Binding-aware cross-reactivity axes are no longer tracked by any open issue #412

The binding-aware cross-reactivity axes — self_mimic_* (nearest self peptide that also binds at least as strongly as the query at this allele), self_strongest_nearby_*, and the structured self_nearest_candidates column — have fallen out of issue tracking entirely.

Where they used to live:

• topiary/self_proteome.py:34-38 still tells the reader they are "not implemented yet ... Tracked under Ship a curated non-CTA reference proteome artifact for nearest-self #124".
• Ship a curated non-CTA reference proteome artifact for nearest-self #124's body was subsequently narrowed to the reference-data half (shipping a curated non-CTA proteome artifact), so it no longer describes them.
• Populate self_nearest_* from a caller-supplied proteome, including paired MHC binding #190, which shipped the sequence-plus-binding computation half (predict_self_nearest, paired MHC binding at the row's allele, the self_nearest DSL scope), is closed.

An issue search for self_mimic / self_strongest_nearby / self_nearest_candidates returns only #124 and #190. The design notes for them exist only in #124's 2026-04-16 comments.

Two things wanted here:

1. Track the axes under their own issue, carrying over the decisions already recorded in those comments: per-(peptide, allele) rather than per-peptide; self_mimic_* as the primary cross-reactivity signal with "none found" meaning no match within the stated reference, search and prediction coverage, never no presentable risk; binding_metric= defaulting to percentile_rank for cross-tool comparability; TCR-facing vs anchor position weighting (TCR-facing substitutions read as more distant, anchor substitutions as less, since the latter keep the same TCR-facing surface).
2. Fix the self_proteome.py docstring, which now points readers at an issue that does not cover the feature.

Found while triaging #124.

Scientific semantics correction (Vaxrank feedback, 2026-10-02): sequence similarity, same-allele predicted binding and healthy-tissue presentation are separate evidence axes, not proof of TCR recognition or absence of risk. Preserve multiple candidate matches and every origin, including CTA/non-CTA shared sequences. Exact self ligands within a vaccine window must be distinguishable from similar peptides relevant to the intended target. Reference/CTA ownership remains oncoref; Vaxrank owns window repair and required-target constraints. Primary study: https://www.nature.com/articles/s41541-023-00713-y .