[ToikY]

Final hosted evidence: original run42111456479, attempt1, SUCCESS.711 repository helper tests/24platform skips and all eight actual normal/optimized package commands passed on Python3.12.16; both worktree/index drift guards passed. Synthetic checkout2a10d6b55153087319ec49c265e1bb0ab1bb68f2 has exactly published treec22d8cf56ce6a5085f39d131d1586f4ceb89da4b and canonical/public parents. Completed169438B plan and35844B replay logs retained once; no rerun. This is finite execution corroboration, not human/kernel/native or full supplier acceptance.

Change and mathematical scope

Three reporting faults in the actual canonical compile_counts pass all retained inline scalar guards while changing complete outputs: omitted local source slots (first rank121), omitted recursive child slots (first126), and zeroed instance/slot invoices (first121). At121, the local-slot fault changes127372 to119684; zeroing U/R leaves B/L/C unchanged. Four other actual controls reject with the recorded source messages.

This is finite inline-guard fault coverage. The healthy compiler is not shown incorrect. Dillon’s PR4521 already independently corroborates all five expanded totals, rawp7 through1000, toys100,44 literal programs and14 controls. Its complete saved-receipt comparison would catch altered totals; this does not bypass that verifier. Healthy counts here are calibration, not a new count theorem.

Task and overlap

Canonical main 7c49fe32c55de02c2c873655c804b16e4c3fef0a, revision8/selected6030. Actual source verify.py blob df9ca12ad0dae5437ed7d65688e962b9abda5af4; six exact inert source/data/license objects132276B retained. Scope: default nested policy, rawp7 through606 and three fixed toys through50/16/20, every rank0..N of all five arrays and every returned summary field.

Formal-interface-lemma card scope CHANGED to finite actual-source fault coverage; original formal/kernel/all-N gate unmet. Initial expanded-count gap claim was wrong: historical PR4521 was found after the sole primary first. The claim was corrected in place, v1 drafts/firsts preserved, and work narrowed after main refresh and separate scope review. All three presealed programs/cases/scientific inputs stayed unchanged. Canonical claim/correction; see CORRECTIONS.md.

Live694 open bodies/heads refreshed before submission; only unrelated H25 PR6744 changed. Visible Hub/active Discussions, targeted all-state queries and local overlap helper checked. Sparse8881 omitted note paths, middle/hidden Discussions, arbitrary closed diffs and unpublished work remain coverage limits; no global novelty claim. PR6742 Flow outputs, PR6743 written default-width law and PR4519 chronological-width recount retain separate credit.

Validation

Two separately authored expected methods (integer copies versus normalized Fractions) and actual pinned source execution agree on the entire1930712B packet, SHA256 da8b0953c9a94165de7e25fb00b793acc86c36a0eb73e0afb37146c65275aeb3:3480 complete vector entries plus four summaries. Source-aware historical flow-method reuse is disclosed; no fresh blind/end-to-end independence. Actual seven-control packet3058B SHA256 dac54fc24ed2f0ca9edacb4bac90d9268be8655c331e4f0e752fee3d5c174354.

• Eight registered warning-strict normal/optimized whole-packet commands passed locally on Python3.12.14, stdlib, serial/one thread.
• 72 actual CLI rejections and8 healthy-reference positives passed. An initial external diagnostic-harness mismatch is preserved; verifier code/firsts were unchanged.
• Exact committed bytes/source history and final export/all11 declared inputs pass. All eight scientific commands passed from a read-only, Git-free, empty-PATH export; only README/TASK/VALIDATION status metadata changed afterward. Scientific freeze pins are unchanged.
• Sealed firsts took0.319/0.316/0.261s below38MB RSS; actual controls0.751s/52.6MB. Separate source-aware code/scope/byte reviews are retained. Original hosted CI42111456479 passed as detailed above; no rerun.

Reproduce: python3.12 -B scripts/research_package.py --package research/direct-compiler-counts-audit. Public head 50027f6de434a8d936871a1b3f776f04d47cd3e2, tree c22d8cf56ce6a5085f39d131d1586f4ceb89da4b,31files, single canonical parent. Start20:38:32UTC; science freeze21:05:20.015014UTC; verification exceeded15min after that freeze before21:38:32cap. Exact finish in durable handoff.

Credit

Dillon/dillona4521; zxv4436 canonical construction/compiler/checks; Jacob Alan Sussman lockstep/Pascal; Cinna Davis/rawp7 and retained native/compiler lineage; RyansonJonathan4456/4476/4519 and Mihir’s distinct chronological policies. PR6742/6743 are not recounted. ToikY with substantial OpenAI Codex assistance, separate source-aware AI authors/review; Apache2 and original LICENSE/NOTICE retained. No independent human or proof-kernel acceptance.

Open next questions

Bind one actual chronological native producer/consumer and its complete productive/paid/ordinary/owned invoice; missing physical contracts are the blocker. More count samples do not close it. Native all-columns, owned erasure, compiler, precision/tape, analytic recovery, outer/full-supplier admission and all five selected assumptions remain. No κ/frontier/adoption change, merge or follow-on reservation. Maintainer edits and auto-merge disabled.