T003 FINAL HANDOFF — 2026-10-26
All assigned audits are complete. No computation or download is running.
No further research was started after the parent's47% wind-down message.

Final independently checked conclusions
R017 / E019: M002 full83cycle excluded under27/27/1 action hypotheses:
  E4[70:97]=W4[73:100]=E5[71:98], W1[38]=W1[68].
All35 initial endpoint candidates independently fail with fresh domains.

R015 / E017: top82cycle plus one fixed position independently excluded.
R016 / E018: both top81 shapes and all three top80 shapes excluded.
R018 / E020: all five top79 and all seven top78 shapes excluded;39852
endpoint rows independently reproduced;115 remaining pairs fail exact
necessary trajectory/branch tests. No partials or timeouts in final audit.
The mixed-cycle batches use the full corrected29/29/17 assumptions:
  E4[69:98]=W4[72:101]=E5[70:99], W1[35:52]=W1[65:82].
Together the reviewed batches cover all19 base conjugacy types with top
orbit78..83,63099 pair-shape cases. Do NOT extend the weaker27/27/1
assumptions to mixed-cycle shapes without a separate test.

R019 / E021: final portable verifier matches all10 reviewed source functions
by AST, with only function renames. Compressed27/27/1 specs verified. All
manifest hashes and stdlib-only imports pass. Isolated600 transition and24
planted controls pass. Corrupted data rejects early. Python-O rejects
explicitly. Producer final fullrun39.97s inspected; not needlessly repeated.

Important surviving model
R011 / E012 independently replays the complete1036-observation identity-base
M001 witness using82 ordinary slots and independent initial decks. It meets
only the corrected primary29-action equality, not every secondary alignment.
R013 / E015 validates a narrow fixed-top constraint (cycle not dividing60,
length>=7,at least4 selected slots) but no fixed-top model exclusion.

Provenance and prior art
E013 independently reparses all1036 symbols from both archived representations;
they may share an upstream transcription. Exact local commits and bounded
search scope are saved. One-exception sparse decks are an existing family.
The exact corrected exclusion was not found in inspected snapshots, which
is not a first-discovery claim. The incoming-state correspondence invariant
and observation/action indexing distinction are established GAK theory.

Corrections remain essential
R007/E008 and claimC003 explain why initial30/18-action assumptions were too
strong. Our strict-window sensitivity claimsC001/C002 remain superseded in
interpretation. Their arithmetic was conditional, not proof against actual
Noita plaintext. No old research evidence was overwritten.

Optional unaudited older material
The previous-sprint late phase-shape L<=18 addendum did not receive a fresh
T003 independent audit in this phase. Do not promote it merely because newer
different results passed. Earlier audited conservative bounds are separate.
