RT 2026-07-24 - area 5 - opus (Opus 4.8) - yield 0

[ms609-agent]

opus (Opus 4.8) | effort: unrecorded | 2026-07-24

Migrated record. This round predates the move to Discussions; it is reproduced
verbatim below from dev/red-team/log.md, which is now closed to new entries. It is
posted here so this area has a round record whose createdAt orders correctly against
the other areas' — the backfill ran oldest-first for that reason. All fourteen posting
timestamps are later than the rounds they describe; the real date is in the title and
in the date: field.

⚠ Scoped round. This was one of the 2026-07-24 directed passes over what was new in
MaximizeParsimony since the previous inspection — not the area's full file list. The
seam verdict it records is about that surface. Read the verdict in the notes rather
than inferring one from the yield in the title: a low number here can mean the new
code was clean, which says nothing about the parts of the area it never opened.

---

area: 5 (Data pipeline & simplification — DIRECTED round: "what's new in MaximizeParsimony since last inspection")
reviewed_by: opus finder (a0586c26) + haiku verifier (a4bf9a63); orchestrator advisor-briefed; pinned @ 1863b2b
date: 2026-07-24
tier: opus (Opus 4.8)
yield: 0 new rows — 1 finding (PACK-COLRED-01, verified REAL-but-inert by haiku) FOLDED into T-335 as a sibling defect (same reduce_sector_columns_ew, same EW-only gate); no standalone row
notes: First area-5 review since the 2026-06-15 CRAN sweep (DAT-001/DAT-002/LS-RSS=0, all fixed PR ms609#252); last rotation round 2026-05-26 (dry). Fresh-surface opus pass on the post-sweep commits: per-block-local-alphabet packing (89ebd870, default-ON), char homoplasy-ordering (d8c59998), ambiguity fixes (53ffb05b {-+}, 942bfe5c multistate overcount). ts_ls.cpp correctly EXCLUDED as fresh surface — it is a behaviour-preserving extraction (created wholesale at 78b74147, 2026-06-19; same extraction that split ts_ratchet.cpp) of LS code already inspected 2026-06-15; divergence-check only (RSS clamp-at-0 + singular→+Inf guards present). Advisor's THIRD severity-prior framing (the one that mattered): data-pipeline corruption is UPSTREAM of the full-rescore firewall — the firewall runs score_tree(tree, ds) against the SAME ds, so a corrupted ds (misaligned min_steps, wrong per-block alphabet, dropped ambiguity state) makes EVERY rescore silently wrong on EVERY path → P1-capable (T-330-class), NOT reach and NOT crash. That P1 hypothesis was tested directly and did NOT reproduce. Prime targets CLEARED by BOTH line-by-line trace AND a from-current-tip empirical on/off differential: TS_PACK_LOCAL default-ON is byte-identical to OFF (score + ntree + candidates_evaluated) across NA/EW/IW/profile/XPIWE on hand-crafted degenerate shapes + full MaximizeParsimony on Sansom2010 (exercises ratchet+sector+incremental-NA). char-reorder (d8c59998) is index-based — parallel arrays (min_steps/pattern_freq/precomputed_steps/info_amounts/eff_k/phi) stay in original-pattern-index space and every scorer dereferences through pattern_index[c], so the T-335 "index rewritten / arrays stale" shape does NOT apply. Packing "?"→alpha narrowing + per-block relabel is Fitch-exact (no set bit dropped); new n_states==1 all-inapp block scores 0 correctly through NA three-pass; per-block DAT-001 analog RULED OUT (no per-block 1u<<32; per-block n_states ≤ 32; global guard intact at ts_data.cpp:31); DAT-002 XPIWE obs==0 guard intact (ts_data.cpp:489); SIMD odd/small-n_states tail-scalar path no over-read. PACK-COLRED-01 (folded into T-335): reduce_sector_columns_ew (ts_sector.cpp:362) leaves the new plane_state field zero-init (CharBlock cb{}) — same recurring "reduced-dataset builder drops the NEW field" class as T-335's stale n_patterns/min_steps/etc.; inert (opt-in TS_SECT_COLREDUCE default-OFF + EW-gated; sole consumer ts_na_debug_char builds a fresh full dataset, never a reduced one). build_reduced_dataset/_collapsed carry plane_state correctly via full struct-copy. Benign (not a bug): packing shrinks blk.n_states to the local alphabet, so wagner_entropy/goloboff_scores tip-addition-order HEURISTIC is now packing-dependent — reach-neutral (Wagner trees are then scored exactly; covered by reach corpus). Seam status: RAN DRY at opus-4.8 (evidenced — empirical byte-identical + full trace, not assumed) → (as recorded 2026-07-24: "per doctrine next area-5 visit escalates to FABLE") — RE-TARGETED 2026-07-27 by the opus version bump: the next area-5 visit is opus (Opus 5) with a fresh-angle brief, NOT fable (version bump before rung bump; fable is the step after Opus 5 also runs dry). The residual below is unchanged, only the rung aimed at it. The residual is THIN. Residual = the ONE untested corner: does TS_PACK_LOCAL interact with hierarchy_blocks/Sankoff (HSJ/XFORM) tip data, which the Rcpp bridge sets AFTER build_dataset (out of this pass's trace depth; area-3 round showed rss/xss SKIP HSJ/XFORM so likely moot, but the packed tip_states layout vs HSJ tip data was not directly verified). NEXT ROTATION AREA = 6 (R↔C++ interface).