Repository navigation
feat(inference): add reversible operation input encoding - #2314
Merged
hidb4ai merged 7 commits intoNov 1, 2026
Merged
Conversation
This was referenced Oct 30, 2026
hidb4ai
marked this pull request as ready for review
October 30, 2026 09:43
hidb4ai
requested review from
AlexStocks,
PsiACE,
frostming and
zhanghuidinah
as code owners
October 30, 2026 09:43
hireview001
reviewed
Oct 31, 2026
hireview001
left a comment
hireview001
left a comment
Collaborator
There was a problem hiding this comment.
- CI failed: tests (3.12, unit) (failure).
- CI failed: tests (3.13, unit) (failure).
- CI failed: tests (3.14, unit) (failure).
- CI failed: tests (3.11, unit) (failure).
- CI failed: Main (failure). Failed steps: tests (3.11, unit) / Run unit tests; tests (3.14, unit) / Run unit tests; tests (3.13, unit) / Run unit tests.
hidb4ai
added a commit
that referenced
this pull request
Nov 2, 2026
…2311) * feat(inference): enforce final request and response privacy checks * feat(context): add exact reuse proofs and final payload admission * test(inference): preserve privacy admission content * docs(context): define admission boundaries for transformed values * test(inference): adapt privacy recovery fixtures to memory descriptions * build(opencode): refresh combined API operation bundles * test(desktop): qualify synchronized context Server protocol * fix(inference): close worker and opaque carrier privacy gaps * fix(context): reject prepared markers split across text blocks * fix(context): enforce permit deadlines throughout request writes * fix: align main upgrade review and test fixtures * test(desktop): qualify synchronized context Server protocol * test(desktop): qualify synced context reuse server * feat(inference): add reversible operation input encoding (#2314) * feat(inference): add reversible operation input encoding * docs(inference): clarify input encoding semantics and acceptance * fix(context): defer crypto imports and check merged Anthropic turns * test: align processing diagnostics fixture with input encoding (cherry picked from commit 20b3b4c) * fix(tests): resolve SeekDB probe imports consistently (cherry picked from commit 19a1c11) * fix(inference): preserve stream identities and refused usage * test(desktop): qualify merged context receipt contract * refactor(inference): remove unused Atomic Memory encoding wrappers * fix(inference): inspect assembled Responses output text * fix(inference): validate Chat stream root model identity * docs: require coordinated context capability client upgrades * fix(inference): preserve provider retries in observation modes * fix(inference): preserve completed sibling privacy refusals * fix(server): reuse the prepared context admission limit * docs(inference): clarify enforcing response buffer limits * test(evaluation): measure Experience input encoding quality and cost (#2315) * feat(inference): add reversible operation input encoding * test(evaluation): measure Experience input encoding quality and cost * docs(inference): clarify input encoding semantics and acceptance * test(evaluation): probe model interpretation of encoded evidence * fix(evaluation): guard identities and isolate failed cases * test: align processing diagnostics fixture with input encoding * fix(evaluation): preserve judge streaming and evidence identity * fix(eval): reject missing external databases before runtime startup * fix(eval): normalize partial database configuration * fix(eval): reject unsupported private identity platforms * fix(inference): guard SDK assembly and evaluation resume identity --------- Co-authored-by: hireview001 <1441951127@qq.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Which issue or RFC does this PR close?
Implements the operation input-encoding portion of RFC #2086. Replaces the corresponding scope of #2264.
Depends on #2311 and targets its branch so this diff contains only this feature. After #2311 merges, rebase onto master and retarget before marking this PR ready; do not merge into the feature base branch.
Rationale for this change
Repeated evidence and metadata can inflate model inputs. Reversible encoding can share repeated values while preserving the exact stage JSON and its evidence identities. Selection must remain bounded, use the same representation as budgeting, and never turn a privacy rejection into an encoding fallback.
What changes are included in this PR?
Are there any user-facing changes?
Encoding defaults to off and adds no model calls, HTTP endpoints, database tables, or migrations. Active encoding is experimental: exact reconstruction and local size estimates do not prove equivalent model quality or lower provider cost. Active candidates reach the model as encoded envelopes; shared values still contain the original evidence and provide no privacy redaction. Literal placeholders are not resolved. The bilingual guide defines model/workflow acceptance for ownership, paths/IPs, types, repeated occurrences, tool intent, and persisted facts. Each operation/model combination needs its own acceptance.
How was this change tested?
make check: passed, including lint, types, locks, and integration manifest checks.AI usage statement
OpenAI Codex assisted with implementation organization, conflict resolution, regression tests, documentation, and validation.