Quarantine eight false MediaDive material aliases with complete audit evidence

[realmarcin]

Scope

Closes #1262. Closes #1263. References #650 and #286; neither umbrella issue is
closed by this change.

Withhold eight evidence-reviewed MediaDive material-name/ChEBI pairs whose
native chemical identities do not describe the source materials:

Source material	Withheld target
SL10	CHEBI:1387
Trace vitamins	CHEBI:1784
Mueller-Hinton broth	CHEBI:748
Difco Marine Broth 2216	CHEBI:2216
Malted wheat meal	CHEBI:1881
K2HSO4	CHEBI:1895
Leibovitz's L-15 medium	CHEBI:1941
Selenite-tungstate solution	CHEBI:88

The immutable source/native/mapping fixture records 289 affected source
observations and thirteen complete erroneous lexical alias rows. This includes
256 nested solution references across seven distinct solution IDs; those IDs,
recipes, quantities and contexts must remain distinct. No replacement identity
is inferred, including for the unresolved source spelling K2HSO4.

Changes and preservation

• Finite name/target exclusions, not global ChEBI bans or unconditional
  all-target material holds. Independently selected alternatives remain visible
  and require appropriate scientific review.
• A required reviewed catalogue preserves all thirteen original mapping rows,
  ordinals and baseline provenance after derived mapping removal. Historical
  claims are audit evidence, never active identity mappings.
• Extend the existing producer-time material audit and finalization/input
  binding, including empty cohorts, full source occurrence/context, current
  grounding candidates and actual retained targets.
• Address Align reviewed material audit coverage with policy spelling and reader target normalization #1263's audit-coverage gaps using the selected finite policy patterns
  and actual reader-compatible target whitespace handling. Eligible current
  rows retain full fields and duplicate row multiplicity.
• Preserve the supported MIM release and pin, valid native names/direct IDs,
  chelator-role, kinetin and TMAC-chloride controls, and unrelated material
  policies. Do not reinterpret weak relations as chemical identity.

Science and reproducible native/source witnesses are documented in
docs/reviews/mediadive-material-aliases-1262.md and
tests/resources/mediadive/material_aliases_1262.json.

Verification checkpoint

• 592 affected tests passed. Independent producer/audit review passed 64 cases:
  24 independent adversarial probes plus 40 authored tests.
• Independent full-stream candidate delta passed: exactly thirteen removed
  rows, zero relabels, 591,933 retained rows with all remaining row bytes,
  ordering and multiplicity unchanged. Metadata changes only the reviewed
  policy fingerprint; original writer provenance is retained.
• Actual cold-reader/second-cycle acceptance passed: all twelve reader indices
  compared, unrelated entries and native controls preserved, no new alternative,
  and a byte-identical second build. The first attempt was retained as rejected
  evidence after a planning-document addition changed directory membership;
  the fresh second attempt reran all phases and passed unchanged-input guards.
• The exact candidate is committed with unchanged supported MIM/pin bytes.
  Post-installation native-pair/policy/audit verification: 193 tests passed.
• Independent review cleared the source/policy/audit implementation and paired
  promotion. The finite full-source comparison helper passed 83 hermetic cases,
  including 17 independent adversarial probes; this is not a source replay.
• Pending before merge: isolated MediaDive replay with full-row/audit comparison
  and manual node review; all four repository gates; exact-head CI.

This checkpoint is not production rebuild, merged-graph acceptance, publication
or release approval. The all-source rebuild and broader #650/#286 acceptance
remain separate work.

[realmarcin]

Final review checkpoint for exact PR head 1805eb8547b74694bf38717f3824ba7549ccd924:

• Independent adversarial review of the producer, policy, audit coverage, paired mapping promotion and exact PR diff completed. The eight contradicted material assignments are withheld without inventing replacement chemical identities.
• All four repository gates passed: full pytest (5,414 passed), tox (all six environments; test environment 5,415 passed), lock validation and generated merge configuration validation. The original runner's post-test BrokenPipe receipt is preserved; the two unstarted checks have a separate successful continuation receipt.
• Exact-head CI run 36598165767 passed all three Python jobs.
• The actual isolated MediaDive replay completed with settled processes, unchanged pinned inputs, source admission and fingerprint verification: 32,133 nodes / 113,918 edges. Original replay receipt SHA-256: 6b440601f15a41fc7257189f3bbef41a1166af486b6599f051e44ff61de4f745.
• Full-row comparison passed with original request f6b583a7... and independently reviewed corrected comparator 27e31e41...; result SHA-256 3215b5c91d06b429fc935ea529b84351489a05a9cbc675b64ea6226324e0c523. Exactly 289 recipe assertions change only their target, eight now-unneeded native-role projections are removed, and every unrelated assertion remains unchanged. All seven nested solution identities and valid controls are preserved.
• Audit preservation is complete: 225 previous rows plus 354 historical mapping claims. The first rejected comparison and original helper remain preserved under Bind native dependencies correctly in the MediaDive source acceptance comparator #1266; no producer rerun or source-output editing was used to obtain the corrected comparison.
• Independent manual node review confirms eight new local ingredient declarations, ten zero-incidence source-node removals, no changed pre-existing nodes, and no unsupported native metadata on the new local nodes. Native ontology assertions remain intact.

The #1266 correction passed 135 tests (96 authored, 22 new independent, 17 inherited independent). This closes the bounded source acceptance for #1262/#1263; it is not release acceptance. #650 and #286 remain open pending the fresh all-15 rebuild, candidate-only merge, model/path/content review and evidence packaging. #1265's additional source-to-merge preservation check must run against that candidate.

Local evidence: data/issue1224-quarantine-20260929.xjuS9H/issue1259-diagnosis.7T9Jn8/ (original replay, source1262-actual-comparison-02/, independent reviews and repository gate receipts).