[Clear1nsight]

Change and mathematical scope

Add research/h25-elf-literal-discovery-controls, an isolated archive of a completed bounded source/control component. The v2 ELF candidate-discovery helper rejects (1) a file-unique literal with a competing virtual LOAD mapping, including partial or zero-fill overlap, and (2) a non-executable entry whose file bytes are aliased by an executable LOAD at another virtual address. The exact rejected v1 source, independent M01/M02 objections, repair diff, and accepted v2 source are retained.

The completed component has 23 passing NEW synthetic cases in normal mode and 23 in optimized mode over 20 literal buffers, independently reviewed from the retained raw outputs, grant, tool receipts and ledger. This is candidate discovery and finite synthetic control evidence only. The actual-inspection round closed partial at 23:55:37 UTC on 2026-11-02 with 0/4 P1-P4 checkpoints complete. No actual GCC image pass, helper.main/decoder execution, actual 60-second grant, operational prefix certificate or feasibility acceptance occurred. The v2/v3 capture guard remains REJECT/HOLD because close can overwrite the first read/pin/write failure; inert reviews and full closure receipt preserve that blocker, with no guard executable exported.

Full C05/V03 remain false. Original activity credit added is zero: 43.5%, 22/49, full stage gate qualification 20% (2/10), overall milestone qualification 10% (1/10), and 24 components/52 helpers are unchanged. Physical costs remain OPEN_NOT_ZERO. No kappa, speedup, supplier admission, selected-frontier change, or closed P1/P2/P3/P4 is claimed.

Task and overlap

Base 7c49fe32c55de02c2c873655c804b16e4c3fef0a, tree e6898c8ba7a27b71587c32ba3f639259ceeeed91. Explicit user-directed incremental prerequisite publication; scope is changed from the pinned portable-replay catalogue card, which supplies genuine infrastructure provenance only. No picker or unrelated task was run.

Related: #6397 (criterion), #6362 (input constructors), #6434 (raw pilot error accounting), and #6454 (conditional phase frame; observed head e83090c937b920affc87593f02b535e6c9773615). This package adds the separate pure literal-discovery mapping repair and finite controls; it does not claim predecessor work. Parent's bounded related-PR graph covers 39 nodes/29 edges; historical heads and finite search limits remain explicit. No exhaustive novelty or incorporation claim follows. #6454's hosted run 41547936328 attempt 1 completed success with replay skipped, supplying no candidate/native/scientific replay credit.

The package is self-contained for its registered synthetic controls. Exact source origins, Git blob IDs, SHA256, sizes and 100644 modes are recorded in SOURCE-MAP/MANIFEST. All transitive executable inputs are package-local: the frozen v2 helper, builder and harness, plus Python standard library. Historical absolute paths in protocol and evidence are inert. A known unrelated base PDF Git blob remains missing in the local alternate and is not repaired or presented as complete object availability.

Validation

Historical accepted evidence: helper source review cdf88f0eeaff9ca22d9964a086875a43b5e87a0647de41c4b637129acb553f37; separate controls-execution review a7bb90acc80f519af72abd87993c31c0b2c3a938d0a6e29508daee7c589520d7; ledger 4955efcccf2160f87e75ab02dfca82da6ed733e2690bad8a65dfd15e4f64c7b9. The full closed-round receipt is e8fc63616c75ad6dd5b5b69620764e29f7e6040ad292c5a5924cd6027a5e2c93.

GNU wall 0.16/0.17 seconds and tool wall 0.065126715/0.071047239 remain separately retained. The declared policy charged 0.35 seconds total; supplemental combined policy time at that historical component checkpoint was 100.130533479 seconds and canonical 94.550533479 seconds is unchanged. Individual GNU RSS is not complete tree peak or a full physical-cost bill.

Reproduction is documented in the package README. VERIFY.json registers only the normal/-O publication adapter. It resolves its own directory, pins the exact 8023-byte frozen harness before compiling those SAME bytes in a fresh non-main module with absolute file, then calls only its main with the absolute sibling v2 helper path and fixed hash. The upstream runner supplies a relative package string; this transparent adapter preserves the unchanged absolute-path requirement and all three accepted source files. Unexpected wrapper arguments are refused. The harness checks source bytes before executing those same bytes in fresh non-main modules, calls only inspect on literal buffers, and keeps explicit requirements active under -O. C01/C02 require real Stop outcomes. The standalone static integrity checker reads bytes/JSON only and is not a registered execution of helper.main. It accepts exactly 0644 or read-only 0444 physical permissions with logical Git100644, rejects links/executable permissions, and checks the file-length cap before and after reading.

Parent A-I validation of preserved v3 commit 4dda053c39b11d16da387440a7c07b4450f5edbd passed, with independent execution/committed-export review d1a42da686b482a07e78231ef35cdab615a3f87a209236fcc34c38a7512617bd. Evidence includes static integrity, official metadata/normal23/optimized23, committed-byte/source-history/modes, official committed export, read-only cold static/normal23/optimized23, and expected one-byte harness tamper refusal (exit 2). Total added A-I policy debit was 3.17 seconds; supplemental combined policy time is 103.300533479, canonical 94.550533479 is unchanged, and physical costs remain OPEN_NOT_ZERO. These exact code/fixture/control outcomes are reused only because all five Python files, VERIFY and scientific/raw evidence remain byte-identical. This v4 successor changes exactly README, SCOPE, SOURCE-MAP and MANIFEST to correct qualification-category labels; the erroneous v3 snapshot and complete evidence remain preserved. The 20% 2/10 counter is FULL_STAGE_GATE qualification, and 10% 1/10 is OVERALL_MILESTONE qualification; neither is a C05 or V03 percentage. Both full flags remain false. Current v4 parent-owned metadata checks all passed: static integrity; official committed-byte/source-history/modes preflight; official export and check-inputs; exact 56-file/305,834-byte checked read-only copy; and read-only static integrity. Current tar SHA256 is 24047aa10d346db522ed89fd4801983bc473f2b068b362662ea669c65871428e. Independent v4 reporting/delta/invalidation review bc0cdeaffdc904a9c37152879d444ac5b2356b58dcb1971ebaa742f2e1fa86f0 accepts exact unchanged-code evidence reuse and rejects the old percentage labels. Metadata-only ledger 73c41786f3bd694cfdad753b3bbba51fa68a4757868be29ac0b90b57df9a7f1c adds 1.59 policy seconds, giving supplemental combined 104.890533479 seconds; canonical 94.550533479 remains unchanged. No new v4 local synthetic-control or tamper replay was performed. V4 local commit d776c09f8817cffe0dbe7b3021eda775dfa2b247, tree ac6eebceca1f56d8a177ea1fd0fe2bba02f7898c, has sole parent 7c49fe32c55de02c2c873655c804b16e4c3fef0a and exactly 56 additions. Normal git write-tree succeeded in the earlier v3 phase; --missing-ok was not used. The unrelated missing base PDF remains unrepaired. The fork publication and Chrome submission use this exact reviewed tree. Current-head hosted checks and public handoff will be recorded in the delivery comment, with any pending/skipped status preserved. No actual-image/helper.main/decoder/native/scientific/Lean/full historical suite, new budget or actual round was authorized by this reporting correction.

S04 checks unvalidated interior candidates; S06 checks nonpromotion after overwrite. S11 role/reachability, S12 relocation effects and S13 private-call/alias semantic falsifiers remain unimplemented. Finite metadata buffers are not claimed to be genuine kernel-loadable executables.

Credit

Prepared for Clear1nsight with substantial OpenAI Codex assistance, including assigned GPT-6.1 Sol Max authoring and independent AI source/evidence reviews. Those reviews are not human or proof-kernel acceptance. Predecessor credit and exact LICENSE/NOTICE snapshots are retained; the genuine base contributor index remains pinned in SOURCES.json. New material is Apache-2.0. No GCC source/binary or external decoder is vendored, and no license claim about external executables is made.

Open next questions

See TASK.json and README. The immediate blocker is first-error preservation through close/cleanup in a separately authorized source round; review the exact frozen objections before considering any genuine actual-image admission. Boundary/role/relocation/private-call qualification would then need its own concrete admission and bounded plan. The current inspection round is closed; this PR does not reopen it, authorize a repair or actual run, add funding, or resume the separate paused application goal.

Workflow feedback (optional)

Separating accepted helper/control evidence from the rejected capture guard permits useful component publication while preserving the actual round's negative result and unchanged original credit.

[Clear1nsight]

Delivery verified — PR #6476

Exact head 44749c3a6ddc8cde48c5aa83073f6794e3546b00, tree ac6eebceca1f56d8a177ea1fd0fe2bba02f7898c, sole parent 7c49fe32c55de02c2c873655c804b16e4c3fef0a. All 56 additions / 305,834 bytes match the independently reviewed transport. Draft/open/unmerged; maintainer modification disabled.

Hosted run 41553748968, attempt 1, completed SUCCESS on this exact head:

• plan — SUCCESS (job 125647453284).
• replay (research/h25-elf-literal-discovery-controls) — SUCCESS (job 125649230596), including explicit bounded package replay and unchanged committed-input check.

VERIFY registers only the normal and optimized synthetic wrapper commands. This hosted success does not execute the actual GCC helper.main/decoder/compiler query or original full-data caller and adds no original qualification credit. Local retained evidence has 23 unique synthetic cases over 20 buffers in each profile, one expected harness-tamper refusal, and read-only cold controls. The reporting-only V4 delta preserves all executable code/fixtures/VERIFY/layout; current V4 integrity, committed-source-history/modes, official export/check and read-only static checks passed. No new V4 local synthetic replay is claimed. Current metadata cross-review is cca18d7f7572828c18f707dd06e8a2a7dbac3ec896600641583f59877fa0a17e; reporting/delta review bc0cdeaffdc904a9c37152879d444ac5b2356b58dcb1971ebaa742f2e1fa86f0.

Canonical result, uncertainty, reproduction and next dependency · short hub pointer. Those handoff posts recorded CI in progress at posting; the final conclusions above now supersede only that hosted-status observation.

The actual inspection round remains closed partial 0/4, with capture guard HOLD because close can replace the primary read/pin/write failure. Full C05/full V03 remain false. Original weighted completion is unchanged at 43.5%, with 22/49 activities; 20% (2/10 full-stage gates) and 10% (1/10 overall milestones) are separate counters, neither a C05/V03 percentage. Original 24 components and 52 helper IDs remain fixed.

Canonical guarded policy remains 94.550533479/2400 seconds. Supplemental combined policy is 104.890533479 seconds after current metadata checks; no new budget. This is not a complete physical-cost bill: preparation, review, Pro, reporting, hosted execution, publication and final metadata costs remain OPEN_NOT_ZERO. No merge, automatic source retry, extension of a closed round, selected-frontier change or application-goal resume is performed. Substantial Codex/GPT-6.1 Sol Max assistance and predecessor credit are disclosed in the PR.