[MattGyverLee]

Decision

Option 1 (honour the picks), done on the live plan path rather than the unused categories_affixes stub. The wizard already folds Stem picks into leaf_item_picks[STEMS], which stems_enumerate_source filters on. The gap was Selection.stem_picks / affix_picks, which nothing read, so wizard affix picks and any direct Selection(stem_picks=...) still took every entry. Preview also said nothing about picks.

Changes

• Selection.leaf_picks_for(STEMS/AFFIXES) falls back to a non-empty stem_picks / affix_picks (lower-cased). An explicit leaf_item_picks entry still wins. Empty picks mean transfer all. Closure dependencies are still pulled by the existing closure walk.
• preview.lexicon_pick_notes, added to RunPlan.category_notes: "no stem picks set -- every stem entry ... is planned" vs "only the N picked stem entries planned (plus the dependencies they need)", for stems and affixes.
• The xfail test_stem_picks_flow_into_compute_plan_owned_children is now a passing test, and the stale docstring in categories_affixes.py is updated.
• Tests: tests/unit/test_117_honour_picks.py (new). test_issue84_lexicon_expansion.py was adjusted for the extra notes. The unit suite passes (5715 passed).

Live evidence (FLEXLIBS_REQUIRE_LIVE=1, Target restored from the Target 2026-07-06 backup, lock "Target-live-flex" held)

tests/integration/test_issue117_honour_picks_live.py (3 passed). The source is the private GT-Src-117, with 2 stems and 2 suffixes authored.

• STEMS run with stem_picks={A}: Target gained only stem A. Stem B and both affixes were not transferred.
• AFFIXES run with affix_picks={a}: Target gained only affix a. Affix b was not transferred.
• Preview notes show "only the 1 picked stem entry" with picks and "no stem picks set" without.

closes #117

🤖 Generated with Claude Code