[iskandr]

Summary

• Use the single-word categories Reproductive, Disposable, Manageable, Protected and Vital, preserving machine IDs and anatomical memberships.
• Require positive Reproductive evidence. Disposable expression is exempt but cannot qualify a candidate or enter the reproductive RNA reference.
• Add tissue-by-tissue selectivity Rmax / (Rmax + tissue RNA), where Rmax is the highest valid raw Reproductive nTPM for the same gene/source/version/species.
• Apply both the absolute ceiling and selectivity minimum using the same local complete negative IHC evidence. Positive IHC, including Low, remains an independent veto.
• Publish reproducible candidate decisions, per-observation RNA/IHC/selectivity audits, comparison figure and human/dog/mouse mapping inventories. Preserve historical reports and the frozen public CTA panels.
• Release package version 1.11.0; data release version is unchanged.

Approved grid

Each entry is maximum nTPM / minimum selectivity.

IHC	Vital	Protected	Manageable
Missing/uncertain/incomplete	1 / 99%	2 / 97%	5 / 95%
Approved negative	2 / 98%	4 / 95%	10 / 90%
Supported negative	3 / 97%	6 / 90%	15 / 80%
Enhanced negative	6 / 95%	12 / 80%	30 / 60%

The thresholds are inclusive research parameters, not validated clinical safety limits. Zero/missing reproductive RNA gives unresolved selectivity. IHC relaxation requires all routine contexts in the RNA organ, using the weakest reliability; narrower regions and other organs cannot clear it.

Results on 2,537 published candidates

Scenario	Retained
v1.10.0 adaptive absolute screen	624
Require reproductive evidence	619
Combined fixed baseline	341
Combined IHC-adaptive gates	377

Five losses are due to reproductive admission (IL17F, KRT82, LCE3C, TCHHL1, TGM7); another 242 fail selectivity. PRAME, CTAG1B/NY-ESO-1, MAGEA3, MAGEA4 and SSX2 remain. Every exclusion is recorded with its tissue, threshold and IHC evidence. Missing IHC earns no relaxation; incomplete routine RNA and existing family/manual-review gates still block retention. Human HPA v23 limits are not transferred numerically to dogs or mice.

Review and verification

• 143 focused tests passed, including inclusive boundaries, Disposable-only evidence, reference selection, missing/zero/invalid RNA, empty inputs and local IHC restrictions.
• Ruff passed; strict MkDocs build passed; comparison figure visually checked.
• All report input, implementation and output SHA256 checksums verified.
• Wheel and source distribution built, checked with Twine and compared against the working sources.
• Full CI matrix required before merge.