exception:
             return s
         return None
     if isinstance(v, dict):
+        if v.get('source') == 'llm':
+            return None
         sel = v.get('selected')
         return str(sel).strip() if isinstance(sel, str) else None
     return None
@@ -840,8 +846,21 @@ async def classify_citation(
         )
 
     # Build or use provided citation text (fall back to combined title/abstract when not provided)
-    citation_text = payload.citation_text or citations_router._build_combined_citation_from_row(
-        row, payload.include_columns,
+    citation_fields = canonical.get('citation_fields') or {}
+    if payload.citation_text:
+        citation_text = payload.citation_text
+    elif citation_fields.get('title') or citation_fields.get('abstract'):
+        citation_text = format_title_abstract_context(row, citation_fields)
+    else:
+        citation_text = citations_router._build_combined_citation_from_row(
+            row, payload.include_columns,
+        )
+
+    item_context = format_item_context(
+        source_item or {
+            'question': payload.question,
+            'answers': [{'label': option} for option in payload.options],
+        },
     )
 
     # Ensure LLM client is available
@@ -919,7 +938,11 @@ async def classify_citation(
         prompt = PROMPT_JSON_TEMPLATE_FULLTEXT.format(
             question=payload.question,
             options=options_listed,
-            xtra=payload.xtra or '',
+            xtra='\n\n'.join(
+                part for part in (
+                    item_context, payload.xtra or '',
+                ) if part
+            ),
             fulltext=fulltext or citation_text,
             tables='\n'.join(tables_md_lines) if tables_md_lines else '(none)',
             figures='\n'.join(figures_lines) if figures_lines else '(none)',
@@ -948,7 +971,11 @@ async def classify_citation(
             question=payload.question,
             cit=citation_text,
             options=options_listed,
-            xtra=payload.xtra or '',
+            xtra='\n\n'.join(
+                part for part in (
+                    item_context, payload.xtra or '',
+                ) if part
+            ),
         )
         llm_response = await azure_openai_client.simple_chat(
             user_message=prompt,
@@ -1026,15 +1053,6 @@ async def classify_citation(
     # Persist into Postgres under a dynamic column name derived from question
     col_name = snake_case_column(payload.question)
 
-    # Human mirror column name (same slug as llm_, but prefixed human_)
-    try:
-        col_core_h = snake_case(
-            payload.question, max_len=56,
-        ) if snake_case else ''
-    except Exception:
-        col_core_h = ''
-    human_col_name = f"human_{col_core_h}" if col_core_h else 'human_col'
-
     try:
         updated = await run_in_threadpool(cits_dp_service.update_jsonb_column, citation_id, col_name, classification_json, table_name)
     except RuntimeError as rexc:
@@ -1053,26 +1071,6 @@ async def classify_citation(
             detail='Citation not found to update',
         )
 
-    # Auto-fill human_* from llm_* if missing (never overwrite)
-    try:
-        human_payload = {
-            **classification_json,
-            'autofilled': True,
-            'source': 'llm',
-            'timestamp': datetime.utcnow().isoformat() + 'Z',
-        }
-        await run_in_threadpool(
-            cits_dp_service.copy_jsonb_if_empty,
-            citation_id,
-            col_name,
-            human_col_name,
-            human_payload,
-            table_name,
-        )
-    except Exception:
-        # best-effort
-        pass
-
     await update_inclusion_decision(sr, citation_id, payload.screening_step, 'llm')
 
     return {'status': 'success', 'sr_id': sr_id, 'citation_id': citation_id, 'column': col_name, 'classification': classification_json}
diff --git a/backend/api/services/citation_field_service.py b/backend/api/services/citation_field_service.py
index 24678512..30d1fbd0 100644
--- a/backend/api/services/citation_field_service.py
+++ b/backend/api/services/citation_field_service.py
@@ -10,6 +10,7 @@
     'human_l1_decision', 'human_l2_decision',
 }
 SYSTEM_PREFIXES = ('llm_', 'human_', 'fulltext', 'parameters_', 'l1_', 'l2_')
+CANONICAL_FIELD_NAMES = {'title': 'Title', 'abstract': 'Abstract'}
 
 
 def _is_user_field(name: str) -> bool:
@@ -26,6 +27,9 @@ def build_citation_field_contract(review: dict[str, Any], columns: list[dict[str
         ).strip()
         if not name or not _is_user_field(name):
             continue
+        name = CANONICAL_FIELD_NAMES.get(name.casefold(), name)
+        if any(field['name'] == name for field in fields):
+            continue
         fields.append({
             'name': name, 'data_type': str(
                 column.get('data_type') or 'text',
@@ -46,9 +50,11 @@ def build_citation_field_contract(review: dict[str, Any], columns: list[dict[str
         ) or criteria.get('include') or []
     ]
     doi = configured.get('doi')
+    title = configured.get('title')
+    abstract = configured.get('abstract')
     available = {field['name'] for field in fields}
     unavailable = [
-        value for value in [*selected, doi]
+        value for value in [*selected, title, abstract, doi]
         if value and value not in available
     ]
     return {
diff --git a/backend/api/services/sr_db_service.py b/backend/api/services/sr_db_service.py
index 727f874c..2f0c205e 100644
--- a/backend/api/services/sr_db_service.py
+++ b/backend/api/services/sr_db_service.py
@@ -642,7 +642,10 @@ def save_criteria_config(
             logger.exception(
                 'Failed to save canonical criteria config: %s', exc,
             )
-            raise
+            raise HTTPException(
+                status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                detail=f'Failed to save criteria configuration: {exc}',
+            ) from exc
 
     def list_systematic_reviews_for_user(self, user_email: str) -> list[dict[str, Any]]:
         """
diff --git a/backend/api/sr/router.py b/backend/api/sr/router.py
index cd18f9de..51132be1 100644
--- a/backend/api/sr/router.py
+++ b/backend/api/sr/router.py
@@ -397,9 +397,14 @@ async def get_criteria_config(
                 review['criteria_yaml'], source_kind='backend_load',
             )
         else:
-            raise ValueError(
-                'No usable criteria configuration is stored for this review.',
+            empty_criteria = CriteriaConfigV2(
+                citation_fields={'title': 'Title', 'abstract': 'Abstract'},
             )
+            return {
+                'criteria': empty_criteria.model_dump(mode='json', exclude_none=True),
+                'revision': review.get('criteria_revision', 0),
+                'warnings': [],
+            }
     except ValueError as exc:
         raise _criteria_error(exc) from exc
 
@@ -426,10 +431,19 @@ async def validate_criteria_config(
     criteria: CriteriaConfigV2,
     current_user: dict[str, Any] = Depends(get_current_active_user),
 ):
-    await _load_criteria_review(sr_id, current_user)
+    review = await _load_criteria_review(sr_id, current_user)
+    available = await run_in_threadpool(discover_citation_fields, review)
+    available_names = {
+        str(field.get('name'))
+        for field in available.get('fields', [])
+    }
+    errors = criteria_configuration_service.required_citation_field_errors(
+        criteria, available_names,
+    )
     return {
-        'valid': True,
+        'valid': not errors,
         'criteria': criteria.model_dump(mode='json', exclude_none=True),
+        'errors': errors,
         'warnings': [],
     }
 
@@ -463,7 +477,26 @@ async def save_criteria_config(
             detail='This SR already has screening data. Pass force=true to confirm criteria invalidation.',
         )
     try:
-        stored = criteria_configuration_service.normalize(review['criteria'])
+        available = await run_in_threadpool(discover_citation_fields, review)
+        available_names = {
+            str(field.get('name'))
+            for field in available.get('fields', [])
+        }
+        required_errors = criteria_configuration_service.required_citation_field_errors(
+            payload.criteria, available_names,
+        )
+        if required_errors:
+            raise HTTPException(
+                status_code=status.HTTP_422_UNPROCESSABLE_ENTITY, detail={
+                    'errors': required_errors,
+                },
+            )
+        previous_criteria = review.get('criteria')
+        stored = None
+        if isinstance(previous_criteria, dict):
+            stored = criteria_configuration_service.normalize(
+                previous_criteria,
+            )
     except (KeyError, ValueError):
         stored = None
     if stored:
diff --git a/backend/tests/citation_field_service_test.py b/backend/tests/citation_field_service_test.py
index 6dba2c3f..3eb27288 100644
--- a/backend/tests/citation_field_service_test.py
+++ b/backend/tests/citation_field_service_test.py
@@ -30,3 +30,24 @@ def test_preserves_order_filters_runtime_columns_and_reports_unavailable(self):
 
     def test_supports_review_without_upload(self):
         self.assertEqual(build_citation_field_contract({}, [])['fields'], [])
+
+    def test_canonicalizes_normalized_uploaded_title_and_abstract_columns(self):
+        result = build_citation_field_contract(
+            {
+                'criteria': {
+                    'citation_fields': {
+                        'title': 'Title', 'abstract': 'Abstract', 'l1_include': [],
+                    },
+                },
+            },
+            [{'column_name': name, 'data_type': 'text'} for name in [
+                'id', 'title', 'abstract',
+            ]],
+        )
+        self.assertEqual(
+            result['fields'], [
+                {'name': 'Title', 'data_type': 'text'},
+                {'name': 'Abstract', 'data_type': 'text'},
+            ],
+        )
+        self.assertEqual(result['unavailable_configured_fields'], [])
diff --git a/backend/tests/criteria_config_router_test.py b/backend/tests/criteria_config_router_test.py
index e61464dc..e053ccbb 100644
--- a/backend/tests/criteria_config_router_test.py
+++ b/backend/tests/criteria_config_router_test.py
@@ -113,6 +113,34 @@ async def test_save_requires_matching_legacy_migration_fingerprint(self, load_re
             caught.exception.detail['fingerprint'].startswith('sha256:'),
         )
 
+    @patch('api.sr.router.srdb_service.save_criteria_config')
+    @patch('api.sr.router.discover_citation_fields')
+    @patch('api.sr.router._load_criteria_review', new_callable=AsyncMock)
+    async def test_save_initial_config_when_criteria_is_missing(
+        self,
+        load_review: AsyncMock,
+        discover,
+        save,
+    ) -> None:
+        load_review.return_value = {
+            'criteria': None,
+            'criteria_revision': 0,
+        }
+        discover.return_value = {
+            'fields': [{'name': 'Title'}, {'name': 'Abstract'}],
+            'unavailable_configured_fields': [],
+        }
+        save.return_value = {'revision': 1}
+        payload = CriteriaConfigSaveRequest(
+            expected_revision=0,
+            criteria=CriteriaConfigV2.model_validate(EMPTY_CONFIG),
+        )
+
+        response = await save_criteria_config('review-id', payload, USER)
+
+        self.assertEqual(response['revision'], 1)
+        save.assert_called_once()
+
 
 if __name__ == '__main__':
     unittest.main()
diff --git a/backend/tests/criteria_context_test.py b/backend/tests/criteria_context_test.py
new file mode 100644
index 00000000..ea2f353c
--- /dev/null
+++ b/backend/tests/criteria_context_test.py
@@ -0,0 +1,84 @@
+from __future__ import annotations
+
+import unittest
+
+from api.criteria.context import format_item_context
+from api.criteria.context import format_title_abstract_context
+from api.criteria.context import match_answer_label
+from api.criteria.context import resolve_existing_human_value
+from api.screen.router import _parse_selected_from_human_payload
+
+
+class CriteriaContextTests(unittest.TestCase):
+    def test_title_abstract_format_uses_configured_headers(self) -> None:
+        row = {
+            'Paper title': 'Configured title',
+            'Paper abstract': 'Configured abstract', 'Year': 2025,
+        }
+        result = format_title_abstract_context(
+            row, {
+                'title': 'Paper title', 'abstract': 'Paper abstract', 'l1_include': ['Paper title', 'Year'],
+            },
+        )
+        self.assertEqual(
+            result, 'Title: Configured title\nAbstract: Configured abstract\nOther fields: Year: 2025',
+        )
+
+    def test_answer_matching_is_formatting_only_and_preserves_raw_value(self) -> None:
+        matched = match_answer_label(
+            '  Yes   ', [{'id': 'yes', 'label': 'Yes'}],
+        )
+        self.assertEqual(matched['status'], 'matched')
+        self.assertEqual(matched['raw'], '  Yes   ')
+        self.assertEqual(
+            match_answer_label(
+                'Probably', [{'id': 'yes', 'label': 'Yes'}],
+            )['status'], 'unmatched',
+        )
+        self.assertEqual(
+            match_answer_label(
+                '   ', [{'id': 'yes', 'label': 'Yes'}],
+            )['status'], 'blank',
+        )
+
+    def test_parameter_mapping_is_authoritative_when_valid(self) -> None:
+        result = resolve_existing_human_value(
+            {'reviewer value': '42'},
+            {'answer_column': 'reviewer value', 'name': 'Rate', 'type': 'text'},
+        )
+        self.assertEqual(result['status'], 'matched')
+        self.assertEqual(result['value'], '42')
+
+    def test_item_context_keeps_option_context_separate(self) -> None:
+        result = format_item_context({
+            'question': 'Eligible?', 'context': 'Item guidance',
+            'answers': [{'label': 'Yes', 'context': 'Yes guidance'}],
+        })
+        self.assertIn('Context: Item guidance', result)
+        self.assertIn('- Yes: Yes guidance', result)
+
+    def test_metrics_ignore_legacy_llm_autofilled_human_values(self) -> None:
+        self.assertIsNone(
+            _parse_selected_from_human_payload(
+                {'selected': 'Yes', 'source': 'llm', 'autofilled': True},
+            ),
+        )
+        self.assertEqual(
+            _parse_selected_from_human_payload(
+                {'selected': 'Yes', 'source': 'csv_upload', 'autofilled': True},
+            ),
+            'Yes',
+        )
+        self.assertEqual(
+            _parse_selected_from_human_payload(
+                {
+                    'selected': 'No', 'human': True,
+                    'reviewer': 'reviewer@example.com',
+                },
+            ),
+            'No',
+        )
+
+
+if __name__ == '__main__':
+    unittest.main()
diff --git a/backend/tests/extract_prompt_test.py b/backend/tests/extract_prompt_test.py
new file mode 100644
index 00000000..a64e77e3
--- /dev/null
+++ b/backend/tests/extract_prompt_test.py
@@ -0,0 +1,9 @@
+from __future__ import annotations
+
+from api.extract.prompts import PARAMETER_PROMPT_JSON
+
+
+def test_parameter_prompt_requires_evidence_for_negative_results():
+    assert 'Use this for both positive and negative conclusions.' in PARAMETER_PROMPT_JSON
+    assert 'list the sentence indices that support that conclusion' in PARAMETER_PROMPT_JSON
+    assert 'Do not return an empty evidence_sentences array when relevant sentences support the explanation.' in PARAMETER_PROMPT_JSON
diff --git a/backend/tests/fulltext_screening_prompt_test.py b/backend/tests/fulltext_screening_prompt_test.py
new file mode 100644
index 00000000..ccd6a332
--- /dev/null
+++ b/backend/tests/fulltext_screening_prompt_test.py
@@ -0,0 +1,21 @@
+from __future__ import annotations
+
+from api.screen.prompts import PROMPT_XML_TEMPLATE_FULLTEXT
+
+
+def test_fulltext_screening_prompt_contains_all_evidence_context():
+    prompt = PROMPT_XML_TEMPLATE_FULLTEXT.format(
+        question='Does the study qualify?',
+        options='Include\nExclude',
+        xtra='Population guidance',
+        fulltext='[0] Full text sentence',
+        tables='Table [T1]\n| result |',
+        figures='Figure [F2] caption: outcome',
+    )
+
+    assert '[0] Full text sentence' in prompt
+    assert 'Table [T1]' in prompt
+    assert 'Figure [F2]' in prompt
+    assert '<evidence_sentences>' in prompt
+    assert '<evidence_tables>' in prompt
+    assert '<evidence_figures>' in prompt
diff --git a/backend/tests/screen_agentic_utils_test.py b/backend/tests/screen_agentic_utils_test.py
index 4ce754af..3f8233f7 100644
--- a/backend/tests/screen_agentic_utils_test.py
+++ b/backend/tests/screen_agentic_utils_test.py
@@ -34,6 +34,20 @@ def test_invalid_confidence_is_reported_missing():
     assert validate_agent_response(parsed, stage='screening') == ['confidence']
 
 
+def test_fulltext_evidence_is_parsed_and_deduplicated():
+    parsed = parse_agent_xml(
+        '<answer>Include</answer><confidence>0.8</confidence>'
+        '<rationale>Supported by the results.</rationale>'
+        '<evidence_sentences>2, 5, 2, invalid</evidence_sentences>'
+        '<evidence_tables>1, 1</evidence_tables>'
+        '<evidence_figures>3, bad, 4</evidence_figures>',
+    )
+
+    assert parsed.evidence_sentences == [2, 5]
+    assert parsed.evidence_tables == [1]
+    assert parsed.evidence_figures == [3, 4]
+
+
 @pytest.mark.asyncio
 async def test_screening_repairs_missing_rationale_once():
     responses = iter([
diff --git a/backend/tests/screening_executor_skip_test.py b/backend/tests/screening_executor_skip_test.py
new file mode 100644
index 00000000..19e5f545
--- /dev/null
+++ b/backend/tests/screening_executor_skip_test.py
@@ -0,0 +1,37 @@
+from __future__ import annotations
+
+from api.jobs.pipelines.screening_executor import _should_skip_ai_output
+from api.jobs.pipelines.screening_executor import _should_skip_human_answer
+
+
+def test_existing_ai_answer_is_skipped_only_when_requested() -> None:
+    assert not _should_skip_ai_output(
+        'existing', force=False, skip_existing_ai=False,
+    )
+    assert _should_skip_ai_output(
+        'existing', force=False, skip_existing_ai=True,
+    )
+    assert not _should_skip_ai_output(
+        None, force=False, skip_existing_ai=True,
+    )
+
+
+def test_existing_human_answer_is_skipped_only_when_requested() -> None:
+    assert not _should_skip_human_answer(
+        'matched', force=False, skip_existing_human=False,
+    )
+    assert _should_skip_human_answer(
+        'matched', force=False, skip_existing_human=True,
+    )
+    assert not _should_skip_human_answer(
+        'blank', force=False, skip_existing_human=True,
+    )
+
+
+def test_force_preserves_rerun_everything_behavior() -> None:
+    assert not _should_skip_ai_output(
+        'existing', force=True, skip_existing_ai=True,
+    )
+    assert not _should_skip_human_answer(
+        'matched', force=True, skip_existing_human=True,
+    )
diff --git a/frontend/app/[lang]/can-sr/l1-screen/view/page.tsx b/frontend/app/[lang]/can-sr/l1-screen/view/page.tsx
index 3d4a7fd0..2f122899 100644
--- a/frontend/app/[lang]/can-sr/l1-screen/view/page.tsx
+++ b/frontend/app/[lang]/can-sr/l1-screen/view/page.tsx
@@ -8,6 +8,7 @@ import { getAuthToken, getTokenType } from '@/lib/auth'
 import { Wand2 } from 'lucide-react'
 import { useDictionary } from '@/app/[lang]/DictionaryProvider'
 import { needsHumanReviewForCriterion } from '@/components/can-sr/needsHumanReview'
+import { ScreeningCitationContext, extractHumanAnswer, humanAnswerStatus, resolveConfiguredValue } from '@/components/can-sr/screening-citation-context'
 
 /*
   Title & Abstract single-citation viewer for L1 screening.
@@ -92,6 +93,8 @@ function extractXmlTag(text: string, tag: string): string {
 type CriteriaData = {
   questions: string[]
   possible_answers: string[][]
+  items?: Array<{ answer_column?: string | null }>
+  citation_fields?: { title?: string | null; abstract?: string | null; l1_include?: string[] }
 }
 
 type LatestAgentRun = {
@@ -403,6 +406,8 @@ export default function CanSrL1ScreenPage() {
             setCriteriaData({
               questions: l1.questions,
               possible_answers: l1.possible_answers,
+              items: Array.isArray(l1.items) ? l1.items : [],
+              citation_fields: parsed?.citation_fields || {},
             })
           } else {
             // If structure is different attempt best-effort mapping
@@ -413,6 +418,8 @@ export default function CanSrL1ScreenPage() {
               possible_answers: Array.isArray(possible_answers)
                 ? possible_answers
                 : [],
+              items: Array.isArray(l1?.items) ? l1.items : [],
+              citation_fields: parsed?.citation_fields || {},
             })
           }
         }
@@ -440,9 +447,9 @@ export default function CanSrL1ScreenPage() {
 
     criteriaData.questions.forEach((q: string, idx: number) => {
       const llmCol = snakeCaseColumn(q)
-      const humanCol = humanScreenColumn(q)
+      const humanCol = criteriaData.items?.[idx]?.answer_column || humanScreenColumn(q)
 
-      const humanRaw = (citation as any)?.[humanCol]
+      const humanRaw = resolveConfiguredValue(citation as any, humanCol)
       const llmRaw = (citation as any)?.[llmCol]
 
       const parseMaybeJson = (v: any) => {
@@ -463,8 +470,8 @@ export default function CanSrL1ScreenPage() {
       // 1) Prefer human_* for dropdown
       if (humanParsed && typeof humanParsed === 'object' && (humanParsed as any).selected !== undefined) {
         newSelections[idx] = (humanParsed as any).selected
-      } else if (typeof humanParsed === 'string' && humanParsed) {
-        newSelections[idx] = humanParsed
+      } else if (extractHumanAnswer(humanParsed)) {
+        newSelections[idx] = extractHumanAnswer(humanParsed)
       }
 
       // 2) Always populate AI panel from llm_* when available
@@ -659,22 +666,10 @@ export default function CanSrL1ScreenPage() {
 
     return (
       <div className="space-y-3">
-        <div>
-          <p className="text-xs text-gray-600">Citation #{citation.id}</p>
-          <h2 className="text-lg font-semibold text-gray-900">
-            {citation.title}
-          </h2>
-        </div>
-
-        <div className="rounded-md border border-gray-200 bg-white p-4">
-          <h3 className="text-sm font-medium text-gray-800">{dict.screening.abstract}</h3>
-          <p className="mt-2 text-sm whitespace-pre-wrap text-gray-800">
-            {citation.abstract || dict.screening.noAbstract}
-          </p>
-        </div>
+        <ScreeningCitationContext citation={citation} fields={criteriaData?.citation_fields} />
       </div>
     )
-  }, [citation, loadingCitation, dict, error])
+  }, [citation, loadingCitation, dict, error, criteriaData])
 
   if (!srId || !citationId) {
     // guard - redirect already handled in effect but keep safe render
@@ -763,6 +758,8 @@ export default function CanSrL1ScreenPage() {
                 <div className="space-y-4">
                   {criteriaData.questions.map((q, idx) => {
                     const options = criteriaData.possible_answers[idx] || []
+                    const answerColumn = criteriaData.items?.[idx]?.answer_column || null
+                    const answerStatus = humanAnswerStatus(citation!, answerColumn, options)
                     const current = selections[idx] ?? ''
                     const aiData = aiPanels[idx]
                     const aiSelected =
@@ -832,6 +829,11 @@ export default function CanSrL1ScreenPage() {
                                 </option>
                               ))}
                             </select>
+                            {answerStatus !== 'unconfigured' && answerStatus !== 'matched' ? (
+                              <p className="mt-1 text-xs text-amber-700" role="status">
+                                Human answer column “{answerColumn}”: {answerStatus === 'missing' ? 'not found in this citation' : answerStatus === 'blank' ? 'blank for this citation' : 'value does not match an available answer'}.
+                              </p>
+                            ) : null}
                           </div>
 
                           <div className="ml-3 flex flex-col items-end space-y-2">
diff --git a/frontend/app/[lang]/can-sr/l2-screen/view/page.tsx b/frontend/app/[lang]/can-sr/l2-screen/view/page.tsx
index bc612c30..09396811 100644
--- a/frontend/app/[lang]/can-sr/l2-screen/view/page.tsx
+++ b/frontend/app/[lang]/can-sr/l2-screen/view/page.tsx
@@ -4,11 +4,19 @@ import React, { useCallback, useEffect, useMemo, useRef, useState } from 'react'
 import { useParams, useRouter, useSearchParams } from 'next/navigation'
 import GCHeader, { SRHeader } from '@/components/can-sr/headers'
 import { ModelSelector } from '@/components/chat'
-import PDFBoundingBoxViewer, { PDFBoundingBoxViewerHandle } from '@/components/can-sr/PDFBoundingBoxViewer'
+import PDFBoundingBoxViewer, {
+  PDFBoundingBoxViewerHandle,
+} from '@/components/can-sr/PDFBoundingBoxViewer'
 import { Wand2 } from 'lucide-react'
 import { getAuthToken, getTokenType } from '@/lib/auth'
 import { useDictionary } from '@/app/[lang]/DictionaryProvider'
 import { needsHumanReviewForCriterion } from '@/components/can-sr/needsHumanReview'
+import {
+  ScreeningCitationContext,
+  extractHumanAnswer,
+  humanAnswerStatus,
+  resolveConfiguredValue,
+} from '@/components/can-sr/screening-citation-context'
 
 type ValidationEntry = { user: string; validated_at: string }
 
@@ -107,6 +115,12 @@ type CriteriaData = {
   questions: string[]
   possible_answers: string[][]
   additional_infos?: (string | null)[] // optional per-question extra guidance when available
+  items?: Array<{ answer_column?: string | null }>
+  citation_fields?: {
+    title?: string | null
+    abstract?: string | null
+    l1_include?: string[]
+  }
 }
 
 type LatestAgentRun = {
@@ -143,7 +157,9 @@ export default function CanSrL2ScreenViewPage() {
   const [citationIdList, setCitationIdList] = useState<number[]>([])
 
   // Autosave indicator per question
-  const [saveStatus, setSaveStatus] = useState<Record<number, 'idle' | 'saving' | 'saved' | 'error'>>({})
+  const [saveStatus, setSaveStatus] = useState<
+    Record<number, 'idle' | 'saving' | 'saved' | 'error'>
+  >({})
 
   // UI state: human selections keyed by question index
   const [selections, setSelections] = useState<Record<number, string>>({})
@@ -158,37 +174,54 @@ export default function CanSrL2ScreenViewPage() {
 
   // Agentic runs (screening_agent_runs) for this citation
   const [agentRuns, setAgentRuns] = useState<LatestAgentRun[]>([])
-  const [titleAbstractRuns, setTitleAbstractRuns] = useState<LatestAgentRun[]>([])
+  const [titleAbstractRuns, setTitleAbstractRuns] = useState<LatestAgentRun[]>(
+    [],
+  )
   const [, setLoadingRuns] = useState(false)
 
   // Per-criterion thresholds for this SR/step
-  const [thresholdByCriterionKey, setThresholdByCriterionKey] = useState<Record<string, number> | null>(null)
+  const [thresholdByCriterionKey, setThresholdByCriterionKey] = useState<Record<
+    string,
+    number
+  > | null>(null)
   const [validating, setValidating] = useState(false)
   const [userEmail, setUserEmail] = useState<string | null>(null)
 
   // Per-criterion AI status (keyed by question index)
-  const [criterionStatus, setCriterionStatus] = useState<Record<number, 'idle' | 'queued' | 'running' | 'done' | 'error'>>({})
+  const [criterionStatus, setCriterionStatus] = useState<
+    Record<number, 'idle' | 'queued' | 'running' | 'done' | 'error'>
+  >({})
   // Run-all progress
-  const [runAllProgress, setRunAllProgress] = useState<{ done: number; total: number } | null>(null)
+  const [runAllProgress, setRunAllProgress] = useState<{
+    done: number
+    total: number
+  } | null>(null)
   const [runningAll, setRunningAll] = useState(false)
   const [runAllError, setRunAllError] = useState<string | null>(null)
   const citationKey = `${srId || ''}:${citationId || ''}`
   const currentCitationKeyRef = useRef(citationKey)
   currentCitationKeyRef.current = citationKey
 
-  const l2Validations = useMemo(() => parseValidations((citation as any)?.l2_validations), [citation])
+  const l2Validations = useMemo(
+    () => parseValidations((citation as any)?.l2_validations),
+    [citation],
+  )
   const l2Checked = useMemo(() => {
     const me = String(userEmail || '')
     if (!me) return false
     return l2Validations.some((v) => v.user === me)
   }, [l2Validations, userEmail])
   const l2ValidationsSorted = useMemo(() => {
-    return [...l2Validations].sort((a, b) => String(b.validated_at || '').localeCompare(String(a.validated_at || '')))
+    return [...l2Validations].sort((a, b) =>
+      String(b.validated_at || '').localeCompare(String(a.validated_at || '')),
+    )
   }, [l2Validations])
 
   // Fulltext PDF viewer linkage
   const [fulltextCoords, setFulltextCoords] = useState<any[] | null>(null)
-  const [fulltextPages, setFulltextPages] = useState<{ width: number; height: number }[] | null>(null)
+  const [fulltextPages, setFulltextPages] = useState<
+    { width: number; height: number }[] | null
+  >(null)
   const [fulltextStr, setFulltextStr] = useState<string | null>(null)
   const viewerRef = useRef<PDFBoundingBoxViewerHandle | null>(null)
 
@@ -212,7 +245,9 @@ export default function CanSrL2ScreenViewPage() {
         const j = await res.json().catch(() => ({}))
         const t = res.ok ? j?.screening_thresholds : null
         const stepMap = t && typeof t === 'object' ? (t as any).l2 : null
-        setThresholdByCriterionKey(stepMap && typeof stepMap === 'object' ? stepMap : {})
+        setThresholdByCriterionKey(
+          stepMap && typeof stepMap === 'object' ? stepMap : {},
+        )
       } catch {
         setThresholdByCriterionKey({})
       }
@@ -259,9 +294,13 @@ export default function CanSrL2ScreenViewPage() {
       if (navSessionKey) {
         const raw = window.sessionStorage.getItem(navSessionKey)
         const parsed = raw ? JSON.parse(raw) : null
-        if (parsed && typeof parsed.startIndex === 'number') startedAt = parsed.startIndex
+        if (parsed && typeof parsed.startIndex === 'number')
+          startedAt = parsed.startIndex
         if (!raw && idx >= 0) {
-          window.sessionStorage.setItem(navSessionKey, JSON.stringify({ startIndex: idx, startedAt: Date.now() }))
+          window.sessionStorage.setItem(
+            navSessionKey,
+            JSON.stringify({ startIndex: idx, startedAt: Date.now() }),
+          )
           startedAt = idx
         }
       }
@@ -289,88 +328,119 @@ export default function CanSrL2ScreenViewPage() {
   }, [])
 
   // Load citation row (and ensure fulltext is extracted if missing)
-  const fetchCitationById = useCallback(async (id: string, signal?: AbortSignal) => {
-    if (!srId || !id) return
-    const requestKey = `${srId}:${id}`
-    setLoadingCitation(true)
-    try {
-      const headers = getAuthHeaders()
-      const res = await fetch(
-        `/api/can-sr/citations/get?sr_id=${encodeURIComponent(srId)}&citation_id=${encodeURIComponent(id)}`,
-        { method: 'GET', headers, signal },
-      )
-      const data = await res.json().catch(() => ({}))
-      if (currentCitationKeyRef.current !== requestKey) return
-      if (!res.ok) {
-        setError(
-          data?.error || data?.detail || `Failed to load citation (${res.status})`,
+  const fetchCitationById = useCallback(
+    async (id: string, signal?: AbortSignal) => {
+      if (!srId || !id) return
+      const requestKey = `${srId}:${id}`
+      setLoadingCitation(true)
+      try {
+        const headers = getAuthHeaders()
+        const res = await fetch(
+          `/api/can-sr/citations/get?sr_id=${encodeURIComponent(srId)}&citation_id=${encodeURIComponent(id)}`,
+          { method: 'GET', headers, signal },
         )
-        setCitation(null)
-      } else {
-        setCitation(data || null)
-
-        // Extract overlay data
-        const parseJson = (v: any) => {
-          if (!v) return null
-          try {
-            return typeof v === 'string' ? JSON.parse(v) : v
-          } catch {
-            return null
+        const data = await res.json().catch(() => ({}))
+        if (currentCitationKeyRef.current !== requestKey) return
+        if (!res.ok) {
+          setError(
+            data?.error ||
+              data?.detail ||
+              `Failed to load citation (${res.status})`,
+          )
+          setCitation(null)
+        } else {
+          setCitation(data || null)
+
+          // Extract overlay data
+          const parseJson = (v: any) => {
+            if (!v) return null
+            try {
+              return typeof v === 'string' ? JSON.parse(v) : v
+            } catch {
+              return null
+            }
           }
-        }
 
-        const ft = typeof (data as any).fulltext === 'string' ? (data as any).fulltext : null
-        setFulltextStr(ft)
-        const coordsAny = parseJson((data as any).fulltext_coords) ?? (data as any).fulltext_coords
-        setFulltextCoords(Array.isArray(coordsAny) ? coordsAny : null)
-        const pagesAny = parseJson((data as any).fulltext_pages) ?? (data as any).fulltext_pages
-        setFulltextPages(Array.isArray(pagesAny) ? pagesAny : null)
-
-        // If coords/pages missing, trigger backend extraction then refetch row
-        const needExtract =
-          !Array.isArray(coordsAny) || coordsAny.length === 0 || !Array.isArray(pagesAny) || pagesAny.length === 0
-        if (needExtract) {
-          try {
-            const res2 = await fetch(
-              `/api/can-sr/citations/full-text?action=extract&sr_id=${encodeURIComponent(
-                srId || '',
-              )}&citation_id=${encodeURIComponent(String(id || ''))}`,
-              { method: 'POST', headers, signal },
-            )
-            if (res2.ok && currentCitationKeyRef.current === requestKey) {
-              const res3 = await fetch(
-                `/api/can-sr/citations/get?sr_id=${encodeURIComponent(srId)}&citation_id=${encodeURIComponent(
-                  String(id),
-                )}`,
-                { headers, signal },
+          const ft =
+            typeof (data as any).fulltext === 'string'
+              ? (data as any).fulltext
+              : null
+          setFulltextStr(ft)
+          const coordsAny =
+            parseJson((data as any).fulltext_coords) ??
+            (data as any).fulltext_coords
+          setFulltextCoords(Array.isArray(coordsAny) ? coordsAny : null)
+          const pagesAny =
+            parseJson((data as any).fulltext_pages) ??
+            (data as any).fulltext_pages
+          setFulltextPages(Array.isArray(pagesAny) ? pagesAny : null)
+
+          // If coords/pages missing, trigger backend extraction then refetch row
+          const needExtract =
+            !Array.isArray(coordsAny) ||
+            coordsAny.length === 0 ||
+            !Array.isArray(pagesAny) ||
+            pagesAny.length === 0
+          if (needExtract) {
+            try {
+              const res2 = await fetch(
+                `/api/can-sr/citations/full-text?action=extract&sr_id=${encodeURIComponent(
+                  srId || '',
+                )}&citation_id=${encodeURIComponent(String(id || ''))}`,
+                { method: 'POST', headers, signal },
               )
-              const row2 = await res3.json().catch(() => ({}))
-              if (currentCitationKeyRef.current !== requestKey) return
-              if (res3.ok) setCitation(row2 || null)
-
-              const ft2 = typeof (row2 as any).fulltext === 'string' ? (row2 as any).fulltext : null
-              setFulltextStr(ft2)
-
-              const coordsAny2 = parseJson((row2 as any).fulltext_coords) ?? (row2 as any).fulltext_coords
-              setFulltextCoords(Array.isArray(coordsAny2) ? coordsAny2 : null)
-
-              const pagesAny2 = parseJson((row2 as any).fulltext_pages) ?? (row2 as any).fulltext_pages
-              setFulltextPages(Array.isArray(pagesAny2) ? pagesAny2 : null)
+              if (res2.ok && currentCitationKeyRef.current === requestKey) {
+                const res3 = await fetch(
+                  `/api/can-sr/citations/get?sr_id=${encodeURIComponent(srId)}&citation_id=${encodeURIComponent(
+                    String(id),
+                  )}`,
+                  { headers, signal },
+                )
+                const row2 = await res3.json().catch(() => ({}))
+                if (currentCitationKeyRef.current !== requestKey) return
+                if (res3.ok) setCitation(row2 || null)
+
+                const ft2 =
+                  typeof (row2 as any).fulltext === 'string'
+                    ? (row2 as any).fulltext
+                    : null
+                setFulltextStr(ft2)
+
+                const coordsAny2 =
+                  parseJson((row2 as any).fulltext_coords) ??
+                  (row2 as any).fulltext_coords
+                setFulltextCoords(Array.isArray(coordsAny2) ? coordsAny2 : null)
+
+                const pagesAny2 =
+                  parseJson((row2 as any).fulltext_pages) ??
+                  (row2 as any).fulltext_pages
+                setFulltextPages(Array.isArray(pagesAny2) ? pagesAny2 : null)
+              }
+            } catch (err: any) {
+              if (
+                err?.name === 'AbortError' ||
+                currentCitationKeyRef.current !== requestKey
+              )
+                return
+              console.warn('Failed to extract fulltext for overlay', err)
             }
-          } catch (err: any) {
-            if (err?.name === 'AbortError' || currentCitationKeyRef.current !== requestKey) return
-            console.warn('Failed to extract fulltext for overlay', err)
           }
         }
+      } catch (err: any) {
+        if (
+          err?.name === 'AbortError' ||
+          currentCitationKeyRef.current !== requestKey
+        )
+          return
+        console.error('Citation fetch error', err)
+        setError(err?.message || 'Network error while fetching citation')
+      } finally {
+        if (currentCitationKeyRef.current === requestKey)
+          setLoadingCitation(false)
       }
-    } catch (err: any) {
-      if (err?.name === 'AbortError' || currentCitationKeyRef.current !== requestKey) return
-      console.error('Citation fetch error', err)
-      setError(err?.message || 'Network error while fetching citation')
-    } finally {
-      if (currentCitationKeyRef.current === requestKey) setLoadingCitation(false)
-    }
-  }, [srId])
+    },
+    [srId],
+  )
 
   useEffect(() => {
     if (!srId || !citationId) return
@@ -397,7 +467,8 @@ export default function CanSrL2ScreenViewPage() {
 
   // Re-usable loader so we can refresh after triggering an agentic run.
   async function loadRuns(context?: { srId: string; citationId: string }) {
-    const runContext = context || (srId && citationId ? { srId, citationId } : null)
+    const runContext =
+      context || (srId && citationId ? { srId, citationId } : null)
     if (!runContext) return
     const requestKey = `${runContext.srId}:${runContext.citationId}`
     setLoadingRuns(true)
@@ -439,16 +510,22 @@ export default function CanSrL2ScreenViewPage() {
         )
         const data = await res.json().catch(() => ({}))
         if (currentCitationKeyRef.current !== requestKey) return
-        setTitleAbstractRuns(res.ok && Array.isArray(data?.runs) ? data.runs : [])
+        setTitleAbstractRuns(
+          res.ok && Array.isArray(data?.runs) ? data.runs : [],
+        )
       } catch {
-        if (currentCitationKeyRef.current === requestKey) setTitleAbstractRuns([])
+        if (currentCitationKeyRef.current === requestKey)
+          setTitleAbstractRuns([])
       }
     })()
     // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [srId, citationId])
 
   const runsByCriterion = useMemo(() => {
-    const by: Record<string, { screening?: LatestAgentRun; critical?: LatestAgentRun }> = {}
+    const by: Record<
+      string,
+      { screening?: LatestAgentRun; critical?: LatestAgentRun }
+    > = {}
     for (const r of agentRuns) {
       const key = String((r as any)?.criterion_key || '')
       if (!key) continue
@@ -491,7 +568,11 @@ export default function CanSrL2ScreenViewPage() {
 
           const parseBlock = (block: any) => {
             if (!block) {
-              return { questions: [], possible_answers: [], additional_infos: [] as (string | null)[] }
+              return {
+                questions: [],
+                possible_answers: [],
+                additional_infos: [] as (string | null)[],
+              }
             }
             if (!Array.isArray(block?.questions)) {
               // mapping form: { question: { option: description } }
@@ -519,17 +600,27 @@ export default function CanSrL2ScreenViewPage() {
                     const clean = String(d).replace(/\s+/g, ' ').trim()
                     return `- ${opt}: ${clean}`
                   })
-                  addInfos.push(guidanceParts.length ? guidanceParts.join('\n') : null)
+                  addInfos.push(
+                    guidanceParts.length ? guidanceParts.join('\n') : null,
+                  )
                 })
-                return { questions: qArr, possible_answers: ansArr, additional_infos: addInfos }
+                return {
+                  questions: qArr,
+                  possible_answers: ansArr,
+                  additional_infos: addInfos,
+                }
               } catch {
                 const questions = block?.questions || []
                 const possible_answers = block?.possible_answers || []
                 const additional_infos = block?.additional_infos || []
                 return {
                   questions: Array.isArray(questions) ? questions : [],
-                  possible_answers: Array.isArray(possible_answers) ? possible_answers : [],
-                  additional_infos: Array.isArray(additional_infos) ? additional_infos : [],
+                  possible_answers: Array.isArray(possible_answers)
+                    ? possible_answers
+                    : [],
+                  additional_infos: Array.isArray(additional_infos)
+                    ? additional_infos
+                    : [],
                 }
               }
             } else {
@@ -538,8 +629,12 @@ export default function CanSrL2ScreenViewPage() {
               const additional_infos = block?.additional_infos || []
               return {
                 questions: Array.isArray(questions) ? questions : [],
-                possible_answers: Array.isArray(possible_answers) ? possible_answers : [],
-                additional_infos: Array.isArray(additional_infos) ? additional_infos : [],
+                possible_answers: Array.isArray(possible_answers)
+                  ? possible_answers
+                  : [],
+                additional_infos: Array.isArray(additional_infos)
+                  ? additional_infos
+                  : [],
               }
             }
           }
@@ -551,13 +646,25 @@ export default function CanSrL2ScreenViewPage() {
           const l2Parsed = parseBlock(l2Block)
 
           const mergedQuestions = [...l1Parsed.questions, ...l2Parsed.questions]
-          const mergedAnswers = [...l1Parsed.possible_answers, ...l2Parsed.possible_answers]
-          const mergedInfos = [...l1Parsed.additional_infos, ...l2Parsed.additional_infos]
+          const mergedAnswers = [
+            ...l1Parsed.possible_answers,
+            ...l2Parsed.possible_answers,
+          ]
+          const mergedInfos = [
+            ...l1Parsed.additional_infos,
+            ...l2Parsed.additional_infos,
+          ]
+          const mergedItems = [
+            ...(Array.isArray(l1Block?.items) ? l1Block.items : []),
+            ...(Array.isArray(l2Block?.items) ? l2Block.items : []),
+          ]
 
           setCriteriaData({
             questions: mergedQuestions,
             possible_answers: mergedAnswers,
             additional_infos: mergedInfos,
+            items: mergedItems,
+            citation_fields: parsed?.citation_fields || {},
           })
           setSourceFlags([
             ...l1Parsed.questions.map(() => 'l1'),
@@ -586,7 +693,8 @@ export default function CanSrL2ScreenViewPage() {
 
     criteriaData.questions.forEach((q: string, idx: number) => {
       const llmCol = snakeCaseColumn(q)
-      const humanCol = humanScreenColumn(q)
+      const humanCol =
+        criteriaData.items?.[idx]?.answer_column || humanScreenColumn(q)
       const criterionKey = llmCol.replace(/^llm_/, '')
       const fulltextRun = runsByCriterion[criterionKey]?.screening
       const titleAbstractRun = titleAbstractRunsByCriterion[criterionKey]
@@ -594,10 +702,12 @@ export default function CanSrL2ScreenViewPage() {
       const runGuardrails = parseObject(fulltextRun?.guardrails)
       const runFulltextMd5 = String(runGuardrails?.fulltext_md5 || '')
       const isCurrentFulltextRun = Boolean(
-        fulltextRun && citationFulltextMd5 && runFulltextMd5 === citationFulltextMd5,
+        fulltextRun &&
+        citationFulltextMd5 &&
+        runFulltextMd5 === citationFulltextMd5,
       )
 
-      const humanRaw = (citation as any)?.[humanCol]
+      const humanRaw = resolveConfiguredValue(citation as any, humanCol)
       const llmRaw = (citation as any)?.[llmCol]
 
       // Parse possible JSON payloads from DB
@@ -621,12 +731,16 @@ export default function CanSrL2ScreenViewPage() {
       // human_* is historically shared by L1 and L2. Only accept values that
       // explicitly identify themselves as Full Text/L2; an untagged legacy value
       // is ambiguous and must not appear as a previous Full Text review.
-      const isFulltextHuman = humanParsed && typeof humanParsed === 'object' && (
-        (humanParsed as any).pipeline === 'fulltext' ||
-        String((humanParsed as any).screening_step || '').toLowerCase() === 'l2'
-      )
+      const isFulltextHuman =
+        humanParsed &&
+        typeof humanParsed === 'object' &&
+        ((humanParsed as any).pipeline === 'fulltext' ||
+          String((humanParsed as any).screening_step || '').toLowerCase() ===
+            'l2')
       if (isFulltextHuman && (humanParsed as any).selected !== undefined) {
         newSelections[idx] = (humanParsed as any).selected
+      } else if (sourceFlags[idx] === 'l2' && extractHumanAnswer(humanParsed)) {
+        newSelections[idx] = extractHumanAnswer(humanParsed)
       }
 
       // 2) For L1 questions: show hint from prior L1 screening (llm_*), but do not prefill from it
@@ -634,7 +748,8 @@ export default function CanSrL2ScreenViewPage() {
         // The shared llm_* column may have been overwritten by either pipeline.
         // Use the pipeline-specific run so a full-text result cannot masquerade
         // as the earlier Title/Abstract answer (or vice versa).
-        if (titleAbstractRun?.answer != null) newHints[idx] = String(titleAbstractRun.answer)
+        if (titleAbstractRun?.answer != null)
+          newHints[idx] = String(titleAbstractRun.answer)
       }
 
       // A Full Text AI panel must only exist when there is a persisted fulltext
@@ -643,9 +758,13 @@ export default function CanSrL2ScreenViewPage() {
         // The latest agent-run endpoint returns the audit answer, while the
         // persisted llm_* payload contains the evidence arrays. Do not discard
         // that payload merely because answer normalization changed the value.
-        const persistedFulltext = llmParsed && typeof llmParsed === 'object' &&
-          ((llmParsed as any).pipeline === 'fulltext' || !(llmParsed as any).pipeline) &&
-          (!(llmParsed as any).fulltext_md5 || (llmParsed as any).fulltext_md5 === citationFulltextMd5)
+        const persistedFulltext =
+          llmParsed &&
+          typeof llmParsed === 'object' &&
+          ((llmParsed as any).pipeline === 'fulltext' ||
+            !(llmParsed as any).pipeline) &&
+          (!(llmParsed as any).fulltext_md5 ||
+            (llmParsed as any).fulltext_md5 === citationFulltextMd5)
         newAiPanels[idx] = {
           ...(persistedFulltext ? llmParsed : {}),
           selected: fulltextRun.answer ?? '',
@@ -657,9 +776,13 @@ export default function CanSrL2ScreenViewPage() {
       }
 
       // For L2 questions: if no human selection present, allow llm_* to prefill the dropdown
-      const hasSelection = newSelections[idx] !== undefined && newSelections[idx] !== ''
+      const hasSelection =
+        newSelections[idx] !== undefined && newSelections[idx] !== ''
       if (sourceFlags[idx] === 'l2' && !hasSelection) {
-        const aiSelected = (newAiPanels[idx] && typeof newAiPanels[idx].selected === 'string') ? newAiPanels[idx].selected : null
+        const aiSelected =
+          newAiPanels[idx] && typeof newAiPanels[idx].selected === 'string'
+            ? newAiPanels[idx].selected
+            : null
         if (aiSelected) {
           newSelections[idx] = aiSelected
         }
@@ -670,7 +793,13 @@ export default function CanSrL2ScreenViewPage() {
     setAiPanels(newAiPanels)
     setPanelOpen(newPanelOpen)
     setHintByIndex(newHints)
-  }, [citation, criteriaData, sourceFlags, runsByCriterion, titleAbstractRunsByCriterion])
+  }, [
+    citation,
+    criteriaData,
+    sourceFlags,
+    runsByCriterion,
+    titleAbstractRunsByCriterion,
+  ])
 
   // Persist human classification
   async function postHumanClassifyPayload(
@@ -720,27 +849,51 @@ export default function CanSrL2ScreenViewPage() {
     if (!criteriaData) return
     const question = criteriaData.questions[questionIndex]
     const ok = await postHumanClassifyPayload(question, value)
-    setSaveStatus((prev) => ({ ...prev, [questionIndex]: ok ? 'saved' : 'error' }))
+    setSaveStatus((prev) => ({
+      ...prev,
+      [questionIndex]: ok ? 'saved' : 'error',
+    }))
   }
 
   // Run agentic fulltext screening for a single criterion (per-question AI button)
   async function classifyQuestion(
     questionIndex: number,
-    context?: { srId: string; citationId: string; questions: string[]; model: string },
+    context?: {
+      srId: string
+      citationId: string
+      questions: string[]
+      model: string
+    },
   ): Promise<boolean> {
-    const runContext = context || (srId && citationId && criteriaData
-      ? { srId, citationId, questions: criteriaData.questions, model: selectedModel }
-      : null)
+    const runContext =
+      context ||
+      (srId && citationId && criteriaData
+        ? {
+            srId,
+            citationId,
+            questions: criteriaData.questions,
+            model: selectedModel,
+          }
+        : null)
     if (!runContext) return false
     const requestKey = `${runContext.srId}:${runContext.citationId}`
     if (currentCitationKeyRef.current !== requestKey) return false
     const q = runContext.questions[questionIndex]
     if (!q) return false
     setRunAllError(null)
-    const ck = q.trim().toLowerCase().replace(/[^\w]+/g, '_').replace(/_+/g, '_').replace(/^_+|_+$/g, '').slice(0, 56)
+    const ck = q
+      .trim()
+      .toLowerCase()
+      .replace(/[^\w]+/g, '_')
+      .replace(/_+/g, '_')
+      .replace(/^_+|_+$/g, '')
+      .slice(0, 56)
     setCriterionStatus((prev) => ({ ...prev, [questionIndex]: 'running' }))
     try {
-      const headers = { 'Content-Type': 'application/json', ...getAuthHeaders() }
+      const headers = {
+        'Content-Type': 'application/json',
+        ...getAuthHeaders(),
+      }
       const res = await fetch('/api/can-sr/screen/fulltext/run', {
         method: 'POST',
         headers,
@@ -755,8 +908,16 @@ export default function CanSrL2ScreenViewPage() {
         }),
       })
       const runData = await res.json().catch(() => ({}))
-      if (!res.ok) throw new Error(runData?.detail || runData?.error || `AI request failed (${res.status})`)
-      if (String(runData?.citation_id ?? runContext.citationId) !== runContext.citationId) {
+      if (!res.ok)
+        throw new Error(
+          runData?.detail ||
+            runData?.error ||
+            `AI request failed (${res.status})`,
+        )
+      if (
+        String(runData?.citation_id ?? runContext.citationId) !==
+        runContext.citationId
+      ) {
         throw new Error('AI response did not match the requested citation')
       }
       if (currentCitationKeyRef.current !== requestKey) return false
@@ -770,20 +931,33 @@ export default function CanSrL2ScreenViewPage() {
           { method: 'GET', headers: getAuthHeaders() },
         )
         const citData = await citRes.json().catch(() => ({}))
-        if (currentCitationKeyRef.current === requestKey && citRes.ok && citData) {
+        if (
+          currentCitationKeyRef.current === requestKey &&
+          citRes.ok &&
+          citData
+        ) {
           const llmRaw = citData[llmColName]
           let llmParsed = llmRaw
           if (typeof llmRaw === 'string') {
-            try { llmParsed = JSON.parse(llmRaw) } catch { llmParsed = llmRaw }
+            try {
+              llmParsed = JSON.parse(llmRaw)
+            } catch {
+              llmParsed = llmRaw
+            }
           }
           if (llmParsed && typeof llmParsed === 'object') {
             setAiPanels((prev) => ({ ...prev, [questionIndex]: llmParsed }))
           }
         }
-      } catch { /* best-effort */ }
+      } catch {
+        /* best-effort */
+      }
 
       // Refresh agent runs (no PDF state changes)
-      await loadRuns({ srId: runContext.srId, citationId: runContext.citationId })
+      await loadRuns({
+        srId: runContext.srId,
+        citationId: runContext.citationId,
+      })
       if (currentCitationKeyRef.current !== requestKey) return false
       setCriterionStatus((prev) => ({ ...prev, [questionIndex]: 'done' }))
       return true
@@ -791,7 +965,9 @@ export default function CanSrL2ScreenViewPage() {
       console.error('Classify API error', err)
       if (currentCitationKeyRef.current === requestKey) {
         setCriterionStatus((prev) => ({ ...prev, [questionIndex]: 'error' }))
-        setRunAllError(err instanceof Error ? err.message : 'Full Text AI failed')
+        setRunAllError(
+          err instanceof Error ? err.message : 'Full Text AI failed',
+        )
       }
       return false
     }
@@ -800,13 +976,22 @@ export default function CanSrL2ScreenViewPage() {
   // Run all criteria sequentially (Run All AI button)
   async function runAllAI() {
     if (!srId || !citationId || !criteriaData || runningAll) return
-    const context = { srId, citationId, questions: [...criteriaData.questions], model: selectedModel }
+    const context = {
+      srId,
+      citationId,
+      questions: [...criteriaData.questions],
+      model: selectedModel,
+    }
     const requestKey = `${srId}:${citationId}`
     const total = context.questions.length
     setRunningAll(true)
     setRunAllProgress({ done: 0, total })
     setRunAllError(null)
-    setCriterionStatus(Object.fromEntries(context.questions.map((_, index) => [index, 'queued'])))
+    setCriterionStatus(
+      Object.fromEntries(
+        context.questions.map((_, index) => [index, 'queued']),
+      ),
+    )
     try {
       const failures: number[] = []
       for (let index = 0; index < context.questions.length; index++) {
@@ -830,7 +1015,11 @@ export default function CanSrL2ScreenViewPage() {
     } catch (err: any) {
       console.error('Run All AI error', err)
       setRunAllError(err?.message || 'Run All AI failed')
-      setCriterionStatus((prev) => Object.fromEntries(Object.keys(prev).map((key) => [Number(key), 'error'])))
+      setCriterionStatus((prev) =>
+        Object.fromEntries(
+          Object.keys(prev).map((key) => [Number(key), 'error']),
+        ),
+      )
     }
   }
 
@@ -876,46 +1065,81 @@ export default function CanSrL2ScreenViewPage() {
   const scrollToArtifact = (kind: 'table' | 'figure', idx: number) => {
     const list = kind === 'table' ? parsedTables : parsedFigures
     const item = list.find((x: any) => Number(x?.index) === Number(idx))
-    console.log('[artifact-click]', { kind, idx, hasViewer: !!viewerRef.current, item })
     if (!item || !viewerRef.current) return
     const bbox = item?.bounding_box
     // We store normalized boxes as an array of {page,x,y,width,height}
-    const first = Array.isArray(bbox) ? bbox[0] : null
-    console.log('[artifact-bbox]', { kind, idx, bbox, first })
+    const first = Array.isArray(bbox)
+      ? bbox.find(
+          (box: any) =>
+            box &&
+            typeof box === 'object' &&
+            Number.isFinite(
+              Number(box?.page ?? box?.page_number ?? box?.pageNum),
+            ),
+        )
+      : null
     if (!first) return
     viewerRef.current.scrollToCoord(first)
   }
 
   const workspace = useMemo(() => {
-    if (error)
-      return <div className="text-sm text-red-600">{error}</div>
+    if (error) return <div className="text-sm text-red-600">{error}</div>
     if (loadingCitation)
-      return <div className="text-sm text-gray-600">{dict.screening.loadingCitation}</div>
+      return (
+        <div className="text-sm text-gray-600">
+          {dict.screening.loadingCitation}
+        </div>
+      )
     if (!citation)
-      return <div className="text-sm text-gray-600">{dict.screening.citationNotFound}</div>
+      return (
+        <div className="text-sm text-gray-600">
+          {dict.screening.citationNotFound}
+        </div>
+      )
 
     return (
-      <PDFBoundingBoxViewer
-        srId={srId || ''}
-        citationId={citationId ?? ''}
-        conversionId={null}
-        fileName={String(
-          citation?.title ||
-          citation?.citation ||
-          citation?.article_title ||
-          'Full text'
-        )}
-        coords={fulltextCoords || []}
-        pages={fulltextPages || []}
-        aiPanels={panelsKeyed.panels}
-        panelOpen={panelsKeyed.open}
-        artifacts={{ tables: parsedTables, figures: parsedFigures }}
-        fulltext={fulltextStr || ''}
-        defaultFitToWidth={true}
-        ref={viewerRef}
-      />
+      <div className="space-y-4">
+        <ScreeningCitationContext
+          citation={citation}
+          fields={criteriaData?.citation_fields}
+          showTitleAbstract={false}
+        />
+        <PDFBoundingBoxViewer
+          srId={srId || ''}
+          citationId={citationId ?? ''}
+          conversionId={null}
+          fileName={String(
+            citation?.title ||
+              citation?.citation ||
+              citation?.article_title ||
+              'Full text',
+          )}
+          coords={fulltextCoords || []}
+          pages={fulltextPages || []}
+          aiPanels={panelsKeyed.panels}
+          panelOpen={panelsKeyed.open}
+          artifacts={{ tables: parsedTables, figures: parsedFigures }}
+          fulltext={fulltextStr || ''}
+          defaultFitToWidth={true}
+          ref={viewerRef}
+        />
+      </div>
     )
-  }, [citation, loadingCitation, srId, citationId, fulltextCoords, fulltextPages, fulltextStr, panelsKeyed, parsedTables, parsedFigures, dict, error])
+  }, [
+    citation,
+    criteriaData,
+    loadingCitation,
+    srId,
+    citationId,
+    fulltextCoords,
+    fulltextPages,
+    fulltextStr,
+    panelsKeyed,
+    parsedTables,
+    parsedFigures,
+    dict,
+    error,
+  ])
 
   if (!srId || !citationId) {
     // guard - redirect already handled in effect but keep safe render
@@ -937,23 +1161,29 @@ export default function CanSrL2ScreenViewPage() {
         }
       />
 
-      <main className="mx-auto max-w-8xl px-3 py-3">
-
+      <main className="max-w-8xl mx-auto px-3 py-3">
         {/* Filter/session progress bar */}
         {progressInfo.total > 0 && progressInfo.idx >= 0 ? (
           <div className="mb-3 rounded-md border border-gray-200 bg-white p-3">
             <div className="flex items-center justify-between text-xs text-gray-600">
               <div>
-                Progress: <span className="font-medium">{progressInfo.idx + 1}</span> / {progressInfo.total}
+                Progress:{' '}
+                <span className="font-medium">{progressInfo.idx + 1}</span> /{' '}
+                {progressInfo.total}
               </div>
               <div>
-                Since start: <span className="font-medium">{Math.max(0, progressInfo.idx - progressInfo.startedAt + 1)}</span>
+                Since start:{' '}
+                <span className="font-medium">
+                  {Math.max(0, progressInfo.idx - progressInfo.startedAt + 1)}
+                </span>
               </div>
             </div>
             <div className="mt-2 h-2 w-full overflow-hidden rounded bg-gray-200">
               <div
                 className="h-2 bg-emerald-600"
-                style={{ width: `${Math.min(100, ((progressInfo.idx + 1) / progressInfo.total) * 100)}%` }}
+                style={{
+                  width: `${Math.min(100, ((progressInfo.idx + 1) / progressInfo.total) * 100)}%`,
+                }}
               />
             </div>
           </div>
@@ -961,31 +1191,53 @@ export default function CanSrL2ScreenViewPage() {
 
         <div className="grid grid-cols-12 gap-3">
           {/* Workspace (left) */}
-          <div className="col-span-9">
-              {workspace}
-          </div>
+          <div className="col-span-9">{workspace}</div>
 
           {/* Selection sidebar (right) */}
           <aside className="col-span-3">
-            <div className="h-full space-y-4 rounded-lg border border-gray-200 bg-white p-4 shadow-sm flex flex-col">
+            <div className="flex h-full flex-col space-y-4 rounded-lg border border-gray-200 bg-white p-4 shadow-sm">
               <div>
-                <h4 className="text-xl font-semibold text-gray-900 text-center">{dict.screening.screeningQuestions}</h4>
+                <h4 className="text-center text-xl font-semibold text-gray-900">
+                  {dict.screening.screeningQuestions}
+                </h4>
               </div>
 
               {/* Run All AI button */}
               <div className="flex items-center justify-between gap-2">
                 <span className="text-xs text-gray-500">
-                  {runAllProgress ? `${runAllProgress.done}/${runAllProgress.total} criteria` : ''}
+                  {runAllProgress
+                    ? `${runAllProgress.done}/${runAllProgress.total} criteria`
+                    : ''}
                 </span>
                 <button
                   onClick={onRunAllAI}
-                  disabled={runningAll || !criteriaData || criteriaData.questions.length === 0 || !(citation as any)?.fulltext_md5}
+                  disabled={
+                    runningAll ||
+                    !criteriaData ||
+                    criteriaData.questions.length === 0 ||
+                    !(citation as any)?.fulltext_md5
+                  }
                   className="inline-flex items-center gap-1 rounded-md border px-2 py-1 text-xs hover:bg-gray-50 disabled:cursor-not-allowed disabled:opacity-50"
                 >
                   {runningAll ? (
-                    <svg className="h-3 w-3 animate-spin" viewBox="0 0 24 24" fill="none">
-                      <circle className="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" strokeWidth="4" />
-                      <path className="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8v8H4z" />
+                    <svg
+                      className="h-3 w-3 animate-spin"
+                      viewBox="0 0 24 24"
+                      fill="none"
+                    >
+                      <circle
+                        className="opacity-25"
+                        cx="12"
+                        cy="12"
+                        r="10"
+                        stroke="currentColor"
+                        strokeWidth="4"
+                      />
+                      <path
+                        className="opacity-75"
+                        fill="currentColor"
+                        d="M4 12a8 8 0 018-8v8H4z"
+                      />
                     </svg>
                   ) : (
                     <Wand2 className="h-3 w-3" />
@@ -995,129 +1247,219 @@ export default function CanSrL2ScreenViewPage() {
                     : 'Run All AI'}
                 </button>
               </div>
-              {runAllError ? <div className="text-xs text-red-600">{runAllError}</div> : null}
+              {runAllError ? (
+                <div className="text-xs text-red-600">{runAllError}</div>
+              ) : null}
 
               {loadingCriteria ? (
-                <div className="text-sm text-gray-600">{dict.screening.loadingCriteria}</div>
+                <div className="text-sm text-gray-600">
+                  {dict.screening.loadingCriteria}
+                </div>
               ) : !criteriaData || criteriaData.questions.length === 0 ? (
                 <div className="text-sm text-gray-600">
                   {dict.screening.noL2Criteria}
                 </div>
               ) : (
-                <div className="rounded-md border border-gray-100 p-3 h-[680px] overflow-y-auto">
+                <div className="h-[680px] overflow-y-auto rounded-md border border-gray-100 p-3">
                   <div className="space-y-4">
-                  {criteriaData.questions.map((q, idx) => {
-                    const options = criteriaData.possible_answers[idx] || []
-                    const current = selections[idx] ?? ''
-                    const aiData = aiPanels[idx]
-                    const aiSelected =
-                      aiData && aiData.selected ? aiData.selected : undefined
-
-                    // Per-question highlight aligned with list/back-end logic.
-                    // - Highlight when this criterion triggers review (low confidence OR critical disagreement OR guardrail issue)
-                    // - BUT do not highlight if this criterion is a confident-exclude (exclude + conf>=thr + critical agrees)
-                    const criterionKey = q
-                      ? q
-                          .trim()
-                          .toLowerCase()
-                          .replace(/[^\w]+/g, '_')
-                          .replace(/_+/g, '_')
-                          .replace(/^_+|_+$/g, '')
-                          .slice(0, 56)
-                      : ''
-
-                    const r = (runsByCriterion as any)?.[criterionKey] || {}
-                    const currentPdfMd5 = String((citation as any)?.fulltext_md5 || '')
-                    const runMatchesCurrentPdf = (run: any) => {
-                      const guardrails = parseObject(run?.guardrails)
-                      return Boolean(run && currentPdfMd5 && String(guardrails?.fulltext_md5 || '') === currentPdfMd5)
-                    }
-                    const scr = runMatchesCurrentPdf(r.screening) ? r.screening : undefined
-                    const crit = runMatchesCurrentPdf(r.critical) ? r.critical : undefined
-                    const perThrRaw = thresholdByCriterionKey ? Number((thresholdByCriterionKey as any)[criterionKey]) : NaN
-                    const thr = Number.isFinite(perThrRaw) ? Math.max(0, Math.min(1, perThrRaw)) : 0.9
-
-                    const { needsHuman, criticalDisagrees: critDisagrees } = needsHumanReviewForCriterion({
-                      threshold: thr,
-                      screening: scr,
-                      critical: crit,
-                    })
-                    const hasAgentic = !!scr || !!crit
-
-                    // Prefer agentic screening run when available to avoid mismatches
-                    const displayConfidence =
-                      Number.isFinite(Number((scr as any)?.confidence))
+                    {criteriaData.questions.map((q, idx) => {
+                      const options = criteriaData.possible_answers[idx] || []
+                      const answerColumn =
+                        criteriaData.items?.[idx]?.answer_column || null
+                      const answerStatus = humanAnswerStatus(
+                        citation!,
+                        answerColumn,
+                        options,
+                      )
+                      const current = selections[idx] ?? ''
+                      const aiData = aiPanels[idx]
+                      const aiSelected =
+                        aiData && aiData.selected ? aiData.selected : undefined
+
+                      // Per-question highlight aligned with list/back-end logic.
+                      // - Highlight when this criterion triggers review (low confidence OR critical disagreement OR guardrail issue)
+                      // - BUT do not highlight if this criterion is a confident-exclude (exclude + conf>=thr + critical agrees)
+                      const criterionKey = q
+                        ? q
+                            .trim()
+                            .toLowerCase()
+                            .replace(/[^\w]+/g, '_')
+                            .replace(/_+/g, '_')
+                            .replace(/^_+|_+$/g, '')
+                            .slice(0, 56)
+                        : ''
+
+                      const r = (runsByCriterion as any)?.[criterionKey] || {}
+                      const currentPdfMd5 = String(
+                        (citation as any)?.fulltext_md5 || '',
+                      )
+                      const runMatchesCurrentPdf = (run: any) => {
+                        const guardrails = parseObject(run?.guardrails)
+                        return Boolean(
+                          run &&
+                          currentPdfMd5 &&
+                          String(guardrails?.fulltext_md5 || '') ===
+                            currentPdfMd5,
+                        )
+                      }
+                      const scr = runMatchesCurrentPdf(r.screening)
+                        ? r.screening
+                        : undefined
+                      const crit = runMatchesCurrentPdf(r.critical)
+                        ? r.critical
+                        : undefined
+                      const perThrRaw = thresholdByCriterionKey
+                        ? Number((thresholdByCriterionKey as any)[criterionKey])
+                        : NaN
+                      const thr = Number.isFinite(perThrRaw)
+                        ? Math.max(0, Math.min(1, perThrRaw))
+                        : 0.9
+
+                      const { needsHuman, criticalDisagrees: critDisagrees } =
+                        needsHumanReviewForCriterion({
+                          threshold: thr,
+                          screening: scr,
+                          critical: crit,
+                        })
+                      const hasAgentic = !!scr || !!crit
+
+                      // Prefer agentic screening run when available to avoid mismatches
+                      const displayConfidence = Number.isFinite(
+                        Number((scr as any)?.confidence),
+                      )
                         ? Number((scr as any)?.confidence)
                         : Number(aiData?.confidence)
 
-                    const aiExpl =
-                      (typeof (aiData as any)?.explanation === 'string' ? String((aiData as any).explanation) : '') ||
-                      (typeof (aiData as any)?.rationale === 'string' ? String((aiData as any).rationale) : '') ||
-                      (typeof (aiData as any)?.llm_raw === 'string' ? String((aiData as any).llm_raw) : '')
-
-                    const scrRationale = typeof (scr as any)?.rationale === 'string' ? String((scr as any).rationale) : ''
-                    const displayExplanationRaw = (scrRationale || aiExpl || '').trim()
-                    const displayExplanation =
-                      extractXmlTag(displayExplanationRaw, 'rationale') || displayExplanationRaw
-                    const screeningGuardrails = parseObject((scr as any)?.guardrails)
-                    const missingFields = Array.isArray(aiData?.missing_fields)
-                      ? aiData.missing_fields.map((field: unknown) => String(field))
-                      : Array.isArray(screeningGuardrails?.missing_fields)
-                        ? screeningGuardrails.missing_fields.map((field: unknown) => String(field))
-                        : []
-                    const missingLabels = missingFields.map((field: string) => {
-                      if (field === 'answer') return 'selection'
-                      if (field === 'confidence') return 'confidence'
-                      if (field === 'rationale') return 'rationale'
-                      return field
-                    })
-
-                    return (
-                      <div
-                        key={idx}
-                        className={
-                          'rounded-md border-2 p-3 ' + (needsHuman ? 'border-amber-400' : 'border-gray-100')
-                        }
-                      >
-                        <div className="flex items-start justify-between">
-                          <div className="flex-1">
-                            <p className="text-sm font-medium text-gray-800">
-                              {q}
-                            </p>
-
-                            {sourceFlags[idx] === 'l1' ? (
-                              <p className="mt-1 text-xs text-gray-500">
-                                {dict.screening.titleAbstractAnswer} {hintByIndex[idx] ?? dict.screening.none}
+                      const aiExpl =
+                        (typeof (aiData as any)?.explanation === 'string'
+                          ? String((aiData as any).explanation)
+                          : '') ||
+                        (typeof (aiData as any)?.rationale === 'string'
+                          ? String((aiData as any).rationale)
+                          : '') ||
+                        (typeof (aiData as any)?.llm_raw === 'string'
+                          ? String((aiData as any).llm_raw)
+                          : '')
+
+                      const scrRationale =
+                        typeof (scr as any)?.rationale === 'string'
+                          ? String((scr as any).rationale)
+                          : ''
+                      const displayExplanationRaw = (
+                        scrRationale ||
+                        aiExpl ||
+                        ''
+                      ).trim()
+                      const displayExplanation =
+                        extractXmlTag(displayExplanationRaw, 'rationale') ||
+                        displayExplanationRaw
+                      const screeningGuardrails = parseObject(
+                        (scr as any)?.guardrails,
+                      )
+                      const missingFields = Array.isArray(
+                        aiData?.missing_fields,
+                      )
+                        ? aiData.missing_fields.map((field: unknown) =>
+                            String(field),
+                          )
+                        : Array.isArray(screeningGuardrails?.missing_fields)
+                          ? screeningGuardrails.missing_fields.map(
+                              (field: unknown) => String(field),
+                            )
+                          : []
+                      const missingLabels = missingFields.map(
+                        (field: string) => {
+                          if (field === 'answer') return 'selection'
+                          if (field === 'confidence') return 'confidence'
+                          if (field === 'rationale') return 'rationale'
+                          return field
+                        },
+                      )
+
+                      return (
+                        <div
+                          key={idx}
+                          className={
+                            'rounded-md border-2 p-3 ' +
+                            (needsHuman
+                              ? 'border-amber-400'
+                              : 'border-gray-100')
+                          }
+                        >
+                          <div className="flex items-start justify-between">
+                            <div className="flex-1">
+                              <p className="text-sm font-medium text-gray-800">
+                                {q}
                               </p>
-                            ) : null}
-
-                            <select
-                              value={current || (aiSelected ?? '')}
-                              onChange={(e) =>
-                                onSelectOption(idx, e.target.value)
-                              }
-                              className="mt-2 w-full rounded-md border border-gray-200 bg-white px-3 py-2 text-sm"
-                            >
-                              <option value="">-- select --</option>
-                              {options.map((opt: string) => (
-                                <option key={opt} value={opt}>
-                                  {opt}
-                                </option>
-                              ))}
-                            </select>
-                          </div>
 
-                          <div className="ml-3 flex flex-col items-end space-y-2">
+                              {sourceFlags[idx] === 'l1' ? (
+                                <p className="mt-1 text-xs text-gray-500">
+                                  {dict.screening.titleAbstractAnswer}{' '}
+                                  {hintByIndex[idx] ?? dict.screening.none}
+                                </p>
+                              ) : null}
+
+                              <select
+                                value={current || (aiSelected ?? '')}
+                                onChange={(e) =>
+                                  onSelectOption(idx, e.target.value)
+                                }
+                                className="mt-2 w-full rounded-md border border-gray-200 bg-white px-3 py-2 text-sm"
+                              >
+                                <option value="">-- select --</option>
+                                {options.map((opt: string) => (
+                                  <option key={opt} value={opt}>
+                                    {opt}
+                                  </option>
+                                ))}
+                              </select>
+                              {answerStatus !== 'unconfigured' &&
+                              answerStatus !== 'matched' ? (
+                                <p
+                                  className="mt-1 text-xs text-amber-700"
+                                  role="status"
+                                >
+                                  Human answer column “{answerColumn}”:{' '}
+                                  {answerStatus === 'missing'
+                                    ? 'not found in this citation'
+                                    : answerStatus === 'blank'
+                                      ? 'blank for this citation'
+                                      : 'value does not match an available answer'}
+                                  .
+                                </p>
+                              ) : null}
+                            </div>
+
+                            <div className="ml-3 flex flex-col items-end space-y-2">
                               <button
                                 onClick={() => classifyQuestion(idx)}
-                                disabled={criterionStatus[idx] === 'running' || runningAll || !(citation as any)?.fulltext_md5}
+                                disabled={
+                                  criterionStatus[idx] === 'running' ||
+                                  runningAll ||
+                                  !(citation as any)?.fulltext_md5
+                                }
                                 className="rounded-md border px-2 py-1 text-xs hover:bg-gray-50 disabled:cursor-not-allowed disabled:opacity-50"
                               >
                                 <span className="inline-flex items-center gap-1">
                                   {criterionStatus[idx] === 'running' ? (
-                                    <svg className="h-3 w-3 animate-spin" viewBox="0 0 24 24" fill="none">
-                                      <circle className="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" strokeWidth="4" />
-                                      <path className="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8v8H4z" />
+                                    <svg
+                                      className="h-3 w-3 animate-spin"
+                                      viewBox="0 0 24 24"
+                                      fill="none"
+                                    >
+                                      <circle
+                                        className="opacity-25"
+                                        cx="12"
+                                        cy="12"
+                                        r="10"
+                                        stroke="currentColor"
+                                        strokeWidth="4"
+                                      />
+                                      <path
+                                        className="opacity-75"
+                                        fill="currentColor"
+                                        d="M4 12a8 8 0 018-8v8H4z"
+                                      />
                                     </svg>
                                   ) : (
                                     <Wand2 className="h-3 w-3" />
@@ -1127,186 +1469,247 @@ export default function CanSrL2ScreenViewPage() {
                               </button>
 
                               {criterionStatus[idx] === 'running' ? (
-                                <span className="text-[10px] text-blue-600">Running…</span>
+                                <span className="text-[10px] text-blue-600">
+                                  Running…
+                                </span>
                               ) : criterionStatus[idx] === 'queued' ? (
-                                <span className="text-[10px] text-gray-500">Queued</span>
+                                <span className="text-[10px] text-gray-500">
+                                  Queued
+                                </span>
                               ) : criterionStatus[idx] === 'done' ? (
-                                <span className="text-[10px] text-emerald-600">Completed</span>
+                                <span className="text-[10px] text-emerald-600">
+                                  Completed
+                                </span>
                               ) : criterionStatus[idx] === 'error' ? (
-                                <span className="text-[10px] text-red-600">Error</span>
+                                <span className="text-[10px] text-red-600">
+                                  Error
+                                </span>
                               ) : saveStatus[idx] === 'saving' ? (
-                                <span className="text-[10px] text-gray-500">{dict.common.save}...</span>
+                                <span className="text-[10px] text-gray-500">
+                                  {dict.common.save}...
+                                </span>
                               ) : saveStatus[idx] === 'saved' ? (
-                                <span className="text-[10px] text-emerald-600">{dict.common.done}</span>
+                                <span className="text-[10px] text-emerald-600">
+                                  {dict.common.done}
+                                </span>
                               ) : saveStatus[idx] === 'error' ? (
-                                <span className="text-[10px] text-red-600">{dict.common.error}</span>
-                              ) : null}
-                          </div>
-                        </div>
-
-                        {/* AI panel: collapsible with evidence chips */}
-                        {aiData ? (
-                          <div className="mt-3">
-                            <div
-                              onClick={() =>
-                                setPanelOpen((prev) => ({
-                                  ...prev,
-                                  [idx]: !Boolean(prev[idx]),
-                                }))
-                              }
-                              style={{ cursor: 'pointer' }}
-                              className="flex items-center justify-between rounded-md bg-gray-50 px-3 py-2"
-                            >
-                              <div className="text-sm">
-                                {dict.screening.aiSuggests}{' '}
-                                <span
-                                  className={
-                                    'ml-1 text-sm font-medium ' +
-                                    (String(aiData.selected ?? '')
-                                      .toLowerCase()
-                                      .includes('(exclude)')
-                                      ? 'text-red-600'
-                                      : 'text-emerald-600')
-                                  }
-                                >
-                                  {aiData.selected ?? dict.screening.noSelection}
+                                <span className="text-[10px] text-red-600">
+                                  {dict.common.error}
                                 </span>
-                              </div>
-                              <div className="text-xs text-gray-500">
-                                {panelOpen[idx] ? dict.screening.minimize : dict.screening.maximize}
-                              </div>
+                              ) : null}
                             </div>
+                          </div>
 
-                            {panelOpen[idx] ? (
-                              <div className="mt-2 rounded-md border border-gray-100 bg-white p-3 text-sm whitespace-pre-wrap text-gray-800">
-                                {missingLabels.length > 0 ? (
-                                  <div
-                                    className="mb-3 rounded-md border border-amber-300 bg-amber-50 p-2 text-xs text-amber-900"
-                                    role="alert"
+                          {/* AI panel: collapsible with evidence chips */}
+                          {aiData ? (
+                            <div className="mt-3">
+                              <div
+                                onClick={() =>
+                                  setPanelOpen((prev) => ({
+                                    ...prev,
+                                    [idx]: !Boolean(prev[idx]),
+                                  }))
+                                }
+                                style={{ cursor: 'pointer' }}
+                                className="flex items-center justify-between rounded-md bg-gray-50 px-3 py-2"
+                              >
+                                <div className="text-sm">
+                                  {dict.screening.aiSuggests}{' '}
+                                  <span
+                                    className={
+                                      'ml-1 text-sm font-medium ' +
+                                      (String(aiData.selected ?? '')
+                                        .toLowerCase()
+                                        .includes('(exclude)')
+                                        ? 'text-red-600'
+                                        : 'text-emerald-600')
+                                    }
                                   >
-                                    The AI response was incomplete after one automatic repair. Missing:{' '}
-                                    <strong>{missingLabels.join(', ')}</strong>. Review the available fields and rerun this criterion.
-                                  </div>
-                                ) : null}
-                                <div className="mt-2">
-                                  <strong>{dict.screening.confidence}</strong>{' '}
-                                  {missingFields.includes('confidence')
-                                    ? 'Not returned by AI'
-                                    : Number.isFinite(displayConfidence)
-                                    ? String(displayConfidence)
-                                    : String(aiData.confidence ?? '')}
+                                    {aiData.selected ??
+                                      dict.screening.noSelection}
+                                  </span>
                                 </div>
-                                <div className="mt-2">
-                                  <strong>{dict.screening.explanation}</strong>
-                                  <div className="mt-1 text-sm text-gray-700">
-                                    {missingFields.includes('rationale')
-                                      ? 'Rationale was not returned by the AI after repair.'
-                                      : displayExplanation || dict.screening.noExplanation}
-                                  </div>
+                                <div className="text-xs text-gray-500">
+                                  {panelOpen[idx]
+                                    ? dict.screening.minimize
+                                    : dict.screening.maximize}
                                 </div>
+                              </div>
 
-                                {hasAgentic && crit ? (
-                                  <div className="mt-3 rounded-md border border-gray-100 bg-gray-50 p-2 text-xs text-gray-700">
-                                    <div className="mt-1 font-semibold text-gray-800">
-                                      {critDisagrees ? (
-                                        <span className="text-amber-700">
-                                          Critical review recommends a different answer: {String((crit as any)?.answer ?? '—')}
-                                        </span>
-                                      ) : (
-                                        <span className="text-emerald-700">Critical review supports the screening answer</span>
-                                      )}
+                              {panelOpen[idx] ? (
+                                <div className="mt-2 rounded-md border border-gray-100 bg-white p-3 text-sm whitespace-pre-wrap text-gray-800">
+                                  {missingLabels.length > 0 ? (
+                                    <div
+                                      className="mb-3 rounded-md border border-amber-300 bg-amber-50 p-2 text-xs text-amber-900"
+                                      role="alert"
+                                    >
+                                      The AI response was incomplete after one
+                                      automatic repair. Missing:{' '}
+                                      <strong>
+                                        {missingLabels.join(', ')}
+                                      </strong>
+                                      . Review the available fields and rerun
+                                      this criterion.
                                     </div>
-                                    <div>Judgment confidence: {String((crit as any)?.confidence ?? '—')}</div>
-                                  </div>
-                                ) : null}
-                                {Array.isArray(aiData?.evidence_sentences) && aiData.evidence_sentences.length > 0 ? (
+                                  ) : null}
                                   <div className="mt-2">
-                                    <strong>{dict.screening.evidence}</strong>
-                                    <div className="mt-1 flex flex-wrap gap-1">
-                                      {aiData.evidence_sentences.map((item: any, k: number) => {
-                                        const isCoord = item && typeof item === 'object'
-                                        const label = isCoord
-                                          ? `${dict.screening.page} ${String(item.page ?? item.page_number ?? item.pageNum ?? '?')}${item.text ? `: ${String(item.text).slice(0, 80)}` : ''}`
-                                          : `${dict.screening.sentence} ${String(item)}`
-                                        const onClick = () => {
-                                          if (!viewerRef.current) return
-                                          if (isCoord) {
-                                            viewerRef.current.scrollToCoord(item)
-                                          } else {
-                                            const idxNum = Number(item)
-                                            if (!Number.isNaN(idxNum)) {
-                                              viewerRef.current.scrollToSentenceIndex(idxNum)
-                                            }
-                                          }
-                                        }
-                                        return (
-                                          <button
-                                            key={k}
-                                            onClick={onClick}
-                                            className="rounded border px-1.5 py-0.5 text-xs hover:bg-gray-50"
-                                            title={label}
-                                            type="button"
-                                          >
-                                            {label}
-                                          </button>
-                                        )
-                                      })}
-                                    </div>
+                                    <strong>{dict.screening.confidence}</strong>{' '}
+                                    {missingFields.includes('confidence')
+                                      ? 'Not returned by AI'
+                                      : Number.isFinite(displayConfidence)
+                                        ? String(displayConfidence)
+                                        : String(aiData.confidence ?? '')}
                                   </div>
-                                ) : null}
-
-                                {Array.isArray(aiData?.evidence_tables) && aiData.evidence_tables.length > 0 ? (
                                   <div className="mt-2">
-                                    <strong>Evidence tables:</strong>
-                                    <div className="mt-1 flex flex-wrap gap-1">
-                                      {aiData.evidence_tables.map((t: any, k: number) => {
-                                        const label = `Table T${String(t)}`
-                                        return (
-                                          <button
-                                            key={k}
-                                            type="button"
-                                            onClick={() => scrollToArtifact('table', Number(t))}
-                                            className="rounded border px-1.5 py-0.5 text-xs text-gray-700 bg-gray-50 hover:bg-gray-100"
-                                            title={label}
-                                          >
-                                            {label}
-                                          </button>
-                                        )
-                                      })}
+                                    <strong>
+                                      {dict.screening.explanation}
+                                    </strong>
+                                    <div className="mt-1 text-sm text-gray-700">
+                                      {missingFields.includes('rationale')
+                                        ? 'Rationale was not returned by the AI after repair.'
+                                        : displayExplanation ||
+                                          dict.screening.noExplanation}
                                     </div>
                                   </div>
-                                ) : null}
 
-                                {Array.isArray(aiData?.evidence_figures) && aiData.evidence_figures.length > 0 ? (
-                                  <div className="mt-2">
-                                    <strong>Evidence figures:</strong>
-                                    <div className="mt-1 flex flex-wrap gap-1">
-                                      {aiData.evidence_figures.map((f: any, k: number) => {
-                                        const label = `Figure F${String(f)}`
-                                        return (
-                                          <button
-                                            key={k}
-                                            type="button"
-                                            onClick={() => scrollToArtifact('figure', Number(f))}
-                                            className="rounded border px-1.5 py-0.5 text-xs text-gray-700 bg-gray-50 hover:bg-gray-100"
-                                            title={label}
-                                          >
-                                            {label}
-                                          </button>
-                                        )
-                                      })}
+                                  {hasAgentic && crit ? (
+                                    <div className="mt-3 rounded-md border border-gray-100 bg-gray-50 p-2 text-xs text-gray-700">
+                                      <div className="mt-1 font-semibold text-gray-800">
+                                        {critDisagrees ? (
+                                          <span className="text-amber-700">
+                                            Critical review recommends a
+                                            different answer:{' '}
+                                            {String(
+                                              (crit as any)?.answer ?? '—',
+                                            )}
+                                          </span>
+                                        ) : (
+                                          <span className="text-emerald-700">
+                                            Critical review supports the
+                                            screening answer
+                                          </span>
+                                        )}
+                                      </div>
+                                      <div>
+                                        Judgment confidence:{' '}
+                                        {String(
+                                          (crit as any)?.confidence ?? '—',
+                                        )}
+                                      </div>
                                     </div>
-                                  </div>
-                                ) : null}
-                              </div>
-                            ) : null}
-                          </div>
-                        ) : null}
-                      </div>
-                    )
-                  })}
+                                  ) : null}
+                                  {Array.isArray(aiData?.evidence_sentences) &&
+                                  aiData.evidence_sentences.length > 0 ? (
+                                    <div className="mt-2">
+                                      <strong>{dict.screening.evidence}</strong>
+                                      <div className="mt-1 flex flex-wrap gap-1">
+                                        {aiData.evidence_sentences.map(
+                                          (item: any, k: number) => {
+                                            const isCoord =
+                                              item && typeof item === 'object'
+                                            const label = isCoord
+                                              ? `${dict.screening.page} ${String(item.page ?? item.page_number ?? item.pageNum ?? '?')}${item.text ? `: ${String(item.text).slice(0, 80)}` : ''}`
+                                              : `${dict.screening.sentence} ${String(item)}`
+                                            const onClick = () => {
+                                              if (!viewerRef.current) return
+                                              if (isCoord) {
+                                                viewerRef.current.scrollToCoord(
+                                                  item,
+                                                )
+                                              } else {
+                                                const idxNum = Number(item)
+                                                if (!Number.isNaN(idxNum)) {
+                                                  viewerRef.current.scrollToSentenceIndex(
+                                                    idxNum,
+                                                  )
+                                                }
+                                              }
+                                            }
+                                            return (
+                                              <button
+                                                key={k}
+                                                onClick={onClick}
+                                                className="rounded border px-1.5 py-0.5 text-xs hover:bg-gray-50"
+                                                title={label}
+                                                type="button"
+                                              >
+                                                {label}
+                                              </button>
+                                            )
+                                          },
+                                        )}
+                                      </div>
+                                    </div>
+                                  ) : null}
+
+                                  {Array.isArray(aiData?.evidence_tables) &&
+                                  aiData.evidence_tables.length > 0 ? (
+                                    <div className="mt-2">
+                                      <strong>Evidence tables:</strong>
+                                      <div className="mt-1 flex flex-wrap gap-1">
+                                        {aiData.evidence_tables.map(
+                                          (t: any, k: number) => {
+                                            const label = `Table T${String(t)}`
+                                            return (
+                                              <button
+                                                key={k}
+                                                type="button"
+                                                onClick={() =>
+                                                  scrollToArtifact(
+                                                    'table',
+                                                    Number(t),
+                                                  )
+                                                }
+                                                className="rounded border bg-gray-50 px-1.5 py-0.5 text-xs text-gray-700 hover:bg-gray-100"
+                                                title={label}
+                                              >
+                                                {label}
+                                              </button>
+                                            )
+                                          },
+                                        )}
+                                      </div>
+                                    </div>
+                                  ) : null}
+
+                                  {Array.isArray(aiData?.evidence_figures) &&
+                                  aiData.evidence_figures.length > 0 ? (
+                                    <div className="mt-2">
+                                      <strong>Evidence figures:</strong>
+                                      <div className="mt-1 flex flex-wrap gap-1">
+                                        {aiData.evidence_figures.map(
+                                          (f: any, k: number) => {
+                                            const label = `Figure F${String(f)}`
+                                            return (
+                                              <button
+                                                key={k}
+                                                type="button"
+                                                onClick={() =>
+                                                  scrollToArtifact(
+                                                    'figure',
+                                                    Number(f),
+                                                  )
+                                                }
+                                                className="rounded border bg-gray-50 px-1.5 py-0.5 text-xs text-gray-700 hover:bg-gray-100"
+                                                title={label}
+                                              >
+                                                {label}
+                                              </button>
+                                            )
+                                          },
+                                        )}
+                                      </div>
+                                    </div>
+                                  ) : null}
+                                </div>
+                              ) : null}
+                            </div>
+                          ) : null}
+                        </div>
+                      )
+                    })}
+                  </div>
                 </div>
-              </div>
               )}
               {/* Validation checkbox — at the bottom, matching L1 screen layout */}
               <div className="mt-2 rounded-md border border-gray-100 bg-gray-50 p-3">
@@ -1340,24 +1743,33 @@ export default function CanSrL2ScreenViewPage() {
                     }}
                   />
                   <span>
-                    Validated by <span className="font-medium">{String(userEmail || '—')}</span>
+                    Validated by{' '}
+                    <span className="font-medium">
+                      {String(userEmail || '—')}
+                    </span>
                   </span>
                 </label>
 
                 {l2ValidationsSorted.length ? (
                   <div className="mt-2 space-y-1">
                     {l2ValidationsSorted.map((v, idx) => (
-                      <div key={`${v.user}-${idx}`} className="text-xs text-gray-600">
-                        Validated on {formatValidationDate(v.validated_at)} by {v.user}
+                      <div
+                        key={`${v.user}-${idx}`}
+                        className="text-xs text-gray-600"
+                      >
+                        Validated on {formatValidationDate(v.validated_at)} by{' '}
+                        {v.user}
                       </div>
                     ))}
                   </div>
                 ) : (
-                  <div className="mt-2 text-xs text-gray-600">Not validated</div>
+                  <div className="mt-2 text-xs text-gray-600">
+                    Not validated
+                  </div>
                 )}
               </div>
 
-              <div className="flex items-center justify-between mt-4">
+              <div className="mt-4 flex items-center justify-between">
                 <button
                   disabled={runningAll}
                   onClick={() => {
@@ -1386,7 +1798,9 @@ export default function CanSrL2ScreenViewPage() {
                     const idx = citationIdList.indexOf(cur)
                     if (idx === -1) return
                     if (idx >= citationIdList.length - 1) {
-                      router.push(`/${lang}/can-sr/l2-screen?sr_id=${encodeURIComponent(srId)}&filter=${encodeURIComponent(filterMode)}`)
+                      router.push(
+                        `/${lang}/can-sr/l2-screen?sr_id=${encodeURIComponent(srId)}&filter=${encodeURIComponent(filterMode)}`,
+                      )
                       return
                     }
                     const target = String(citationIdList[idx + 1])
diff --git a/frontend/app/[lang]/can-sr/setup/page.tsx b/frontend/app/[lang]/can-sr/setup/page.tsx
index 2c9875b5..7b9a16c7 100644
--- a/frontend/app/[lang]/can-sr/setup/page.tsx
+++ b/frontend/app/[lang]/can-sr/setup/page.tsx
@@ -27,6 +27,7 @@ export default function CanSrSetupPage() {
   const [uploading, setUploading] = useState(false)
   const [uploadResult, setUploadResult] = useState<any>(null)
   const [uploadError, setUploadError] = useState<string | null>(null)
+  const [criteriaReloadKey, setCriteriaReloadKey] = useState(0)
 
   const [, setSrLoading] = useState<boolean>(true)
   const [, setSrError] = useState<string | null>(null)
@@ -107,6 +108,7 @@ export default function CanSrSetupPage() {
         setUploadError(errMsg)
       } else {
         setUploadResult(data)
+        setCriteriaReloadKey((value) => value + 1)
         void loadSr()
       }
     } catch (err: any) {
@@ -195,6 +197,7 @@ export default function CanSrSetupPage() {
           <CriteriaEditor
             srId={srId || ''}
             hasScreeningData={hasExistingScreeningData}
+            reloadKey={criteriaReloadKey}
             labels={dict.setup.criteriaBuilder as unknown as Record<string, string>}
           />
 
diff --git a/frontend/components/can-sr/CitationListPage.tsx b/frontend/components/can-sr/CitationListPage.tsx
index 1eaa2ead..b879a01f 100644
--- a/frontend/components/can-sr/CitationListPage.tsx
+++ b/frontend/components/can-sr/CitationListPage.tsx
@@ -215,7 +215,8 @@ export default function CitationsListPage({
   const [savingThresholds, setSavingThresholds] = useState<boolean>(false)
 
   // Run-all job tracking (persist across modal close / refresh)
-  const [runAllForce, setRunAllForce] = useState<boolean>(false)
+  const [runAllSkipExistingAI, setRunAllSkipExistingAI] = useState<boolean>(false)
+  const [runAllSkipExistingHuman, setRunAllSkipExistingHuman] = useState<boolean>(false)
   const [runAllJobId, setRunAllJobId] = useState<string | null>(null)
   const [runAllJob, setRunAllJob] = useState<any | null>(null)
   const [runAllStarting, setRunAllStarting] = useState<boolean>(false)
@@ -264,15 +265,18 @@ export default function CitationsListPage({
       setPdfLinkageActive(active)
     }
     const changed = () => void refreshJobs()
+    const screeningChanged = () => setMetricsRefreshKey((value) => value + 1)
     void refreshJobs()
     window.addEventListener('jobs:changed', changed)
     window.addEventListener('run-all:changed', changed)
+    window.addEventListener('screening:changed', screeningChanged)
     const timer = window.setInterval(refreshJobs, 5000)
     return () => {
       alive = false
       window.clearInterval(timer)
       window.removeEventListener('jobs:changed', changed)
       window.removeEventListener('run-all:changed', changed)
+      window.removeEventListener('screening:changed', screeningChanged)
     }
   }, [srId])
 
@@ -518,6 +522,7 @@ export default function CitationsListPage({
   useEffect(() => {
     if (!runAllJobId) return
     let alive = true
+    let processedCount = 0
     const headers = getAuthHeaders()
 
     const fetchStatus = async () => {
@@ -539,6 +544,14 @@ export default function CitationsListPage({
         const latest = await fetchStatus()
         if (!alive) return
         setRunAllJob(latest)
+        const nextProcessedCount =
+          Number(latest?.done || 0) +
+          Number(latest?.skipped || 0) +
+          Number(latest?.failed || 0)
+        if (nextProcessedCount > processedCount) {
+          processedCount = nextProcessedCount
+          setMetricsRefreshKey((value) => value + 1)
+        }
         const st = String(latest?.status || '').toLowerCase()
         if (['done', 'failed', 'canceled'].includes(st)) {
           clearRunAll()
@@ -586,7 +599,9 @@ export default function CitationsListPage({
           body: JSON.stringify({
             step: (screeningStep as any) || 'l1',
             model: selectedModel,
-            force: runAllForce,
+            force: false,
+            skip_existing_ai: runAllSkipExistingAI,
+            skip_existing_human: runAllSkipExistingHuman,
             chunk_size: 5,
             // For l1 we explicitly send the filtered list IDs (entire list, not just page).
             // For l2/extract we let the backend compute eligible IDs so it can enforce
@@ -756,11 +771,19 @@ export default function CitationsListPage({
             <label className="flex items-center gap-2 text-sm text-gray-700">
               <input
                 type="checkbox"
-                checked={runAllForce}
-                onChange={(e) => setRunAllForce(e.target.checked)}
+                checked={runAllSkipExistingAI}
+                onChange={(e) => setRunAllSkipExistingAI(e.target.checked)}
+              />
+              {dict?.screening?.skipExistingAI || 'Skip existing AI answers'}
+            </label>
+
+            <label className="flex items-center gap-2 text-sm text-gray-700">
+              <input
+                type="checkbox"
+                checked={runAllSkipExistingHuman}
+                onChange={(e) => setRunAllSkipExistingHuman(e.target.checked)}
               />
-              {dict?.screening?.forceRerun ||
-                'Force re-run (overwrite existing AI results)'}
+              {dict?.screening?.skipExistingHuman || 'Skip existing human answers'}
             </label>
 
             <DialogFooter className="gap-2">
diff --git a/frontend/components/can-sr/PDFBoundingBoxViewer.tsx b/frontend/components/can-sr/PDFBoundingBoxViewer.tsx
index f5fabd4f..88ed7ebd 100644
--- a/frontend/components/can-sr/PDFBoundingBoxViewer.tsx
+++ b/frontend/components/can-sr/PDFBoundingBoxViewer.tsx
@@ -80,8 +80,10 @@ function extractSentenceArray(fulltext?: string): string[] {
       out[idx] = m[2]
     }
   }
-  // fill holes if any
-  return out.filter((v) => typeof v === 'string')
+  // Preserve the model-facing indices. Removing sparse entries here would shift
+  // every later evidence index and make a valid sentence chip highlight the
+  // wrong sentence.
+  return out
 }
 
 export type PDFBoundingBoxViewerHandle = {
diff --git a/frontend/components/can-sr/PagedList.tsx b/frontend/components/can-sr/PagedList.tsx
index 4377abca..1d8e36a8 100644
--- a/frontend/components/can-sr/PagedList.tsx
+++ b/frontend/components/can-sr/PagedList.tsx
@@ -473,6 +473,7 @@ export default function PagedList({
         }),
       })
       setLlmClassified((prev: Record<number, boolean>) => ({ ...prev, [id]: true }))
+      window.dispatchEvent(new Event('screening:changed'))
       return
     }
 
@@ -489,6 +490,7 @@ export default function PagedList({
         }),
       })
       setLlmClassified((prev: Record<number, boolean>) => ({ ...prev, [id]: true }))
+      window.dispatchEvent(new Event('screening:changed'))
       return
     }
 
diff --git a/frontend/components/can-sr/ScreeningMetricsPanel.tsx b/frontend/components/can-sr/ScreeningMetricsPanel.tsx
index 91c16a99..6886ffc8 100644
--- a/frontend/components/can-sr/ScreeningMetricsPanel.tsx
+++ b/frontend/components/can-sr/ScreeningMetricsPanel.tsx
@@ -38,7 +38,12 @@ export type ScreeningCriterionMetrics = {
   recall?: number | null
   npv?: number | null
   confusion_matrix?: { tp: number; fp: number; fn: number; tn: number } | null
-  queue_confusion_matrix?: { tp: number; fp: number; fn: number; tn: number } | null
+  queue_confusion_matrix?: {
+    tp: number
+    fp: number
+    fn: number
+    tn: number
+  } | null
   human_total_count_all?: number | null
 }
 
@@ -131,7 +136,9 @@ export type ScreeningMetricsPanelProps = {
   showFilter?: boolean
 
   filterMode: 'needs' | 'validated' | 'unvalidated' | 'not_screened' | 'all'
-  onFilterModeChange: (v: 'needs' | 'validated' | 'unvalidated' | 'not_screened' | 'all') => void
+  onFilterModeChange: (
+    v: 'needs' | 'validated' | 'unvalidated' | 'not_screened' | 'all',
+  ) => void
   stats?: ScreeningMetricsStats
 }
 
@@ -159,7 +166,29 @@ export default function ScreeningMetricsPanel({
   onFilterModeChange,
   stats: _stats,
 }: ScreeningMetricsPanelProps) {
-  const [thresholdText, setThresholdText] = React.useState<Record<string, string>>({})
+  const [sliderPercents, setSliderPercents] = React.useState<
+    Record<string, number>
+  >({})
+  const pendingSliderPercents = React.useRef<Record<string, number>>({})
+
+  React.useEffect(() => {
+    if (!criterionMetrics?.length) return
+    setSliderPercents((previous) => {
+      const next = { ...previous }
+      for (const criterion of criterionMetrics) {
+        const incoming = Math.round(
+          Math.max(0, Math.min(1, criterion.threshold)) * 100,
+        )
+        const pending = pendingSliderPercents.current[criterion.criterion_key]
+        if (pending === undefined || incoming === pending) {
+          next[criterion.criterion_key] = incoming
+          if (incoming === pending)
+            delete pendingSliderPercents.current[criterion.criterion_key]
+        }
+      }
+      return next
+    })
+  }, [criterionMetrics])
 
   // Kept for backwards-compatibility with callers that still compute page-local stats.
   void _stats
@@ -170,20 +199,6 @@ export default function ScreeningMetricsPanel({
     return m
   }, [calibration])
 
-  // Keep a stable text representation so users can type freely.
-  React.useEffect(() => {
-    if (!criterionMetrics?.length) return
-    setThresholdText((prev) => {
-      const next = { ...prev }
-      for (const c of criterionMetrics) {
-        const k = c.criterion_key
-        if (!(k in next)) {
-          next[k] = Number.isFinite(c.threshold) ? String(c.threshold) : '0.9'
-        }
-      }
-      return next
-    })
-  }, [criterionMetrics])
   const total = summary?.total_citations ?? 0
   const validatedAll = summary?.validated_all ?? 0
   const notScreened = summary?.not_screened_yet ?? 0
@@ -195,7 +210,10 @@ export default function ScreeningMetricsPanel({
 
   // Everything that is NOT validated, NOT in the remaining queue, and NOT not-screened.
   // (Typically: confident excludes + confident includes at current thresholds.)
-  const resolvedRemaining = Math.max(0, total - validatedAll - queueRemaining - notScreened)
+  const resolvedRemaining = Math.max(
+    0,
+    total - validatedAll - queueRemaining - notScreened,
+  )
 
   const validatedPct = total > 0 ? (validatedAll / total) * 100 : 0
   const queueRemainingPct = total > 0 ? (queueRemaining / total) * 100 : 0
@@ -219,42 +237,81 @@ export default function ScreeningMetricsPanel({
         </div>
 
         {/* Headline metrics */}
-        {criterionMetrics?.length ? (() => {
-          const accVals = criterionMetrics.filter((c) => typeof c.accuracy_all === 'number').map((c) => c.accuracy_all as number)
-          const avgAcc = accVals.length ? Math.round((accVals.reduce((a, b) => a + b, 0) / accVals.length) * 100) : null
-          const f1Vals = criterionMetrics.filter((c) => typeof c.f1_score === 'number').map((c) => c.f1_score as number)
-          const avgF1 = f1Vals.length ? Math.round((f1Vals.reduce((a, b) => a + b, 0) / f1Vals.length) * 100) : null
-          const precVals = criterionMetrics.filter((c) => typeof c.precision === 'number').map((c) => c.precision as number)
-          const avgPrec = precVals.length ? Math.round((precVals.reduce((a, b) => a + b, 0) / precVals.length) * 100) : null
-          const recVals = criterionMetrics.filter((c) => typeof c.recall === 'number').map((c) => c.recall as number)
-          const avgRec = recVals.length ? Math.round((recVals.reduce((a, b) => a + b, 0) / recVals.length) * 100) : null
-          const screened = total - notScreened
-          const wr = screened > 0 ? Math.round((1 - queueTotal / screened) * 100) : null
-          return (
-            <div className="mt-2 flex flex-wrap items-center gap-2">
-              <div className="flex items-center gap-1.5 rounded-md bg-emerald-50 px-2 py-1">
-                <span className="text-[10px] text-emerald-600">Acc</span>
-                <span className="text-xs font-semibold text-emerald-900">{avgAcc === null ? '—' : `${avgAcc}%`}</span>
-              </div>
-              <div className="flex items-center gap-1.5 rounded-md bg-indigo-50 px-2 py-1">
-                <span className="text-[10px] text-indigo-600">F1</span>
-                <span className="text-xs font-semibold text-indigo-900">{avgF1 === null ? '—' : `${avgF1}%`}</span>
-              </div>
-              <div className="flex items-center gap-1.5 rounded-md bg-sky-50 px-2 py-1">
-                <span className="text-[10px] text-sky-600">Prec</span>
-                <span className="text-xs font-semibold text-sky-900">{avgPrec === null ? '—' : `${avgPrec}%`}</span>
-              </div>
-              <div className="flex items-center gap-1.5 rounded-md bg-violet-50 px-2 py-1">
-                <span className="text-[10px] text-violet-600">Recall</span>
-                <span className="text-xs font-semibold text-violet-900">{avgRec === null ? '—' : `${avgRec}%`}</span>
-              </div>
-              <div className="flex items-center gap-1.5 rounded-md bg-amber-50 px-2 py-1">
-                <span className="text-[10px] text-amber-600">Workload ↓</span>
-                <span className="text-xs font-semibold text-amber-900">{wr === null ? '—' : `${Math.max(0, wr)}%`}</span>
+        {criterionMetrics?.length ? (
+          (() => {
+            const accVals = criterionMetrics
+              .filter((c) => typeof c.accuracy_all === 'number')
+              .map((c) => c.accuracy_all as number)
+            const avgAcc = accVals.length
+              ? Math.round(
+                  (accVals.reduce((a, b) => a + b, 0) / accVals.length) * 100,
+                )
+              : null
+            const f1Vals = criterionMetrics
+              .filter((c) => typeof c.f1_score === 'number')
+              .map((c) => c.f1_score as number)
+            const avgF1 = f1Vals.length
+              ? Math.round(
+                  (f1Vals.reduce((a, b) => a + b, 0) / f1Vals.length) * 100,
+                )
+              : null
+            const precVals = criterionMetrics
+              .filter((c) => typeof c.precision === 'number')
+              .map((c) => c.precision as number)
+            const avgPrec = precVals.length
+              ? Math.round(
+                  (precVals.reduce((a, b) => a + b, 0) / precVals.length) * 100,
+                )
+              : null
+            const recVals = criterionMetrics
+              .filter((c) => typeof c.recall === 'number')
+              .map((c) => c.recall as number)
+            const avgRec = recVals.length
+              ? Math.round(
+                  (recVals.reduce((a, b) => a + b, 0) / recVals.length) * 100,
+                )
+              : null
+            const screened = total - notScreened
+            const wr =
+              screened > 0
+                ? Math.round((1 - queueTotal / screened) * 100)
+                : null
+            return (
+              <div className="mt-2 flex flex-wrap items-center gap-2">
+                <div className="flex items-center gap-1.5 rounded-md bg-emerald-50 px-2 py-1">
+                  <span className="text-[10px] text-emerald-600">Acc</span>
+                  <span className="text-xs font-semibold text-emerald-900">
+                    {avgAcc === null ? '—' : `${avgAcc}%`}
+                  </span>
+                </div>
+                <div className="flex items-center gap-1.5 rounded-md bg-indigo-50 px-2 py-1">
+                  <span className="text-[10px] text-indigo-600">F1</span>
+                  <span className="text-xs font-semibold text-indigo-900">
+                    {avgF1 === null ? '—' : `${avgF1}%`}
+                  </span>
+                </div>
+                <div className="flex items-center gap-1.5 rounded-md bg-sky-50 px-2 py-1">
+                  <span className="text-[10px] text-sky-600">Prec</span>
+                  <span className="text-xs font-semibold text-sky-900">
+                    {avgPrec === null ? '—' : `${avgPrec}%`}
+                  </span>
+                </div>
+                <div className="flex items-center gap-1.5 rounded-md bg-violet-50 px-2 py-1">
+                  <span className="text-[10px] text-violet-600">Recall</span>
+                  <span className="text-xs font-semibold text-violet-900">
+                    {avgRec === null ? '—' : `${avgRec}%`}
+                  </span>
+                </div>
+                <div className="flex items-center gap-1.5 rounded-md bg-amber-50 px-2 py-1">
+                  <span className="text-[10px] text-amber-600">Workload ↓</span>
+                  <span className="text-xs font-semibold text-amber-900">
+                    {wr === null ? '—' : `${Math.max(0, wr)}%`}
+                  </span>
+                </div>
               </div>
-            </div>
-          )
-        })() : (
+            )
+          })()
+        ) : (
           <p className="mt-1 text-xs text-gray-600">
             Threshold + validation workload controls.
           </p>
@@ -262,7 +319,9 @@ export default function ScreeningMetricsPanel({
       </div>
 
       <div className="space-y-3">
-        {typeof threshold === 'number' && onThresholdChange && !criterionMetrics?.length ? (
+        {typeof threshold === 'number' &&
+        onThresholdChange &&
+        !criterionMetrics?.length ? (
           <div className="flex items-center justify-between gap-3">
             <label className="text-sm text-gray-700">Threshold</label>
             <input
@@ -312,7 +371,9 @@ export default function ScreeningMetricsPanel({
                     const screened = total - notScreened
                     if (screened === 0 || total === 0) return '—'
                     // Workload reduction = % of screened citations that don't need human review
-                    const reduction = Math.round((1 - (queueTotal / screened)) * 100)
+                    const reduction = Math.round(
+                      (1 - queueTotal / screened) * 100,
+                    )
                     return `${Math.max(0, Math.min(100, reduction))}%`
                   })()}
                 </span>
@@ -324,8 +385,10 @@ export default function ScreeningMetricsPanel({
               <div className="relative h-3 w-full overflow-hidden rounded bg-gray-200">
                 {/* Validated (green) */}
                 <div
-                  className="absolute left-0 top-0 h-3 bg-emerald-600"
-                  style={{ width: `${Math.min(100, Math.max(0, validatedPct))}%` }}
+                  className="absolute top-0 left-0 h-3 bg-emerald-600"
+                  style={{
+                    width: `${Math.min(100, Math.max(0, validatedPct))}%`,
+                  }}
                 />
 
                 {/* Remaining human review queue (amber) */}
@@ -360,16 +423,23 @@ export default function ScreeningMetricsPanel({
 
             <div className="mt-2 grid grid-cols-1 gap-1 text-[11px] text-gray-600">
               <div>
-                <span className="font-medium text-gray-700">Validated:</span> {validatedAll} / {total}
+                <span className="font-medium text-gray-700">Validated:</span>{' '}
+                {validatedAll} / {total}
               </div>
               <div>
-                <span className="font-medium text-gray-700">Human review queue:</span> {queueRemaining} remaining (of {queueTotal})
+                <span className="font-medium text-gray-700">
+                  Human review queue:
+                </span>{' '}
+                {queueRemaining} remaining (of {queueTotal})
               </div>
               {/* <div>
                 <span className="font-medium text-gray-700">Resolved (no review needed):</span> {resolvedRemaining}
               </div> */}
               <div>
-                <span className="font-medium text-gray-700">Not screened yet:</span> {notScreened}
+                <span className="font-medium text-gray-700">
+                  Not screened yet:
+                </span>{' '}
+                {notScreened}
               </div>
             </div>
           </div>
@@ -400,126 +470,149 @@ export default function ScreeningMetricsPanel({
 
             <div className="mt-2 space-y-2">
               {criterionMetrics.map((c) => {
-                const textVal = thresholdText[c.criterion_key] ?? (Number.isFinite(c.threshold) ? String(c.threshold) : '0.9')
+                const thresholdPercent =
+                  sliderPercents[c.criterion_key] ??
+                  Math.round(Math.max(0, Math.min(1, c.threshold)) * 100)
                 const cal = calibByKey.get(c.criterion_key)
                 const rec =
                   cal && typeof cal.recommended_threshold === 'number'
-                    ? Math.round(cal.recommended_threshold * 100) / 100
+                    ? Math.max(0, Math.min(1, cal.recommended_threshold))
                     : null
-
-                const curve = Array.isArray(cal?.curve) ? cal!.curve : []
-                const recPoint =
-                  rec === null
-                    ? null
-                    : curve.find((p) => Math.abs(p.threshold - rec) < 1e-9) || null
-                const wr =
-                  typeof recPoint?.workload_reduction === 'number'
-                    ? Math.round(recPoint.workload_reduction * 100)
-                    : null
-                const recall =
-                  typeof recPoint?.recall === 'number' ? Math.round(recPoint.recall * 100) : null
-                const fpr = typeof recPoint?.fpr === 'number' ? Math.round(recPoint.fpr * 100) : null
+                const cm = c.confusion_matrix
+                const npvPct =
+                  typeof c.npv === 'number'
+                    ? Math.round(c.npv * 100)
+                    : cm && cm.fn === 0
+                      ? 100
+                      : null
+                const updateThreshold = (percent: number) => {
+                  const value = Math.max(0, Math.min(100, Math.round(percent)))
+                  pendingSliderPercents.current[c.criterion_key] = value
+                  setSliderPercents((previous) => ({
+                    ...previous,
+                    [c.criterion_key]: value,
+                  }))
+                  onCriterionThresholdChange?.(c.criterion_key, value / 100)
+                }
+                const commitThreshold = (percent: number) => {
+                  const value = Math.max(0, Math.min(100, Math.round(percent)))
+                  onCriterionThresholdCommit?.(c.criterion_key, value / 100)
+                }
 
                 return (
-                  <details key={c.criterion_key} className="rounded border border-gray-100 bg-white p-2">
-                    <summary className="cursor-pointer list-none">
-                      <div className="flex items-center justify-between gap-2">
-                        <div className="min-w-0 flex-1">
-                          <div className="truncate text-xs font-medium text-gray-800">{c.label}</div>
-                          <div className="mt-1 space-y-0.5 text-[11px] text-gray-500">
-                            {rec !== null ? <div>Recommended thr: {rec}</div> : null}
-                            {(() => {
-                              // NPV: default to 100% if confusion matrix exists but fn === 0
-                              const cm = c.confusion_matrix
-                              const rawNpv = typeof c.npv === 'number' ? c.npv : null
-                              const npvPct = rawNpv !== null ? Math.round(rawNpv * 100) : (cm && cm.fn === 0 ? 100 : null)
-                              // Corrected NPV delta
-                              const qcm = c.queue_confusion_matrix
-                              let npvDelta: number | null = null
-                              if (npvPct !== null && cm && qcm) {
-                                const corrTn = cm.tn + qcm.fp
-                                const corrFn = cm.fn - qcm.fn
-                                const corrNpv = (corrTn + corrFn) > 0 ? corrTn / (corrTn + corrFn) : 1
-                                npvDelta = Math.round(corrNpv * 100) - npvPct
-                              }
-                              return npvPct !== null ? (
-                                <div>NPV: {npvPct}%{npvDelta !== null && npvDelta > 0 ? <span className="ml-1 text-emerald-600 font-medium">+{npvDelta}%</span> : null}</div>
-                              ) : null
-                            })()}
-                            {(() => {
-                              // Workload ↓ general = 1 - (needs_human_review_count / has_run_count)
-                              if (c.has_run_count <= 0) return null
-                              const wrGen = Math.round((1 - c.needs_human_review_count / c.has_run_count) * 100)
-                              return <div>Workload ↓: {Math.max(0, wrGen)}%</div>
-                            })()}
+                  <div
+                    key={c.criterion_key}
+                    className="flex items-start gap-4 rounded border border-gray-100 bg-white p-2"
+                  >
+                    <div className="min-w-0 flex-1">
+                      <div className="truncate text-xs font-medium text-gray-800">
+                        {c.label}
+                      </div>
+                      <div className="mt-1 space-y-0.5 text-[11px] text-gray-500">
+                        {npvPct !== null ? <div>NPV: {npvPct}%</div> : null}
+                        {c.has_run_count > 0 ? (
+                          <div>
+                            Workload ↓:{' '}
+                            {Math.max(
+                              0,
+                              Math.round(
+                                (1 -
+                                  c.needs_human_review_count /
+                                    c.has_run_count) *
+                                  100,
+                              ),
+                            )}
+                            %
                           </div>
-                        </div>
+                        ) : null}
+                      </div>
+                    </div>
 
-                        <div className="flex items-center gap-2">
-                          <label className="text-[11px] text-gray-600">Thr</label>
-                          <input
-                            type="text"
-                            inputMode="decimal"
-                            value={textVal}
-                            onChange={(e: React.ChangeEvent<HTMLInputElement>) => {
-                              const raw = e.target.value
-                              setThresholdText((p) => ({ ...p, [c.criterion_key]: raw }))
-                              const v = Number(raw)
-                              if (!Number.isFinite(v)) return
-                              onCriterionThresholdChange?.(c.criterion_key, Math.max(0, Math.min(1, v)))
-                            }}
-                            onKeyDown={(e: React.KeyboardEvent<HTMLInputElement>) => {
-                              if (e.key !== 'Enter') return
-                              const v = Number((e.currentTarget as HTMLInputElement).value)
-                              if (!Number.isFinite(v)) return
-                              onCriterionThresholdCommit?.(c.criterion_key, Math.max(0, Math.min(1, v)))
+                    <div className="w-44 shrink-0">
+                      <div className="mb-1 flex items-center gap-1 text-[10px] text-gray-500">
+                        <span>Threshold</span>
+                        <span className="font-medium text-gray-700">
+                          {thresholdPercent}%
+                        </span>
+                      </div>
+                      <div
+                        className="relative cursor-pointer pt-7"
+                        role="slider"
+                        tabIndex={0}
+                        aria-label={`Threshold for ${c.label}`}
+                        aria-valuemin={0}
+                        aria-valuemax={100}
+                        aria-valuenow={thresholdPercent}
+                        onPointerDown={(e) => {
+                          e.currentTarget.setPointerCapture(e.pointerId)
+                          const bounds = e.currentTarget.getBoundingClientRect()
+                          updateThreshold(
+                            ((e.clientX - bounds.left) / bounds.width) * 100,
+                          )
+                        }}
+                        onPointerMove={(e) => {
+                          if (!e.currentTarget.hasPointerCapture(e.pointerId))
+                            return
+                          const bounds = e.currentTarget.getBoundingClientRect()
+                          updateThreshold(
+                            ((e.clientX - bounds.left) / bounds.width) * 100,
+                          )
+                        }}
+                        onPointerUp={(e) => {
+                          const bounds = e.currentTarget.getBoundingClientRect()
+                          commitThreshold(
+                            ((e.clientX - bounds.left) / bounds.width) * 100,
+                          )
+                          e.currentTarget.releasePointerCapture(e.pointerId)
+                        }}
+                        onKeyDown={(e) => {
+                          let next: number | null = null
+                          if (e.key === 'ArrowLeft' || e.key === 'ArrowDown')
+                            next = thresholdPercent - 1
+                          if (e.key === 'ArrowRight' || e.key === 'ArrowUp')
+                            next = thresholdPercent + 1
+                          if (e.key === 'Home') next = 0
+                          if (e.key === 'End') next = 100
+                          if (e.key === 'PageDown') next = thresholdPercent - 10
+                          if (e.key === 'PageUp') next = thresholdPercent + 10
+                          if (next === null) return
+                          e.preventDefault()
+                          updateThreshold(next)
+                          commitThreshold(next)
+                        }}
+                      >
+                        {rec !== null ? (
+                          <div
+                            className="pointer-events-none absolute top-0 bottom-0 z-10 w-px"
+                            style={{ left: `${rec * 100}%` }}
+                            aria-hidden="true"
+                          >
+                            <span className="absolute top-0 left-1/2 -translate-x-1/2 text-[9px] font-medium whitespace-nowrap text-amber-700">
+                              Rec {Math.round(rec * 100)}%
+                            </span>
+                            <span className="absolute top-3 bottom-0 left-0 w-px bg-amber-500" />
+                          </div>
+                        ) : null}
+                        <div className="relative h-1 w-full rounded-full bg-gray-300">
+                          <div
+                            className="absolute inset-y-0 left-0 rounded-full bg-blue-600"
+                            style={{ width: `${thresholdPercent}%` }}
+                          />
+                          <div
+                            className="absolute top-1/2 h-4 w-4 rounded-full bg-blue-600 shadow-sm"
+                            style={{
+                              left: `${thresholdPercent}%`,
+                              transform: 'translate(-50%, -50%)',
                             }}
-                            className="w-20 rounded-md border border-gray-200 px-2 py-1 text-xs"
                           />
-                          <span className="text-[11px] text-gray-400">▾</span>
                         </div>
                       </div>
-                    </summary>
-
-                    <div className="mt-2 space-y-2 text-[11px] text-gray-700">
-                      <div className="grid grid-cols-2 gap-2">
-                        <div className="rounded border border-gray-100 bg-gray-50 p-2">
-                          Low confidence: {c.low_confidence_count}
-                        </div>
-                        <div className="rounded border border-gray-100 bg-gray-50 p-2">
-                          Critical disagreements: {c.critical_disagreement_count}
-                        </div>
+                      <div className="mt-0.5 flex justify-between text-[9px] text-gray-400">
+                        <span>0%</span>
+                        <span>100%</span>
                       </div>
-
-                      {cal ? (
-                        <div className="rounded border border-gray-100 bg-gray-50 p-2">
-                          <div className="text-[11px] font-medium text-gray-700">Calibration (validated set)</div>
-                          <div className="mt-1 grid grid-cols-2 gap-2">
-                            <div className="rounded border border-gray-100 bg-white p-2">
-                              Recommended thr: {rec === null ? '—' : rec}
-                            </div>
-                            <div className="rounded border border-gray-100 bg-white p-2">
-                              Validated n: {cal.validated_n}
-                            </div>
-                            <div className="rounded border border-gray-100 bg-white p-2">
-                              Recall @ rec: {recall === null ? '—' : `${recall}%`}
-                            </div>
-                            <div className="rounded border border-gray-100 bg-white p-2">
-                              FPR @ rec: {fpr === null ? '—' : `${fpr}%`}
-                            </div>
-                            <div className="col-span-2 rounded border border-gray-100 bg-white p-2">
-                              Workload ↓ @ rec: {wr === null ? '—' : `${wr}%`}
-                            </div>
-                          </div>
-                          {cal.recommended_reason ? (
-                            <div className="mt-2 text-[11px] text-gray-500">
-                              {cal.recommended_reason}
-                            </div>
-                          ) : null}
-                        </div>
-                      ) : null}
                     </div>
-                  </details>
+                  </div>
                 )
               })}
             </div>
diff --git a/frontend/components/can-sr/screening-citation-context.test.tsx b/frontend/components/can-sr/screening-citation-context.test.tsx
new file mode 100644
index 00000000..b4dc0389
--- /dev/null
+++ b/frontend/components/can-sr/screening-citation-context.test.tsx
@@ -0,0 +1,14 @@
+import { describe, expect, it } from 'vitest'
+import { humanAnswerStatus, resolveConfiguredValue } from './screening-citation-context'
+
+describe('screening citation context helpers', () => {
+  it('returns undefined when the citation row is unavailable', () => {
+    expect(resolveConfiguredValue(null, 'Title')).toBeUndefined()
+    expect(resolveConfiguredValue(undefined, 'Abstract')).toBeUndefined()
+  })
+
+  it('reports a missing configured answer when the citation row is unavailable', () => {
+    expect(humanAnswerStatus(null, 'Screening answer', ['Include', 'Exclude'])).toBe('missing')
+    expect(humanAnswerStatus(undefined, 'Screening answer', ['Include', 'Exclude'])).toBe('missing')
+  })
+})
diff --git a/frontend/components/can-sr/screening-citation-context.tsx b/frontend/components/can-sr/screening-citation-context.tsx
new file mode 100644
index 00000000..16668343
--- /dev/null
+++ b/frontend/components/can-sr/screening-citation-context.tsx
@@ -0,0 +1,126 @@
+'use client'
+
+import React from 'react'
+
+export type CitationFields = {
+  title?: string | null
+  abstract?: string | null
+  doi?: string | null
+  l1_include?: string[]
+}
+
+export function resolveConfiguredValue(
+  row: Record<string, any> | null | undefined,
+  header?: string | null,
+) {
+  if (!row || !header) return undefined
+  if (Object.prototype.hasOwnProperty.call(row, header)) return row[header]
+  const normalize = (value: string) =>
+    value
+      .trim()
+      .toLowerCase()
+      .replace(/[^a-z0-9]+/g, '_')
+      .replace(/^_+|_+$/g, '')
+  const key = normalize(header)
+  const matches = Object.keys(row).filter(
+    (candidate) => normalize(candidate) === key,
+  )
+  return matches.length === 1 ? row[matches[0]] : undefined
+}
+
+export function extractHumanAnswer(value: any): string {
+  if (value === undefined || value === null) return ''
+  if (typeof value === 'object')
+    return String(value.selected ?? value.value ?? '')
+  if (typeof value === 'string') {
+    try {
+      const parsed = JSON.parse(value)
+      return typeof parsed === 'object'
+        ? String(parsed?.selected ?? parsed?.value ?? '')
+        : value
+    } catch {
+      return value.trim()
+    }
+  }
+  return String(value)
+}
+
+export function humanAnswerStatus(
+  row: Record<string, any> | null | undefined,
+  answerColumn: string | null | undefined,
+  options: string[],
+): 'unconfigured' | 'missing' | 'blank' | 'matched' | 'unmatched' {
+  if (!answerColumn) return 'unconfigured'
+  const raw = resolveConfiguredValue(row, answerColumn)
+  if (raw === undefined) return 'missing'
+  const answer = extractHumanAnswer(raw).trim()
+  if (!answer) return 'blank'
+  if (!options.length) return 'matched'
+  const normalized = answer.toLowerCase().replace(/\s+/g, ' ')
+  return options.some(
+    (option) => option.trim().toLowerCase().replace(/\s+/g, ' ') === normalized,
+  )
+    ? 'matched'
+    : 'unmatched'
+}
+
+export function ScreeningCitationContext({
+  citation,
+  fields,
+  showTitleAbstract = true,
+}: {
+  citation: Record<string, any>
+  fields?: CitationFields | null
+  showTitleAbstract?: boolean
+}) {
+  const title =
+    resolveConfiguredValue(citation, fields?.title) ?? citation.title
+  const abstract =
+    resolveConfiguredValue(citation, fields?.abstract) ?? citation.abstract
+  const other = (fields?.l1_include || [])
+    .filter((header) => header !== fields?.title && header !== fields?.abstract)
+    .map((header) => ({
+      header,
+      value: resolveConfiguredValue(citation, header),
+    }))
+    .filter(
+      ({ value }) =>
+        value !== undefined && value !== null && String(value).trim() !== '',
+    )
+
+  return (
+    <div className="space-y-3">
+      {showTitleAbstract ? (
+        <>
+          <div>
+            <p className="text-xs text-gray-600">Citation #{citation.id}</p>
+            <h2 className="text-lg font-semibold text-gray-900">
+              {String(title || '')}
+            </h2>
+          </div>
+          <div className="rounded-md border border-gray-200 bg-white p-4">
+            <h3 className="text-sm font-medium text-gray-800">Abstract</h3>
+            <p className="mt-2 text-sm whitespace-pre-wrap text-gray-800">
+              {String(abstract || 'No abstract available')}
+            </p>
+          </div>
+        </>
+      ) : null}
+      {other.length > 0 && (
+        <div className="rounded-md border border-gray-200 bg-gray-50 p-4">
+          <h3 className="text-sm font-medium text-gray-800">
+            Other citation fields
+          </h3>
+          <dl className="mt-2 space-y-1 text-sm text-gray-700">
+            {other.map(({ header, value }) => (
+              <div key={header}>
+                <dt className="inline font-medium">{header}: </dt>
+                <dd className="inline whitespace-pre-wrap">{String(value)}</dd>
+              </div>
+            ))}
+          </dl>
+        </div>
+      )}
+    </div>
+  )
+}
diff --git a/frontend/components/can-sr/setup/citation-field-selector.tsx b/frontend/components/can-sr/setup/citation-field-selector.tsx
index 2e7e0dc9..114017aa 100644
--- a/frontend/components/can-sr/setup/citation-field-selector.tsx
+++ b/frontend/components/can-sr/setup/citation-field-selector.tsx
@@ -1,5 +1,5 @@
 import { ArrowDown, ArrowUp, ChevronDown, ChevronUp, Plus, X } from 'lucide-react'
-import { useEffect, useRef, useState, type Dispatch } from 'react'
+import { useState, type Dispatch } from 'react'
 import type { CriteriaDraftAction, CriteriaDraftState } from './criteria-types'
 
 export type CitationFieldContract = {
@@ -7,80 +7,69 @@ export type CitationFieldContract = {
   unavailable_configured_fields: string[]
 }
 
+type Row = { kind: 'title' | 'abstract' | 'additional'; column: string | null; index?: number }
+
 export default function CitationFieldSelector({ state, dispatch, contract, labels }: {
   state: CriteriaDraftState
   dispatch: Dispatch<CriteriaDraftAction>
   contract: CitationFieldContract
   labels: Record<string, string>
 }) {
-  const selected = state.criteria.citation_fields.l1_include
-  const available = new Set(contract.fields.map((field) => field.name))
-  const addable = contract.fields.filter((field) => !selected.includes(field.name))
-  const [addMenuOpen, setAddMenuOpen] = useState(false)
   const [collapsed, setCollapsed] = useState(false)
-  const addMenuRef = useRef<HTMLDivElement>(null)
-  const setSelected = (value: string[]) => dispatch({ type: 'set-citation-fields', value })
+  const available = new Set(contract.fields.map((field) => field.name))
+  const title = state.criteria.citation_fields.title || null
+  const abstract = state.criteria.citation_fields.abstract || null
+  const additional = state.criteria.citation_fields.l1_include
+  const rows: Row[] = [
+    { kind: 'title', column: title },
+    { kind: 'abstract', column: abstract },
+    ...additional.map((column, index) => ({ kind: 'additional' as const, column, index })),
+  ]
+  const selected = new Set([title, abstract, ...additional].filter(Boolean) as string[])
+  const options = (current: string | null) => [
+    ...(current && !available.has(current) ? [{ name: current, unavailable: true }] : []),
+    ...contract.fields.filter((field) => !selected.has(field.name) || field.name === current).map((field) => ({ name: field.name, unavailable: false })),
+  ]
+  const setColumn = (row: Row, value: string | null) => {
+    if (row.kind === 'title') dispatch({ type: 'set-title-field', value })
+    else if (row.kind === 'abstract') dispatch({ type: 'set-abstract-field', value })
+    else if (row.index !== undefined) dispatch({ type: 'set-citation-fields', value: additional.map((item, index) => index === row.index ? value || '' : item).filter(Boolean) })
+  }
   const move = (index: number, offset: number) => {
-    const next = [...selected]
-    ;[next[index], next[index + offset]] = [next[index + offset], next[index]]
-    setSelected(next)
+    const target = index + offset
+    if (target < 0 || target >= additional.length) return
+    const next = [...additional]
+    ;[next[index], next[target]] = [next[target], next[index]]
+    dispatch({ type: 'set-citation-fields', value: next })
   }
-  useEffect(() => {
-    if (!addMenuOpen) return
-    const closeOnOutsideClick = (event: MouseEvent) => {
-      if (!addMenuRef.current?.contains(event.target as Node)) setAddMenuOpen(false)
-    }
-    const closeOnEscape = (event: KeyboardEvent) => {
-      if (event.key === 'Escape') setAddMenuOpen(false)
-    }
-    document.addEventListener('mousedown', closeOnOutsideClick)
-    document.addEventListener('keydown', closeOnEscape)
-    return () => {
-      document.removeEventListener('mousedown', closeOnOutsideClick)
-      document.removeEventListener('keydown', closeOnEscape)
-    }
-  }, [addMenuOpen])
+  const labelFor = (kind: Row['kind']) => kind === 'title' ? labels.titleField : kind === 'abstract' ? labels.abstractField : labels.additionalField
+  const statusFor = (row: Row) => !row.column ? labels.requiredField : available.has(row.column) ? labels.available : labels.unavailableField
+
   return <section id="criteria-l1-fields" tabIndex={-1} className="space-y-3" aria-labelledby="citation-fields-heading">
     <div className="flex items-center gap-2 rounded-md bg-gray-50 p-2">
-      <button type="button" aria-expanded={!collapsed} aria-controls="citation-fields-body" aria-label={`${collapsed ? labels.expand : labels.collapse} ${labels.citationFields}`} title={collapsed ? labels.expand : labels.collapse} onClick={() => { setCollapsed((value) => !value); setAddMenuOpen(false) }} className="rounded p-1 text-gray-600 hover:bg-gray-200">
+      <button type="button" aria-expanded={!collapsed} aria-controls="citation-fields-body" aria-label={`${collapsed ? labels.expand : labels.collapse} ${labels.citationFields}`} title={collapsed ? labels.expand : labels.collapse} onClick={() => setCollapsed((value) => !value)} className="rounded p-1 text-gray-600 hover:bg-gray-200">
         {collapsed ? <ChevronUp className="h-5 w-5" /> : <ChevronDown className="h-5 w-5" />}
       </button>
-      <h4 id="citation-fields-heading" className="font-semibold">{labels.citationFields} <span className="font-normal text-gray-500">({selected.length})</span></h4>
-      <div ref={addMenuRef} className="relative ml-auto">
-        <button type="button" aria-label={labels.addField} aria-haspopup="menu" aria-expanded={addMenuOpen} aria-controls="title-abstract-field-menu" disabled={addable.length === 0} onClick={() => setAddMenuOpen((open) => !open)} className="inline-flex items-center gap-2 rounded-md bg-emerald-600 px-3 py-2 text-sm font-medium text-white hover:bg-emerald-700 disabled:cursor-not-allowed disabled:opacity-40">
-          <Plus className="h-4 w-4" />{labels.addField}
-        </button>
-        {addMenuOpen ? <div id="title-abstract-field-menu" role="menu" className="absolute right-0 z-10 mt-2 min-w-48 overflow-hidden rounded-md border bg-white py-1 shadow-lg">
-          {addable.map((field) => <button key={field.name} type="button" role="menuitem" onClick={() => { setSelected([...selected, field.name]); setAddMenuOpen(false) }} className="block w-full px-3 py-2 text-left text-sm hover:bg-gray-100 focus:bg-gray-100">{field.name}</button>)}
-        </div> : null}
-      </div>
+      <h4 id="citation-fields-heading" className="font-semibold">{labels.citationFields}</h4>
+      {!collapsed && <button type="button" aria-label={labels.addAdditionalField} disabled={options(null).length === 0} onClick={() => dispatch({ type: 'set-citation-fields', value: [...additional, options(null)[0]?.name || ''] })} className="ml-auto inline-flex items-center gap-2 rounded-md bg-emerald-600 px-3 py-2 text-sm font-medium text-white hover:bg-emerald-700 disabled:cursor-not-allowed disabled:opacity-40"><Plus className="h-4 w-4" />{labels.addAdditionalField}</button>}
     </div>
     {!collapsed ? <div id="citation-fields-body">
-    <p className="text-sm text-gray-600">{labels.titleAbstractFieldsDescription}</p>
-    <div className="mt-4 overflow-x-auto rounded-md border">
-      <table className="w-full text-left text-sm">
-        <thead className="bg-gray-50">
-          <tr><th scope="col" className="px-3 py-2 font-medium">{labels.field}</th><th scope="col" className="px-3 py-2 font-medium">{labels.status}</th><th scope="col" className="px-3 py-2 text-right font-medium">{labels.actions}</th></tr>
-        </thead>
-        <tbody className="divide-y">
-          {selected.length === 0 ? <tr><td colSpan={3} className="px-3 py-4 text-center text-gray-600">{labels.noSelectedFields}</td></tr> : selected.map((name, index) => <tr key={name} className={available.has(name) ? '' : 'bg-amber-50'}>
-            <td className="px-3 py-2 font-medium">{name}</td>
-            <td className="px-3 py-2">{available.has(name) ? labels.available : <span className="text-amber-800">{labels.unavailableField}</span>}</td>
-            <td className="px-3 py-2"><div className="flex justify-end gap-2">
-              <button type="button" disabled={index === 0} aria-label={`${labels.moveUp} ${name}`} onClick={() => move(index, -1)} className="rounded border p-1 disabled:opacity-40"><ArrowUp className="h-4 w-4" /></button>
-              <button type="button" disabled={index === selected.length - 1} aria-label={`${labels.moveDown} ${name}`} onClick={() => move(index, 1)} className="rounded border p-1 disabled:opacity-40"><ArrowDown className="h-4 w-4" /></button>
-              <button type="button" aria-label={`${labels.removeField} ${name}`} onClick={() => setSelected(selected.filter((value) => value !== name))} className="rounded border border-red-200 p-1 text-red-700"><X className="h-4 w-4" /></button>
-            </div></td>
-          </tr>)}
-        </tbody>
-      </table>
-    </div>
-    <label className="mt-4 block text-sm font-medium" htmlFor="criteria-doi-field">{labels.doiField}</label>
-    <select id="criteria-doi-field" value={state.criteria.citation_fields.doi || ''} onChange={(event) => dispatch({ type: 'set-doi', value: event.target.value || null })} className="mt-1 w-full rounded-md border px-3 py-2">
-      <option value="">{labels.noDoiField}</option>
-      {state.criteria.citation_fields.doi && !available.has(state.criteria.citation_fields.doi) ? <option value={state.criteria.citation_fields.doi}>{state.criteria.citation_fields.doi} · {labels.unavailableField}</option> : null}
-      {contract.fields.map((field) => <option key={field.name} value={field.name}>{field.name}</option>)}
-    </select>
+      <p className="text-sm text-gray-600">{labels.titleAbstractFieldsDescription}</p>
+      <div className="mt-4 overflow-x-auto rounded-md border">
+        <table className="w-full text-left text-sm">
+          <thead className="bg-gray-50"><tr><th scope="col" className="px-3 py-2 font-medium">{labels.field}</th><th scope="col" className="px-3 py-2 font-medium">{labels.selection}</th><th scope="col" className="px-3 py-2 font-medium">{labels.status}</th><th scope="col" className="px-3 py-2 text-right font-medium">{labels.actions}</th></tr></thead>
+          <tbody className="divide-y">
+            {rows.map((row) => <tr key={`${row.kind}-${row.index ?? 0}`} className={!row.column || !available.has(row.column) ? 'bg-amber-50' : ''}>
+              <td className="px-3 py-2 font-medium">{labelFor(row.kind)}{row.kind !== 'additional' && <span className="ml-2 text-xs font-normal text-gray-500">{labels.required}</span>}</td>
+              <td className="px-3 py-2"><select aria-label={`${labelFor(row.kind)} ${labels.selection}`} value={row.column || ''} onChange={(event) => setColumn(row, event.target.value || null)} className="w-full rounded-md border px-2 py-1"><option value="">{labels.selectColumn}</option>{options(row.column).map((option) => <option key={option.name} value={option.name}>{option.name}{option.unavailable ? ` · ${labels.unavailableField}` : ''}</option>)}</select></td>
+              <td className="px-3 py-2">{!row.column ? <span className="text-amber-800">{statusFor(row)}</span> : available.has(row.column) ? labels.available : <span className="text-amber-800">{labels.unavailableField}</span>}</td>
+              <td className="px-3 py-2"><div className="flex justify-end gap-2">{row.kind === 'additional' && row.index !== undefined && <><button type="button" disabled={row.index === 0} aria-label={`${labels.moveUp} ${row.column || labels.additionalField}`} onClick={() => move(row.index!, -1)} className="rounded border p-1 disabled:opacity-40"><ArrowUp className="h-4 w-4" /></button><button type="button" disabled={row.index === additional.length - 1} aria-label={`${labels.moveDown} ${row.column || labels.additionalField}`} onClick={() => move(row.index!, 1)} className="rounded border p-1 disabled:opacity-40"><ArrowDown className="h-4 w-4" /></button><button type="button" aria-label={`${labels.removeField} ${row.column || labels.additionalField}`} onClick={() => dispatch({ type: 'set-citation-fields', value: additional.filter((_item, index) => index !== row.index) })} className="rounded border border-red-200 p-1 text-red-700"><X className="h-4 w-4" /></button></>}</div></td>
+            </tr>)}
+          </tbody>
+        </table>
+      </div>
+      <label className="mt-4 block text-sm font-medium" htmlFor="criteria-doi-field">{labels.doiField}</label>
+      <select id="criteria-doi-field" value={state.criteria.citation_fields.doi || ''} onChange={(event) => dispatch({ type: 'set-doi', value: event.target.value || null })} className="mt-1 w-full rounded-md border px-3 py-2"><option value="">{labels.noDoiField}</option>{state.criteria.citation_fields.doi && !available.has(state.criteria.citation_fields.doi) ? <option value={state.criteria.citation_fields.doi}>{state.criteria.citation_fields.doi} · {labels.unavailableField}</option> : null}{contract.fields.map((field) => <option key={field.name} value={field.name}>{field.name}</option>)}</select>
     </div> : null}
   </section>
 }
diff --git a/frontend/components/can-sr/setup/criteria-builder.tsx b/frontend/components/can-sr/setup/criteria-builder.tsx
index 3646ae0a..81d4da54 100644
--- a/frontend/components/can-sr/setup/criteria-builder.tsx
+++ b/frontend/components/can-sr/setup/criteria-builder.tsx
@@ -31,6 +31,7 @@ function QuestionCard({
   referencedAnswerIds,
   collapsed,
   onToggle,
+  citationFields,
 }: {
   question: ScreeningQuestion
   index: number
@@ -44,6 +45,7 @@ function QuestionCard({
   referencedAnswerIds: Set<string>
   collapsed: boolean
   onToggle: () => void
+  citationFields: CitationFieldContract
 }) {
   const prefix = `${stage}-${question.id}`
   const [contextAnswerId, setContextAnswerId] = useState<string | null>(null)
@@ -69,6 +71,25 @@ function QuestionCard({
       {!collapsed ? <div id={`${prefix}-body`}>
       <label className="mt-3 block text-sm font-medium" htmlFor={`${prefix}-question`}>{labels.questionText}</label>
       <input id={`${prefix}-question`} value={question.question} onChange={(event) => dispatch({ type: 'update-question', stage, questionId: question.id, field: 'question', value: event.target.value })} className="mt-1 w-full rounded-md border px-3 py-2" />
+      <div className="mt-3 rounded-md border border-dashed border-emerald-300 bg-emerald-50/40 p-3">
+        <label className="flex items-center gap-2 text-sm font-medium" htmlFor={`${prefix}-retrospective-validation`}>
+          <input
+            id={`${prefix}-retrospective-validation`}
+            type="checkbox"
+            checked={Boolean(question.answer_column)}
+            onChange={(event) => dispatch({ type: 'set-answer-column', itemId: question.id, value: event.target.checked ? citationFields.fields[0]?.name || null : null })}
+          />
+          {labels.enableRetrospectiveValidation || 'Enable retrospective validation'}
+        </label>
+        {question.answer_column ? <>
+          <p className="mt-1 text-xs text-gray-600">{labels.answerColumnTooltip}</p>
+          <select id={`${prefix}-answer-column`} value={question.answer_column} onChange={(event) => dispatch({ type: 'set-answer-column', itemId: question.id, value: event.target.value || null })} className="mt-2 w-full rounded-md border px-2 py-1.5 text-sm" title={labels.answerColumnTooltip}>
+            {!citationFields.fields.some((field) => field.name === question.answer_column) ? <option value={question.answer_column}>{question.answer_column} · {labels.unavailableField}</option> : null}
+            <option value="">{labels.noAnswerColumn}</option>
+            {citationFields.fields.map((field) => <option key={field.name} value={field.name}>{field.name}</option>)}
+          </select>
+        </> : null}
+      </div>
       <fieldset className="mt-4 space-y-3">
         <legend className="text-sm font-semibold">{labels.answers}</legend>
         {question.answers.map((answer, answerIndex) => (
@@ -222,13 +243,13 @@ export default function CriteriaBuilder({ state, dispatch, labels, diagnostics =
           {state.criteria[stage].length === 0 ? <p className="rounded-md border border-dashed p-4 text-sm text-gray-500">{labels.noQuestions}</p> : null}
           {state.criteria[stage].map((question, index) => {
             const position = ordered.findIndex((item) => item.question.id === question.id)
-            return <QuestionCard key={question.id} question={question} index={index} count={state.criteria[stage].length} stage={stage} dispatch={dispatch} labels={labels} sources={ordered.slice(0, position)} diagnostics={diagnostics.filter((item) => item.itemId === question.id)} sourceReferenced={conditions.some((condition) => condition.source_item_id === question.id)} referencedAnswerIds={new Set(conditions.filter((condition) => condition.source_item_id === question.id).map((condition) => condition.option_id))} collapsed={collapsedQuestions.has(question.id)} onToggle={() => setCollapsedQuestions((current) => { const next = new Set(current); if (next.has(question.id)) next.delete(question.id); else next.add(question.id); return next })} />
+            return <QuestionCard key={question.id} question={question} index={index} count={state.criteria[stage].length} stage={stage} dispatch={dispatch} labels={labels} sources={ordered.slice(0, position)} citationFields={citationFields} diagnostics={diagnostics.filter((item) => item.itemId === question.id)} sourceReferenced={conditions.some((condition) => condition.source_item_id === question.id)} referencedAnswerIds={new Set(conditions.filter((condition) => condition.source_item_id === question.id).map((condition) => condition.option_id))} collapsed={collapsedQuestions.has(question.id)} onToggle={() => setCollapsedQuestions((current) => { const next = new Set(current); if (next.has(question.id)) next.delete(question.id); else next.add(question.id); return next })} />
           })}
           <div className="flex justify-end"><button type="button" onClick={() => addQuestion(stage)} className="inline-flex items-center gap-2 rounded-md border border-emerald-600 px-3 py-2 text-sm font-medium text-emerald-700"><Plus className="h-4 w-4" />{labels.addQuestion}</button></div>
           </div> : null}
         </section>
       ))}
-      <ParameterBuilder state={state} dispatch={dispatch} labels={labels} diagnostics={diagnostics} />
+      <ParameterBuilder state={state} dispatch={dispatch} labels={labels} diagnostics={diagnostics} citationFields={citationFields} />
     </div>
   )
 }
diff --git a/frontend/components/can-sr/setup/criteria-editor.tsx b/frontend/components/can-sr/setup/criteria-editor.tsx
index 04174902..56036065 100644
--- a/frontend/components/can-sr/setup/criteria-editor.tsx
+++ b/frontend/components/can-sr/setup/criteria-editor.tsx
@@ -19,9 +19,9 @@ import {
   AlertDialogTitle,
 } from '@/components/ui/alert-dialog'
 
-type Props = { srId: string; labels: Record<string, string>; hasScreeningData: boolean }
+type Props = { srId: string; labels: Record<string, string>; hasScreeningData: boolean; reloadKey?: number }
 
-export default function CriteriaEditor({ srId, labels, hasScreeningData }: Props) {
+export default function CriteriaEditor({ srId, labels, hasScreeningData, reloadKey = 0 }: Props) {
   const [state, dispatch] = useReducer(criteriaDraftReducer, { criteria: emptyCriteria(), revision: 0, dirty: false })
   const [status, setStatus] = useState(labels.loading)
   const [migrationFingerprint, setMigrationFingerprint] = useState<string | null>(null)
@@ -31,21 +31,36 @@ export default function CriteriaEditor({ srId, labels, hasScreeningData }: Props
   const [recoveryMode, setRecoveryMode] = useState<RecoveryMode | null>(null)
   const [saveConfirmOpen, setSaveConfirmOpen] = useState(false)
   const [citationFields, setCitationFields] = useState<CitationFieldContract>({ fields: [], unavailable_configured_fields: [] })
-  const clientDiagnostics = useMemo(() => validateCriteriaDraft(state.criteria), [state.criteria])
+  const availableFieldNames = useMemo(() => new Set(citationFields.fields.map((field) => field.name)), [citationFields.fields])
+  const clientDiagnostics = useMemo(() => validateCriteriaDraft(state.criteria, availableFieldNames), [availableFieldNames, state.criteria])
   const diagnostics = [...clientDiagnostics, ...serverDiagnostics]
 
-  const endpoint = `/api/can-sr/reviews/criteria-config?sr_id=${encodeURIComponent(srId)}`
+  // Include the upload revision in the request URL so the callback is recreated
+  // after a successful upload without changing the effect dependency shape.
+  const endpoint = `/api/can-sr/reviews/criteria-config?sr_id=${encodeURIComponent(srId)}&reload=${reloadKey}`
   const load = useCallback(async () => {
+    if (!srId) return
     setStatus(labels.loading)
-    const response = await authenticatedFetch(endpoint)
+    const fieldsEndpoint = `/api/can-sr/reviews/citation-fields?sr_id=${encodeURIComponent(srId)}&reload=${reloadKey}`
+    const [response, fieldsResponse] = await Promise.all([
+      authenticatedFetch(endpoint, { cache: 'no-store' }),
+      authenticatedFetch(fieldsEndpoint, { cache: 'no-store' }),
+    ])
     const data = await response.json().catch(() => ({}))
-    if (!response.ok) { setStatus(data?.detail || labels.loadFailed); return }
+    const fieldsData = await fieldsResponse.json().catch(() => ({}))
+    if (fieldsResponse.ok) setCitationFields(fieldsData)
+    if (!response.ok) {
+      const detail = data?.detail
+      const message = typeof detail === 'string'
+        ? detail
+        : backendDiagnostics(detail)[0]?.message || labels.loadFailed
+      setStatus(message)
+      return
+    }
     dispatch({ type: 'replace', criteria: data.criteria as CriteriaConfig, revision: data.revision || 0 })
-    const fieldsResponse = await authenticatedFetch(`/api/can-sr/reviews/citation-fields?sr_id=${encodeURIComponent(srId)}`)
-    if (fieldsResponse.ok) setCitationFields(await fieldsResponse.json())
     setMigrationFingerprint(data?.migration?.fingerprint || null)
     setStatus(data?.migration?.requires_confirmation ? labels.migrationWarning : '')
-  }, [endpoint, labels, srId])
+  }, [endpoint, labels, reloadKey, srId])
 
   useEffect(() => { void load() }, [load])
   useEffect(() => { setServerDiagnostics([]) }, [state.criteria])
@@ -66,8 +81,8 @@ export default function CriteriaEditor({ srId, labels, hasScreeningData }: Props
       method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify(state.criteria),
     })
     const validationData = await validationResponse.json().catch(() => ({}))
-    if (!validationResponse.ok) {
-      const nextDiagnostics = backendDiagnostics(validationData?.detail).map((diagnostic) => {
+    if (!validationResponse.ok || validationData?.valid === false) {
+      const nextDiagnostics = backendDiagnostics(validationData?.detail ?? validationData).map((diagnostic) => {
         const parts = diagnostic.path.split('.')
         const collection = parts[0] as 'l1' | 'l2' | 'parameters'
         const index = Number(parts[1])
@@ -91,7 +106,9 @@ export default function CriteriaEditor({ srId, labels, hasScreeningData }: Props
         setStatus(labels.conflictDetected)
         return
       }
-      setStatus(typeof detail === 'string' ? detail : detail?.message || labels.saveFailed)
+      setStatus(typeof detail === 'string'
+        ? detail
+        : detail?.message || backendDiagnostics(detail)[0]?.message || labels.saveFailed)
       return
     }
     dispatch({ type: 'replace', criteria: data.criteria, revision: data.revision })
@@ -112,7 +129,7 @@ export default function CriteriaEditor({ srId, labels, hasScreeningData }: Props
       method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify({ criteria_yaml: await file.text() }),
     })
     const data = await response.json().catch(() => ({}))
-    if (!response.ok) { setStatus(data?.detail?.errors?.[0]?.message || labels.importFailed); return }
+    if (!response.ok) { setStatus(backendDiagnostics(data?.detail)[0]?.message || labels.importFailed); return }
     setImportPreview(data as CriteriaImportPreview)
     setStatus(labels.importReady)
   }
diff --git a/frontend/components/can-sr/setup/criteria-types.ts b/frontend/components/can-sr/setup/criteria-types.ts
index cfce5835..354226ca 100644
--- a/frontend/components/can-sr/setup/criteria-types.ts
+++ b/frontend/components/can-sr/setup/criteria-types.ts
@@ -12,6 +12,7 @@ export type ScreeningQuestion = {
   question: string
   answers: ScreeningAnswer[]
   trigger: { all: Array<{ source_item_id: string; option_id: string }> }
+  answer_column?: string | null
 }
 
 export type ParameterOption = {
@@ -28,6 +29,7 @@ type ParameterBase = {
   calculation?: string | null
   trigger: ScreeningQuestion['trigger']
   legacy_category?: string | null
+  answer_column?: string | null
 }
 
 export type TextParameter = ParameterBase & { type: 'text' }
@@ -40,7 +42,7 @@ export type Parameter = TextParameter | SelectionParameter
 
 export type CriteriaConfig = {
   schema_version: 2
-  citation_fields: { l1_include: string[]; doi?: string | null }
+  citation_fields: { l1_include: string[]; title?: string | null; abstract?: string | null; doi?: string | null }
   l1: ScreeningQuestion[]
   l2: ScreeningQuestion[]
   parameters: Parameter[]
@@ -55,7 +57,10 @@ export type CriteriaDraftState = {
 export type CriteriaDraftAction =
   | { type: 'replace'; criteria: CriteriaConfig; revision: number }
   | { type: 'set-citation-fields'; value: string[] }
+  | { type: 'set-title-field'; value: string | null }
+  | { type: 'set-abstract-field'; value: string | null }
   | { type: 'set-doi'; value: string | null }
+  | { type: 'set-answer-column'; itemId: string; value: string | null }
   | { type: 'add-question'; stage: 'l1' | 'l2' }
   | { type: 'delete-question'; stage: 'l1' | 'l2'; questionId: string }
   | { type: 'move-question'; stage: 'l1' | 'l2'; questionId: string; direction: -1 | 1 }
@@ -81,7 +86,7 @@ export type CriteriaDraftAction =
 
 export const emptyCriteria = (): CriteriaConfig => ({
   schema_version: 2,
-  citation_fields: { l1_include: [], doi: null },
+  citation_fields: { l1_include: [], title: 'Title', abstract: 'Abstract', doi: null },
   l1: [],
   l2: [],
   parameters: [],
@@ -147,6 +152,14 @@ export function criteriaDraftReducer(
   if (action.type === 'set-doi') {
     return { ...state, dirty: true, criteria: { ...state.criteria, citation_fields: { ...state.criteria.citation_fields, doi: action.value } } }
   }
+  if (action.type === 'set-title-field' || action.type === 'set-abstract-field') {
+    const key = action.type === 'set-title-field' ? 'title' : 'abstract'
+    return { ...state, dirty: true, criteria: { ...state.criteria, citation_fields: { ...state.criteria.citation_fields, [key]: action.value } } }
+  }
+  if (action.type === 'set-answer-column') {
+    const updateItem = <T extends { id: string }>(item: T): T => item.id === action.itemId ? { ...item, answer_column: action.value } : item
+    return { ...state, dirty: true, criteria: { ...state.criteria, l1: state.criteria.l1.map(updateItem), l2: state.criteria.l2.map(updateItem), parameters: state.criteria.parameters.map(updateItem) } }
+  }
   if (action.type === 'add-question') {
     return updateStage(state, action.stage, (questions) => [...questions, createQuestion()])
   }
diff --git a/frontend/components/can-sr/setup/criteria-validation.test.tsx b/frontend/components/can-sr/setup/criteria-validation.test.tsx
index da27e459..308bfd28 100644
--- a/frontend/components/can-sr/setup/criteria-validation.test.tsx
+++ b/frontend/components/can-sr/setup/criteria-validation.test.tsx
@@ -20,4 +20,15 @@ describe('criteria validation', () => {
       { path: 'parameters.0.name', message: 'Field required' },
     ])
   })
+
+  it('reports required citation fields that are not available in the uploaded data', () => {
+    const criteria = emptyCriteria()
+    criteria.citation_fields.title = 'title'
+    criteria.citation_fields.abstract = 'abstract'
+
+    expect(validateCriteriaDraft(criteria, new Set(['Title', 'Abstract']))).toEqual(expect.arrayContaining([
+      { path: 'citation_fields.title', message: 'Title source column title is not available.' },
+      { path: 'citation_fields.abstract', message: 'Abstract source column abstract is not available.' },
+    ]))
+  })
 })
diff --git a/frontend/components/can-sr/setup/criteria-validation.ts b/frontend/components/can-sr/setup/criteria-validation.ts
index 07236c8f..6185c4ff 100644
--- a/frontend/components/can-sr/setup/criteria-validation.ts
+++ b/frontend/components/can-sr/setup/criteria-validation.ts
@@ -4,12 +4,17 @@ export type CriteriaDiagnostic = { path: string; message: string; itemId?: strin
 
 const idPattern = /^[a-z][a-z0-9_-]{2,63}$/
 
-export function validateCriteriaDraft(criteria: CriteriaConfig): CriteriaDiagnostic[] {
+export function validateCriteriaDraft(criteria: CriteriaConfig, availableFields?: Set<string>): CriteriaDiagnostic[] {
   const diagnostics: CriteriaDiagnostic[] = []
   const seenIds = new Set<string>()
   const available = new Map<string, Set<string> | null>()
   const add = (path: string, message: string, itemId?: string) => diagnostics.push({ path, message, itemId })
 
+  if (!criteria.citation_fields.title?.trim()) add('citation_fields.title', 'A Title source column is required.')
+  else if (availableFields && !availableFields.has(criteria.citation_fields.title)) add('citation_fields.title', `Title source column ${criteria.citation_fields.title} is not available.`)
+  if (!criteria.citation_fields.abstract?.trim()) add('citation_fields.abstract', 'An Abstract source column is required.')
+  else if (availableFields && !availableFields.has(criteria.citation_fields.abstract)) add('citation_fields.abstract', `Abstract source column ${criteria.citation_fields.abstract} is not available.`)
+
   const validateId = (id: string, path: string, itemId: string) => {
     if (!idPattern.test(id)) add(path, 'The stable ID has an invalid format.', itemId)
     if (seenIds.has(id)) add(path, 'The stable ID must be unique across the review.', itemId)
@@ -69,6 +74,12 @@ export function backendDiagnostics(detail: unknown): CriteriaDiagnostic[] {
   return (errors || []).map((error) => {
     const value = error as { loc?: Array<string | number>; path?: Array<string | number>; msg?: string; message?: string }
     const location = value.loc?.filter((part) => part !== 'body') || value.path || []
-    return { path: location.join('.'), message: value.message || value.msg || 'Invalid criteria configuration.' }
+    const rawMessage = value.message || value.msg
+    const message = typeof rawMessage === 'string'
+      ? rawMessage
+      : rawMessage == null
+        ? 'Invalid criteria configuration.'
+        : JSON.stringify(rawMessage)
+    return { path: location.join('.'), message }
   })
 }
diff --git a/frontend/components/can-sr/setup/parameter-builder.tsx b/frontend/components/can-sr/setup/parameter-builder.tsx
index 039beb6e..fd719126 100644
--- a/frontend/components/can-sr/setup/parameter-builder.tsx
+++ b/frontend/components/can-sr/setup/parameter-builder.tsx
@@ -6,6 +6,7 @@ import type { CriteriaDraftAction, CriteriaDraftState, Parameter, ParameterOptio
 import type { CriteriaDiagnostic } from './criteria-validation'
 import ContextEditorDialog from './context-editor-dialog'
 import ParameterDetailsDialog, { type ParameterDetails } from './parameter-details-dialog'
+import type { CitationFieldContract } from './citation-field-selector'
 
 type TriggerSource = { stage: 'l1' | 'l2' | 'parameters'; id: string; label: string; options: ParameterOption[] }
 
@@ -14,7 +15,7 @@ const questionSource = (stage: 'l1' | 'l2', question: ScreeningQuestion): Trigge
   options: question.answers.map(({ id, label, context }) => ({ id, label, context })),
 })
 
-function ParameterCard({ parameter, index, count, sources, dependants, referencedOptionIds, dispatch, labels, diagnostics, collapsed, onToggle }: {
+function ParameterCard({ parameter, index, count, sources, dependants, referencedOptionIds, dispatch, labels, diagnostics, collapsed, onToggle, citationFields }: {
   parameter: Parameter
   index: number
   count: number
@@ -26,6 +27,7 @@ function ParameterCard({ parameter, index, count, sources, dependants, reference
   diagnostics: CriteriaDiagnostic[]
   collapsed: boolean
   onToggle: () => void
+  citationFields: CitationFieldContract
 }) {
   const prefix = `parameter-${parameter.id}`
   const [contextOptionId, setContextOptionId] = useState<string | null>(null)
@@ -55,6 +57,21 @@ function ParameterCard({ parameter, index, count, sources, dependants, reference
       <div><label className="block text-sm font-medium" htmlFor={`${prefix}-name`}>{labels.parameterName}</label><input id={`${prefix}-name`} value={parameter.name} onChange={(event) => update('name', event.target.value)} className="mt-1 w-full rounded-md border px-3 py-2" /></div>
       <div><label className="block text-sm font-medium" htmlFor={`${prefix}-type`}>{labels.parameterType}</label><select id={`${prefix}-type`} value={parameter.type} onChange={(event) => dispatch({ type: 'set-parameter-type', parameterId: parameter.id, value: event.target.value as 'text' | 'selection' })} className="mt-1 w-full rounded-md border px-3 py-2"><option value="text" disabled={typeChangeBlocked}>{labels.freeText}</option><option value="selection">{labels.selectionList}</option></select></div>
     </div>
+    <div className="mt-3 rounded-md border border-dashed border-emerald-300 bg-emerald-50/40 p-3">
+      <label className="flex items-center gap-2 text-sm font-medium" htmlFor={`${prefix}-retrospective-validation`}>
+        <input
+          id={`${prefix}-retrospective-validation`}
+          type="checkbox"
+          checked={Boolean(parameter.answer_column)}
+          onChange={(event) => dispatch({ type: 'set-answer-column', itemId: parameter.id, value: event.target.checked ? citationFields.fields[0]?.name || null : null })}
+        />
+        {labels.enableRetrospectiveValidation || 'Enable retrospective validation'}
+      </label>
+      {parameter.answer_column ? <>
+        <p className="mt-1 text-xs text-gray-600">{labels.answerColumnTooltip}</p>
+        <select id={`${prefix}-answer-column`} value={parameter.answer_column} onChange={(event) => dispatch({ type: 'set-answer-column', itemId: parameter.id, value: event.target.value || null })} title={labels.answerColumnTooltip} className="mt-2 w-full rounded-md border px-2 py-1.5 text-sm"><option value="">{labels.noAnswerColumn}</option>{!citationFields.fields.some((field) => field.name === parameter.answer_column) ? <option value={parameter.answer_column}>{parameter.answer_column} · {labels.unavailableField}</option> : null}{citationFields.fields.map((field) => <option key={field.name} value={field.name}>{field.name}</option>)}</select>
+      </> : null}
+    </div>
     {typeChangeBlocked ? <p role="alert" className="mt-2 rounded bg-amber-50 p-2 text-sm text-amber-900">{labels.typeChangeBlocked}: {dependants.join(', ')}</p> : null}
     <div className="mt-3 flex items-center gap-3 rounded-md bg-gray-50 p-3">
       <div className="min-w-0"><p className="text-sm font-medium">{labels.parameterDetails}</p><p className="text-xs text-gray-600">{detailCount ? `${detailCount} ${labels.detailsConfigured}` : labels.noDetailsConfigured}</p></div>
@@ -62,7 +79,7 @@ function ParameterCard({ parameter, index, count, sources, dependants, reference
     </div>
     {parameter.type === 'selection' ? <fieldset className="mt-4 space-y-3">
       <legend className="text-sm font-semibold">{labels.selectionOptions}</legend>
-      <label className="block text-sm font-medium" htmlFor={`${prefix}-mode`}>{labels.selectionMode}</label><select id={`${prefix}-mode`} value={parameter.selection_mode} onChange={(event) => dispatch({ type: 'set-selection-mode', parameterId: parameter.id, value: event.target.value as 'single' | 'multiple' })} className="w-full rounded-md border px-3 py-2 md:w-64"><option value="single">{labels.singleSelection}</option><option value="multiple">{labels.multipleSelection}</option></select>
+      <label className="block text-sm font-medium" htmlFor={`${prefix}-mode`}>{labels.selectionMode}</label><select id={`${prefix}-mode`} value={parameter.selection_mode} onChange={(event) => dispatch({ type: 'set-selection-mode', parameterId: parameter.id, value: event.target.value as 'single' | 'multiple' })} className="w-full rounded-md border px-3 py-2 md:w-64"><option value="single">Select one answer</option><option value="multiple">Select multiple answers</option></select>
       {parameter.options.map((option, optionIndex) => <div key={option.id} className="grid gap-2 rounded-md bg-gray-50 p-3 md:grid-cols-[1fr_auto_auto]">
         <div><label className="block text-xs font-medium" htmlFor={`${prefix}-${option.id}-label`}>{labels.optionLabel} {optionIndex + 1}</label><input id={`${prefix}-${option.id}-label`} value={option.label} onChange={(event) => dispatch({ type: 'update-option', parameterId: parameter.id, optionId: option.id, field: 'label', value: event.target.value })} className="mt-1 w-full rounded border px-2 py-1.5" />{option.context?.trim() ? <p className="mt-1 text-xs font-medium text-emerald-700">{labels.contextAdded}</p> : null}</div>
         <button type="button" aria-label={`${labels.editOptionContext} ${optionIndex + 1}`} onClick={() => setContextOptionId(option.id)} className="self-end rounded border p-2 text-gray-700" title={labels.editOptionContext}><Pencil className="h-4 w-4" /></button>
@@ -94,7 +111,7 @@ function ParameterCard({ parameter, index, count, sources, dependants, reference
   </article>
 }
 
-export default function ParameterBuilder({ state, dispatch, labels, diagnostics = [] }: { state: CriteriaDraftState; dispatch: Dispatch<CriteriaDraftAction>; labels: Record<string, string>; diagnostics?: CriteriaDiagnostic[] }) {
+export default function ParameterBuilder({ state, dispatch, labels, diagnostics = [], citationFields = { fields: [], unavailable_configured_fields: [] } }: { state: CriteriaDraftState; dispatch: Dispatch<CriteriaDraftAction>; labels: Record<string, string>; diagnostics?: CriteriaDiagnostic[]; citationFields?: CitationFieldContract }) {
   const [sectionCollapsed, setSectionCollapsed] = useState(false)
   const [collapsedParameters, setCollapsedParameters] = useState<Set<string>>(() => new Set())
   const pendingParameterCount = useRef<number | null>(null)
@@ -141,7 +158,7 @@ export default function ParameterBuilder({ state, dispatch, labels, diagnostics
     {state.criteria.parameters.map((parameter, index) => {
       const earlierParameters: TriggerSource[] = state.criteria.parameters.slice(0, index).filter((item) => item.type === 'selection').map((item) => ({ stage: 'parameters', id: item.id, label: item.name, options: item.options }))
       const dependants = state.criteria.parameters.filter((item) => item.id !== parameter.id && item.trigger.all.some((condition) => condition.source_item_id === parameter.id)).map((item) => itemLabels.get(item.id) || item.id)
-      return <ParameterCard key={parameter.id} parameter={parameter} index={index} count={state.criteria.parameters.length} sources={[...screeningSources, ...earlierParameters]} dependants={dependants} referencedOptionIds={new Set(allConditions.filter((condition) => condition.source_item_id === parameter.id).map((condition) => condition.option_id))} dispatch={dispatch} labels={labels} diagnostics={diagnostics.filter((item) => item.itemId === parameter.id)} collapsed={collapsedParameters.has(parameter.id)} onToggle={() => setCollapsedParameters((current) => { const next = new Set(current); if (next.has(parameter.id)) next.delete(parameter.id); else next.add(parameter.id); return next })} />
+      return <ParameterCard key={parameter.id} parameter={parameter} index={index} count={state.criteria.parameters.length} sources={[...screeningSources, ...earlierParameters]} dependants={dependants} referencedOptionIds={new Set(allConditions.filter((condition) => condition.source_item_id === parameter.id).map((condition) => condition.option_id))} dispatch={dispatch} labels={labels} citationFields={citationFields} diagnostics={diagnostics.filter((item) => item.itemId === parameter.id)} collapsed={collapsedParameters.has(parameter.id)} onToggle={() => setCollapsedParameters((current) => { const next = new Set(current); if (next.has(parameter.id)) next.delete(parameter.id); else next.add(parameter.id); return next })} />
     })}
     <div className="flex justify-end"><button type="button" onClick={addParameter} className="inline-flex items-center gap-2 rounded-md border border-emerald-600 px-3 py-2 text-sm font-medium text-emerald-700"><Plus className="h-4 w-4" />{labels.addParameter}</button></div>
     </div> : null}
diff --git a/frontend/dictionaries/en.json b/frontend/dictionaries/en.json
index 095281a8..e1da11d1 100644
--- a/frontend/dictionaries/en.json
+++ b/frontend/dictionaries/en.json
@@ -149,21 +149,146 @@
     "saving": "Saving...",
     "savedSuccessfully": "Saved successfully",
     "criteriaBuilder": {
-      "builderTitle": "Criteria Builder", "builderDescription": "Configure title/abstract fields and screening questions.",
-      "citationFields": "Title/Abstract Fields", "l1Fields": "L1 fields (comma separated)", "doiField": "DOI source field",
-      "titleAbstractFieldsDescription": "Choose and order the citation fields used for title and abstract screening.", "field": "Field", "status": "Status", "actions": "Actions", "available": "Available", "noSelectedFields": "No title/abstract fields configured.", "addField": "Add field", "noCitationFields": "Upload a CSV or RIS file to discover fields.", "unavailableField": "Not found in the current upload", "removeField": "Remove field", "noDoiField": "No DOI field",
-      "l1": "Title and abstract screening (L1)", "l2": "Full-text screening (L2)", "question": "Question", "questionText": "Question text", "answerContext": "Answer context",
-      "answers": "Answers", "answerLabel": "Answer label", "decision": "Decision", "include": "Include", "exclude": "Exclude",
-      "addQuestion": "Add question", "addAnswer": "Add answer", "deleteQuestion": "Delete question", "deleteAnswer": "Delete answer", "moveUp": "Move up", "moveDown": "Move down", "noQuestions": "No questions configured.", "expand": "Expand", "collapse": "Collapse", "minimizeAll": "Minimize section", "maximizeSection": "Show section with items minimized", "maximizeAll": "Expand all items", "untitledQuestion": "Untitled question", "untitledParameter": "Untitled parameter", "errors": "errors",
-      "editAnswerContext": "Edit answer context", "editOptionContext": "Edit option context", "answerContextFor": "Add guidance for the answer", "optionContextFor": "Add guidance for the option", "contextAdded": "Context added", "saveContext": "Save context", "cancel": "Cancel",
+      "builderTitle": "Criteria Builder",
+      "builderDescription": "Configure title/abstract fields and screening questions.",
+      "citationFields": "Title/Abstract Fields",
+      "l1Fields": "L1 fields (comma separated)",
+      "doiField": "DOI source field",
+      "titleAbstractFieldsDescription": "Choose the required Title and Abstract columns, then add and order optional citation fields.",
+      "field": "Field",
+      "status": "Status",
+      "actions": "Actions",
+      "available": "Available",
+      "noSelectedFields": "No title/abstract fields configured.",
+      "addField": "Add field",
+      "noCitationFields": "Upload a CSV or RIS file to discover fields.",
+      "unavailableField": "Not found in the current upload",
+      "removeField": "Remove field",
+      "noDoiField": "No DOI field",
+      "l1": "Title and abstract screening (L1)",
+      "l2": "Full-text screening (L2)",
+      "question": "Question",
+      "questionText": "Question text",
+      "answerContext": "Answer context", "answerColumn": "Answer column", "addAnswerColumn": "Add Answer Column", "enableRetrospectiveValidation": "Enable retrospective validation", "answerColumnTooltip": "Match an existing human answer from an imported file when one exists. Valid matches are authoritative; blank or unmatched values are warned about.", "noAnswerColumn": "No answer column", "unavailableField": "Unavailable field",
+      "answers": "Answers",
+      "answerLabel": "Answer label",
+      "decision": "Decision",
+      "include": "Include",
+      "exclude": "Exclude",
+      "addQuestion": "Add question",
+      "addAnswer": "Add answer",
+      "deleteQuestion": "Delete question",
+      "deleteAnswer": "Delete answer",
+      "moveUp": "Move up",
+      "moveDown": "Move down",
+      "noQuestions": "No questions configured.",
+      "expand": "Expand",
+      "collapse": "Collapse",
+      "minimizeAll": "Minimize section",
+      "maximizeSection": "Show section with items minimized",
+      "maximizeAll": "Expand all items",
+      "untitledQuestion": "Untitled question",
+      "untitledParameter": "Untitled parameter",
+      "errors": "errors",
+      "editAnswerContext": "Edit answer context",
+      "editOptionContext": "Edit option context",
+      "answerContextFor": "Add guidance for the answer",
+      "optionContextFor": "Add guidance for the option",
+      "contextAdded": "Context added",
+      "saveContext": "Save context",
+      "cancel": "Continue editing",
       "deleteDependencyWarning": "This item may be referenced by a later condition. Delete it and leave affected conditions marked for repair?",
-      "triggers": "Conditional visibility", "triggerDescription": "Show this question only when all of the following are true.", "alwaysShown": "Always shown", "conditionsConfigured": "conditions configured", "triggerSource": "Earlier question", "triggerAnswer": "Required answer", "addTrigger": "Add condition", "removeTrigger": "Remove condition", "invalidReference": "Invalid reference", "triggerOrderError": "This condition references a missing, current, or later question. Retarget it or remove the condition.", "missingAnswerError": "The referenced answer no longer exists. Select another answer or remove the condition.",
-      "parameters": "Extraction parameters", "parameter": "Parameter", "addParameter": "Add parameter", "deleteParameter": "Delete parameter", "moveUpParameter": "Move parameter up", "moveDownParameter": "Move parameter down", "noParameters": "No extraction parameters configured.", "parameterName": "Name", "parameterDescription": "Description", "parameterType": "Type", "freeText": "Free text", "selectionList": "Selection from list", "unitInstructions": "Units and reporting instructions", "calculation": "Calculation instructions", "parameterDetails": "Description and instructions", "editDetails": "Edit details", "editParameterDetails": "Edit parameter details", "parameterDetailsFor": "Configure description and instructions for", "detailsConfigured": "of 3 details configured", "noDetailsConfigured": "No description or instructions configured", "saveDetails": "Save details", "selectionOptions": "Selection options", "selectionMode": "Selection mode", "singleSelection": "One answer", "multipleSelection": "Multiple answers", "optionLabel": "Option label", "optionContext": "Option context", "addOption": "Add option", "deleteOption": "Delete option", "typeChangeBlocked": "Free text is unavailable because later parameters depend on this list. Remove or retarget these conditions first",
-      "validationPassed": "Draft validation passed.", "validationFailed": "Fix validation errors before saving.", "validationErrors": "{count} validation error(s)", "questionsCount": "questions", "parametersCount": "parameters",
-      "importPreviewTitle": "Review YAML import", "importPreviewDescription": "The backend normalized this file into canonical criteria. Review the summary and warnings before replacing your current draft.", "importFormat": "Detected format", "legacyYaml": "Legacy YAML (normalized to v2)", "canonicalYaml": "Canonical criteria v2", "importWarnings": "Import notices", "importNoWarnings": "No import warnings were reported.", "importConfirmationRequired": "This migration contains inferred or ambiguous values that require review before the first save.", "cancelImport": "Keep current draft", "replaceDraft": "Replace draft", "importReady": "YAML import is ready for review.", "importCancelled": "YAML import cancelled; the current draft was preserved.",
-      "conflictTitle": "A newer criteria revision is available", "conflictDescription": "Another editor saved criteria after you loaded this draft. Reload the latest revision before making further changes.", "conflictPreserved": "Your local draft has not been overwritten. Download it before reloading if you need to retain your changes.", "reloadTitle": "Discard unsaved criteria changes?", "reloadDescription": "Reloading replaces your current draft with the latest saved configuration.", "exportLocalDraft": "Download local draft", "discardAndReload": "Discard and reload", "cancel": "Continue editing", "conflictDetected": "Save stopped because a newer criteria revision exists.",
-      "importYaml": "Import", "downloadYaml": "Download", "reload": "Reload", "save": "Save criteria", "loading": "Loading criteria…", "saving": "Saving…", "saved": "Saved successfully.", "unsaved": "Unsaved changes", "imported": "YAML imported into the draft.", "confirmSaveTitle": "Save criteria changes?", "confirmSaveDescription": "Updating these criteria may invalidate existing screening results.", "cancel": "Continue editing",
-      "loadFailed": "Unable to load criteria.", "saveFailed": "Unable to save criteria.", "importFailed": "Unable to import YAML.", "downloadFailed": "Unable to download YAML.", "migrationWarning": "Legacy YAML was normalized. Review inferred decisions before saving.", "confirmInvalidation": "I understand this change may invalidate existing screening results."
+      "triggers": "Conditional visibility",
+      "triggerDescription": "Show this question only when all of the following are true.",
+      "alwaysShown": "Always shown",
+      "conditionsConfigured": "conditions configured",
+      "triggerSource": "Earlier question",
+      "triggerAnswer": "Required answer",
+      "addTrigger": "Add condition",
+      "removeTrigger": "Remove condition",
+      "invalidReference": "Invalid reference",
+      "triggerOrderError": "This condition references a missing, current, or later question. Retarget it or remove the condition.",
+      "missingAnswerError": "The referenced answer no longer exists. Select another answer or remove the condition.",
+      "parameters": "Extraction parameters",
+      "parameter": "Parameter",
+      "addParameter": "Add parameter",
+      "deleteParameter": "Delete parameter",
+      "moveUpParameter": "Move parameter up",
+      "moveDownParameter": "Move parameter down",
+      "noParameters": "No extraction parameters configured.",
+      "parameterName": "Name",
+      "parameterDescription": "Description",
+      "parameterType": "Type",
+      "freeText": "Free text",
+      "selectionList": "Selection from list",
+      "unitInstructions": "Units and reporting instructions",
+      "calculation": "Calculation instructions",
+      "parameterDetails": "Description and instructions",
+      "editDetails": "Edit details",
+      "editParameterDetails": "Edit parameter details",
+      "parameterDetailsFor": "Configure description and instructions for",
+      "detailsConfigured": "of 3 details configured",
+      "noDetailsConfigured": "No description or instructions configured",
+      "saveDetails": "Save details",
+      "selectionOptions": "Selection options",
+      "selectionMode": "Selection mode",
+      "singleSelection": "Select one answer",
+      "multipleSelection": "Select multiple answers",
+      "optionLabel": "Option label",
+      "optionContext": "Option context",
+      "addOption": "Add option",
+      "deleteOption": "Delete option",
+      "typeChangeBlocked": "Free text is unavailable because later parameters depend on this list. Remove or retarget these conditions first",
+      "validationPassed": "Draft validation passed.",
+      "validationFailed": "Fix validation errors before saving.",
+      "validationErrors": "{count} validation error(s)",
+      "questionsCount": "questions",
+      "parametersCount": "parameters",
+      "importPreviewTitle": "Review YAML import",
+      "importPreviewDescription": "The backend normalized this file into canonical criteria. Review the summary and warnings before replacing your current draft.",
+      "importFormat": "Detected format",
+      "legacyYaml": "Legacy YAML (normalized to v2)",
+      "canonicalYaml": "Canonical criteria v2",
+      "importWarnings": "Import notices",
+      "importNoWarnings": "No import warnings were reported.",
+      "importConfirmationRequired": "This migration contains inferred or ambiguous values that require review before the first save.",
+      "cancelImport": "Keep current draft",
+      "replaceDraft": "Replace draft",
+      "importReady": "YAML import is ready for review.",
+      "importCancelled": "YAML import cancelled; the current draft was preserved.",
+      "conflictTitle": "A newer criteria revision is available",
+      "conflictDescription": "Another editor saved criteria after you loaded this draft. Reload the latest revision before making further changes.",
+      "conflictPreserved": "Your local draft has not been overwritten. Download it before reloading if you need to retain your changes.",
+      "reloadTitle": "Discard unsaved criteria changes?",
+      "reloadDescription": "Reloading replaces your current draft with the latest saved configuration.",
+      "exportLocalDraft": "Download local draft",
+      "discardAndReload": "Discard and reload",
+      "conflictDetected": "Save stopped because a newer criteria revision exists.",
+      "importYaml": "Import",
+      "downloadYaml": "Download",
+      "reload": "Reload",
+      "save": "Save criteria",
+      "loading": "Loading criteria…",
+      "saving": "Saving…",
+      "saved": "Saved successfully.",
+      "unsaved": "Unsaved changes",
+      "imported": "YAML imported into the draft.",
+      "confirmSaveTitle": "Save criteria changes?",
+      "confirmSaveDescription": "Updating these criteria may invalidate existing screening results.",
+      "loadFailed": "Unable to load criteria.",
+      "saveFailed": "Unable to save criteria.",
+      "importFailed": "Unable to import YAML.",
+      "downloadFailed": "Unable to download YAML.",
+      "migrationWarning": "Legacy YAML was normalized. Review inferred decisions before saving.",
+      "confirmInvalidation": "I understand this change may invalidate existing screening results.",
+      "selection": "Selection",
+      "required": "Required",
+      "requiredField": "Required selection",
+      "selectColumn": "Select a column",
+      "titleField": "Title",
+      "abstractField": "Abstract",
+      "additionalField": "Additional field",
+      "addAdditionalField": "Add additional field"
     }
   },
   "search": {
@@ -220,7 +345,8 @@
     "runAllL1Desc": "Runs title/abstract screening on all citations in the list.",
     "runAllL2Desc": "Runs full-text screening on citations that passed L1 and have an uploaded PDF.",
     "runAllExtractDesc": "Runs parameter extraction on citations that passed L2 and have an uploaded PDF.",
-    "forceRerun": "Force re-run (overwrite existing AI results)",
+    "skipExistingAI": "Skip existing AI answers",
+    "skipExistingHuman": "Skip existing human answers",
     "runningAllAI": "Running all AI...",
     "confirmRunAllAI": "Run AI for all citations in this list? This may take a while.",
     "runAllFailed": "Run all AI failed",
diff --git a/frontend/dictionaries/fr.json b/frontend/dictionaries/fr.json
index 7cc57053..2a03c495 100644
--- a/frontend/dictionaries/fr.json
+++ b/frontend/dictionaries/fr.json
@@ -149,21 +149,146 @@
     "saving": "Enregistrement...",
     "savedSuccessfully": "Enregistré avec succès",
     "criteriaBuilder": {
-      "builderTitle": "Générateur de critères", "builderDescription": "Configurez les champs de titre/résumé et les questions.",
-      "citationFields": "Champs de titre/résumé", "l1Fields": "Champs L1 (séparés par des virgules)", "doiField": "Champ source du DOI",
-      "titleAbstractFieldsDescription": "Choisissez et ordonnez les champs de citation utilisés pour la sélection du titre et du résumé.", "field": "Champ", "status": "État", "actions": "Actions", "available": "Disponible", "noSelectedFields": "Aucun champ de titre/résumé configuré.", "addField": "Ajouter un champ", "noCitationFields": "Téléversez un fichier CSV ou RIS pour découvrir les champs.", "unavailableField": "Introuvable dans le téléversement actuel", "removeField": "Supprimer le champ", "noDoiField": "Aucun champ DOI",
-      "l1": "Sélection titre et résumé (L1)", "l2": "Sélection du texte intégral (L2)", "question": "Question", "questionText": "Texte de la question", "answerContext": "Contexte de la réponse",
-      "answers": "Réponses", "answerLabel": "Libellé de réponse", "decision": "Décision", "include": "Inclure", "exclude": "Exclure",
-      "addQuestion": "Ajouter une question", "addAnswer": "Ajouter une réponse", "deleteQuestion": "Supprimer la question", "deleteAnswer": "Supprimer la réponse", "moveUp": "Monter", "moveDown": "Descendre", "noQuestions": "Aucune question configurée.", "expand": "Développer", "collapse": "Réduire", "minimizeAll": "Réduire la section", "maximizeSection": "Afficher la section avec les éléments réduits", "maximizeAll": "Développer tous les éléments", "untitledQuestion": "Question sans titre", "untitledParameter": "Paramètre sans titre", "errors": "erreurs",
-      "editAnswerContext": "Modifier le contexte de la réponse", "editOptionContext": "Modifier le contexte de l’option", "answerContextFor": "Ajouter des directives pour la réponse", "optionContextFor": "Ajouter des directives pour l’option", "contextAdded": "Contexte ajouté", "saveContext": "Enregistrer le contexte", "cancel": "Annuler",
+      "builderTitle": "Générateur de critères",
+      "builderDescription": "Configurez les champs de titre/résumé et les questions.",
+      "citationFields": "Champs de titre/résumé",
+      "l1Fields": "Champs L1 (séparés par des virgules)",
+      "doiField": "Champ source du DOI",
+      "titleAbstractFieldsDescription": "Choisissez les colonnes obligatoires Titre et Résumé, puis ajoutez et ordonnez les champs de citation facultatifs.",
+      "field": "Champ",
+      "status": "État",
+      "actions": "Actions",
+      "available": "Disponible",
+      "noSelectedFields": "Aucun champ de titre/résumé configuré.",
+      "addField": "Ajouter un champ",
+      "noCitationFields": "Téléversez un fichier CSV ou RIS pour découvrir les champs.",
+      "unavailableField": "Introuvable dans le téléversement actuel",
+      "removeField": "Supprimer le champ",
+      "noDoiField": "Aucun champ DOI",
+      "l1": "Sélection titre et résumé (L1)",
+      "l2": "Sélection du texte intégral (L2)",
+      "question": "Question",
+      "questionText": "Texte de la question",
+      "answerContext": "Contexte de réponse", "answerColumn": "Colonne de réponse", "addAnswerColumn": "Ajouter une colonne de réponse", "enableRetrospectiveValidation": "Activer la validation rétrospective", "answerColumnTooltip": "Associer une réponse humaine existante provenant d’un fichier importé. Les correspondances valides sont prioritaires; les valeurs vides ou non correspondantes sont signalées.", "noAnswerColumn": "Aucune colonne de réponse", "unavailableField": "Colonne indisponible",
+      "answers": "Réponses",
+      "answerLabel": "Libellé de réponse",
+      "decision": "Décision",
+      "include": "Inclure",
+      "exclude": "Exclure",
+      "addQuestion": "Ajouter une question",
+      "addAnswer": "Ajouter une réponse",
+      "deleteQuestion": "Supprimer la question",
+      "deleteAnswer": "Supprimer la réponse",
+      "moveUp": "Monter",
+      "moveDown": "Descendre",
+      "noQuestions": "Aucune question configurée.",
+      "expand": "Développer",
+      "collapse": "Réduire",
+      "minimizeAll": "Réduire la section",
+      "maximizeSection": "Afficher la section avec les éléments réduits",
+      "maximizeAll": "Développer tous les éléments",
+      "untitledQuestion": "Question sans titre",
+      "untitledParameter": "Paramètre sans titre",
+      "errors": "erreurs",
+      "editAnswerContext": "Modifier le contexte de la réponse",
+      "editOptionContext": "Modifier le contexte de l’option",
+      "answerContextFor": "Ajouter des directives pour la réponse",
+      "optionContextFor": "Ajouter des directives pour l’option",
+      "contextAdded": "Contexte ajouté",
+      "saveContext": "Enregistrer le contexte",
+      "cancel": "Continuer la modification",
       "deleteDependencyWarning": "Cet élément peut être référencé par une condition ultérieure. Le supprimer et laisser les conditions touchées marquées pour correction?",
-      "triggers": "Visibilité conditionnelle", "triggerDescription": "Afficher cette question uniquement lorsque toutes les conditions suivantes sont remplies.", "alwaysShown": "Toujours affichée", "conditionsConfigured": "conditions configurées", "triggerSource": "Question précédente", "triggerAnswer": "Réponse requise", "addTrigger": "Ajouter une condition", "removeTrigger": "Supprimer une condition", "invalidReference": "Référence non valide", "triggerOrderError": "Cette condition fait référence à une question manquante, courante ou ultérieure. Modifiez-la ou supprimez-la.", "missingAnswerError": "La réponse référencée n’existe plus. Sélectionnez une autre réponse ou supprimez la condition.",
-      "parameters": "Paramètres d’extraction", "parameter": "Paramètre", "addParameter": "Ajouter un paramètre", "deleteParameter": "Supprimer le paramètre", "moveUpParameter": "Monter le paramètre", "moveDownParameter": "Descendre le paramètre", "noParameters": "Aucun paramètre d’extraction configuré.", "parameterName": "Nom", "parameterDescription": "Description", "parameterType": "Type", "freeText": "Texte libre", "selectionList": "Sélection dans une liste", "unitInstructions": "Unités et instructions de présentation", "calculation": "Instructions de calcul", "parameterDetails": "Description et instructions", "editDetails": "Modifier les détails", "editParameterDetails": "Modifier les détails du paramètre", "parameterDetailsFor": "Configurer la description et les instructions pour", "detailsConfigured": "détails sur 3 configurés", "noDetailsConfigured": "Aucune description ni instruction configurée", "saveDetails": "Enregistrer les détails", "selectionOptions": "Options de sélection", "selectionMode": "Mode de sélection", "singleSelection": "Une réponse", "multipleSelection": "Plusieurs réponses", "optionLabel": "Libellé de l’option", "optionContext": "Contexte de l’option", "addOption": "Ajouter une option", "deleteOption": "Supprimer l’option", "typeChangeBlocked": "Le texte libre est indisponible car des paramètres ultérieurs dépendent de cette liste. Supprimez ou modifiez d’abord ces conditions",
-      "validationPassed": "La validation du brouillon a réussi.", "validationFailed": "Corrigez les erreurs de validation avant d’enregistrer.", "validationErrors": "{count} erreur(s) de validation", "questionsCount": "questions", "parametersCount": "paramètres",
-      "importPreviewTitle": "Vérifier l’importation YAML", "importPreviewDescription": "Le serveur a normalisé ce fichier en critères canoniques. Vérifiez le résumé et les avertissements avant de remplacer le brouillon actuel.", "importFormat": "Format détecté", "legacyYaml": "YAML hérité (normalisé en v2)", "canonicalYaml": "Critères canoniques v2", "importWarnings": "Avis d’importation", "importNoWarnings": "Aucun avertissement d’importation n’a été signalé.", "importConfirmationRequired": "Cette migration contient des valeurs déduites ou ambiguës qui doivent être vérifiées avant le premier enregistrement.", "cancelImport": "Conserver le brouillon actuel", "replaceDraft": "Remplacer le brouillon", "importReady": "L’importation YAML est prête à être vérifiée.", "importCancelled": "Importation YAML annulée; le brouillon actuel a été conservé.",
-      "conflictTitle": "Une version plus récente des critères est disponible", "conflictDescription": "Une autre personne a enregistré les critères après le chargement de ce brouillon. Rechargez la dernière version avant de poursuivre.", "conflictPreserved": "Votre brouillon local n’a pas été remplacé. Téléchargez-le avant de recharger si vous devez conserver vos modifications.", "reloadTitle": "Abandonner les modifications non enregistrées?", "reloadDescription": "Le rechargement remplace le brouillon actuel par la dernière configuration enregistrée.", "exportLocalDraft": "Télécharger le brouillon local", "discardAndReload": "Abandonner et recharger", "cancel": "Continuer la modification", "conflictDetected": "L’enregistrement a été interrompu, car une version plus récente des critères existe.",
-      "importYaml": "Importer", "downloadYaml": "Télécharger", "reload": "Recharger", "save": "Enregistrer les critères", "loading": "Chargement des critères…", "saving": "Enregistrement…", "saved": "Enregistré avec succès.", "unsaved": "Modifications non enregistrées", "imported": "YAML importé dans le brouillon.", "confirmSaveTitle": "Enregistrer les modifications des critères?", "confirmSaveDescription": "La mise à jour de ces critères peut invalider les résultats de sélection existants.", "cancel": "Continuer la modification",
-      "loadFailed": "Impossible de charger les critères.", "saveFailed": "Impossible d’enregistrer les critères.", "importFailed": "Impossible d’importer le YAML.", "downloadFailed": "Impossible de télécharger le YAML.", "migrationWarning": "Le YAML hérité a été normalisé. Vérifiez les décisions déduites avant l’enregistrement.", "confirmInvalidation": "Je comprends que ce changement peut invalider les résultats de sélection existants."
+      "triggers": "Visibilité conditionnelle",
+      "triggerDescription": "Afficher cette question uniquement lorsque toutes les conditions suivantes sont remplies.",
+      "alwaysShown": "Toujours affichée",
+      "conditionsConfigured": "conditions configurées",
+      "triggerSource": "Question précédente",
+      "triggerAnswer": "Réponse requise",
+      "addTrigger": "Ajouter une condition",
+      "removeTrigger": "Supprimer une condition",
+      "invalidReference": "Référence non valide",
+      "triggerOrderError": "Cette condition fait référence à une question manquante, courante ou ultérieure. Modifiez-la ou supprimez-la.",
+      "missingAnswerError": "La réponse référencée n’existe plus. Sélectionnez une autre réponse ou supprimez la condition.",
+      "parameters": "Paramètres d’extraction",
+      "parameter": "Paramètre",
+      "addParameter": "Ajouter un paramètre",
+      "deleteParameter": "Supprimer le paramètre",
+      "moveUpParameter": "Monter le paramètre",
+      "moveDownParameter": "Descendre le paramètre",
+      "noParameters": "Aucun paramètre d’extraction configuré.",
+      "parameterName": "Nom",
+      "parameterDescription": "Description",
+      "parameterType": "Type",
+      "freeText": "Texte libre",
+      "selectionList": "Sélection dans une liste",
+      "unitInstructions": "Unités et instructions de présentation",
+      "calculation": "Instructions de calcul",
+      "parameterDetails": "Description et instructions",
+      "editDetails": "Modifier les détails",
+      "editParameterDetails": "Modifier les détails du paramètre",
+      "parameterDetailsFor": "Configurer la description et les instructions pour",
+      "detailsConfigured": "détails sur 3 configurés",
+      "noDetailsConfigured": "Aucune description ni instruction configurée",
+      "saveDetails": "Enregistrer les détails",
+      "selectionOptions": "Options de sélection",
+      "selectionMode": "Mode de sélection",
+      "singleSelection": "Sélectionner une réponse",
+      "multipleSelection": "Sélectionner plusieurs réponses",
+      "optionLabel": "Libellé de l’option",
+      "optionContext": "Contexte de l’option",
+      "addOption": "Ajouter une option",
+      "deleteOption": "Supprimer l’option",
+      "typeChangeBlocked": "Le texte libre est indisponible car des paramètres ultérieurs dépendent de cette liste. Supprimez ou modifiez d’abord ces conditions",
+      "validationPassed": "La validation du brouillon a réussi.",
+      "validationFailed": "Corrigez les erreurs de validation avant d’enregistrer.",
+      "validationErrors": "{count} erreur(s) de validation",
+      "questionsCount": "questions",
+      "parametersCount": "paramètres",
+      "importPreviewTitle": "Vérifier l’importation YAML",
+      "importPreviewDescription": "Le serveur a normalisé ce fichier en critères canoniques. Vérifiez le résumé et les avertissements avant de remplacer le brouillon actuel.",
+      "importFormat": "Format détecté",
+      "legacyYaml": "YAML hérité (normalisé en v2)",
+      "canonicalYaml": "Critères canoniques v2",
+      "importWarnings": "Avis d’importation",
+      "importNoWarnings": "Aucun avertissement d’importation n’a été signalé.",
+      "importConfirmationRequired": "Cette migration contient des valeurs déduites ou ambiguës qui doivent être vérifiées avant le premier enregistrement.",
+      "cancelImport": "Conserver le brouillon actuel",
+      "replaceDraft": "Remplacer le brouillon",
+      "importReady": "L’importation YAML est prête à être vérifiée.",
+      "importCancelled": "Importation YAML annulée; le brouillon actuel a été conservé.",
+      "conflictTitle": "Une version plus récente des critères est disponible",
+      "conflictDescription": "Une autre personne a enregistré les critères après le chargement de ce brouillon. Rechargez la dernière version avant de poursuivre.",
+      "conflictPreserved": "Votre brouillon local n’a pas été remplacé. Téléchargez-le avant de recharger si vous devez conserver vos modifications.",
+      "reloadTitle": "Abandonner les modifications non enregistrées?",
+      "reloadDescription": "Le rechargement remplace le brouillon actuel par la dernière configuration enregistrée.",
+      "exportLocalDraft": "Télécharger le brouillon local",
+      "discardAndReload": "Abandonner et recharger",
+      "conflictDetected": "L’enregistrement a été interrompu, car une version plus récente des critères existe.",
+      "importYaml": "Importer",
+      "downloadYaml": "Télécharger",
+      "reload": "Recharger",
+      "save": "Enregistrer les critères",
+      "loading": "Chargement des critères…",
+      "saving": "Enregistrement…",
+      "saved": "Enregistré avec succès.",
+      "unsaved": "Modifications non enregistrées",
+      "imported": "YAML importé dans le brouillon.",
+      "confirmSaveTitle": "Enregistrer les modifications des critères?",
+      "confirmSaveDescription": "La mise à jour de ces critères peut invalider les résultats de sélection existants.",
+      "loadFailed": "Impossible de charger les critères.",
+      "saveFailed": "Impossible d’enregistrer les critères.",
+      "importFailed": "Impossible d’importer le YAML.",
+      "downloadFailed": "Impossible de télécharger le YAML.",
+      "migrationWarning": "Le YAML hérité a été normalisé. Vérifiez les décisions déduites avant l’enregistrement.",
+      "confirmInvalidation": "Je comprends que ce changement peut invalider les résultats de sélection existants.",
+      "selection": "Sélection",
+      "required": "Obligatoire",
+      "requiredField": "Sélection obligatoire",
+      "selectColumn": "Sélectionner une colonne",
+      "titleField": "Titre",
+      "abstractField": "Résumé",
+      "additionalField": "Champ supplémentaire",
+      "addAdditionalField": "Ajouter un champ supplémentaire"
     }
   },
   "search": {
@@ -220,7 +345,8 @@
     "runAllL1Desc": "Exécute le dépistage titre/résumé sur toutes les citations de la liste.",
     "runAllL2Desc": "Exécute le dépistage en texte intégral sur les citations qui ont passé L1 et qui ont un PDF téléversé.",
     "runAllExtractDesc": "Exécute l’extraction des paramètres sur les citations qui ont passé L2 et qui ont un PDF téléversé.",
-    "forceRerun": "Forcer la ré-exécution (écraser les résultats IA existants)",
+    "skipExistingAI": "Ignorer les réponses IA existantes",
+    "skipExistingHuman": "Ignorer les réponses humaines existantes",
     "runningAllAI": "Exécution de toute l’IA...",
     "confirmRunAllAI": "Exécuter l’IA pour toutes les citations de cette liste? Cela peut prendre un certain temps.",
     "runAllFailed": "Échec de l’exécution de toute l’IA",