Drop purge_asymmetric_pollution + KNOWN_BAD_NARROWMATCH (Codex-#558 closed upstream)

[realmarcin]

Summary

MIM PRs #2, #3, #4 closed the Codex-#558 hardening pass on the upstream side. The kg-microbe-side workarounds added to compensate for upstream invariant violations are now redundant.

Removed:

• purge_asymmetric_pollution() (95 LOC) + its main() call site in scripts/consolidate_chemical_mappings.py. This swept stray child-label synonyms and MIM xrefs that earlier consolidator runs leaked from narrowMatch-MIM subjects onto OBO parents.
• Test docstring reference to purge_asymmetric_pollution updated to point at the upstream MIM PRs that now enforce the invariant (tests/test_chemical_mapping_utils.py).

Already removed (this PR confirms it stays gone):

• KNOWN_BAD_NARROWMATCH constant + filter call site — removed in commit ed421d0e ("Sync MIM SSSOM and remove redundant KNOWN_BAD_NARROWMATCH filter") on this branch. Its 5 entries no longer appear in the published MIM SSSOM (PR ingest trait table data from #2 removed them upstream).

Why these are now redundant

Workaround	Upstream resolution
purge_asymmetric_pollution	MIM PR #4 Rule B4 (canonical object_label) refuses rows where label doesn't match source ontology's canonical name. MIM PR #2 Rule A refuses auto-classifier rows with zero token overlap. Label drift that fed the pollution is gone at the source.
KNOWN_BAD_NARROWMATCH	MIM PR #2 removed all 5 rows from the published SSSOM. Defense-in-depth: claw build_mim_ingredient_sssom.py token-overlap gate (commit e00d4f5) refuses CONSIDER_SPECIFIC overrides where MIM and kg-microbe labels share zero tokens; categorize_residual_p25.py producer-side guards (commit a3e36db) prevent the residual JSON from emitting bad CONSIDER_SPECIFIC entries.

Behaviorally, KNOWN_BAD_NARROWMATCH lookup is now unconditionally false (input data is hardened upstream), and purge_asymmetric_pollution() was a no-op (no parent records to clean) under the new invariants — so the deletions are safe.

Test plan

• Consolidator runs end-to-end cleanly. Output: 595,755 lines (595,681 mappings) vs baseline 595,436 — net delta near 0 as expected. The "Purged N stray child-label synonym(s) and M stray MIM xref(s)..." log line is gone (no longer called).
• pytest tests/test_chemical_mapping_utils.py tests/test_chemical_stereochemistry.py -x: 76/76 passing. test_vermont_soil_resolves_to_child still resolves correctly to the kgmicrobe.ingredient:vermont_soil child without the purge step.
• parent_relations and _PARENT_INDEX runtime indexes preserved (still used by find_chebi_by_name / get_parents lookups and by export_unified_sssom).

Diff stats

 scripts/consolidate_chemical_mappings.py | 105 -------------------------------
 tests/test_chemical_mapping_utils.py     |   4 +-
 2 files changed, 2 insertions(+), 107 deletions(-)

Cross-repo references

• MIM PRs: ingest trait table data from #2 (Rule A — zero-token-overlap refusal), map metabolism / use of oxygen (aerobic etc) column in big trait table #3, Map cell_shape in big trait table #4 (Rule B4 — canonical object_label) — all merged on MediaIngredientMech.
• claw commits: e00d4f5 (defensive token-overlap gate in build_mim_ingredient_sssom.py), a3e36db (producer-side guards in categorize_residual_p25.py).
• kg-microbe prior commit: ed421d0e (removed KNOWN_BAD_NARROWMATCH).

[Copilot]

Pull request overview

Removes kg-microbe-side defensive cleanup logic that was previously needed to compensate for upstream MediaIngredientMech (MIM) SSSOM invariant violations, now resolved upstream (Codex-#558 follow-ups).

Changes:

• Deleted purge_asymmetric_pollution() and its main() call site from scripts/consolidate_chemical_mappings.py.
• Updated the regression-test assertion message in tests/test_chemical_mapping_utils.py to reference upstream MIM PRs instead of the removed purge step.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
scripts/consolidate_chemical_mappings.py	Removes the now-redundant asymmetric pollution purge step from the consolidator workflow.
tests/test_chemical_mapping_utils.py	Updates regression-test messaging to point at upstream hardening rather than a local cleanup routine.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[realmarcin]

Superseded — applied directly on mapping-cleanup-canonical-hub as commit 937a285 per team-lead course correction.