Skip to content

Repair bounded G12 custody cleanup and publication-size checks - #6346

Draft
Clear1nsight wants to merge 1 commit into
CrocSwap:mainfrom
Clear1nsight:codex/p07-g12-custody-repairs-20261102
Draft

Clear1nsight wants to merge 1 commit into
CrocSwap:mainfrom
Clear1nsight:codex/p07-g12-custody-repairs-20261102

Conversation

@Clear1nsight

Copy link
Copy Markdown
Contributor

Change and mathematical scope

Adds a fixed-source successor for the shared P07/N22 G12 custody path. Seven raw-error ledgers move before local acquisition, retained pidfds receive an independent final release attempt after failed process observation, and the consumer requires exact integer publication size.

The package implements the three narrow repairs from a source-pinned ChatGPT Pro consultation and adds 41 synthetic source-function/window cases, including three retained allocation-failure counterexamples, plus 94 separate byte/AST/diff checks in normal and optimized Python. Selected AST-extracted helpers execute against descriptor/process tokens; the full target modules and original H25 caller are not imported or run.

Partial source repair only. The preflight ledger/snapshot/error-recording counterexamples, acquisition-to-registration gaps, full alias/schema review and actual source/mount/pin derivation remain open. No integrated source acceptance, N22/P07 body credit, adoption, complete caller qualification, gain, kappa or frontier change is claimed. Launch/handoff remain None. Original C01-C15 costs and unknown final tails remain OPEN_NOT_ZERO.

Task and overlap

Explicit user assignment: use the completed P07/G12 consultation package to repair the roadblock and submit scoped upstream work.

The detailed AI source review covers all18 negative-branch categories and CP01-CP15 with conditional arguments, counterexamples and remaining gaps distinguished. Category coverage is not acceptance of each predicate.

Validation

Executed locally on Python3.12.3:

python3 scripts/research_package.py --package research/h25-g12-custody-repairs --check-only
python3 scripts/research_package.py --package research/h25-g12-custody-repairs
python3 scripts/check_research_commit.py --package research/h25-g12-custody-repairs

Metadata/pins and the registered normal/optimized replay pass. Each mode reports source_cases=41 metadata_checks=94 unresolved_preflight_allocation=3 no_integrated_acceptance. The committed-byte preflight checked98 package/input files on the local reviewed commit; GitHub readback confirms the public tree equals that exact reviewed tree, with one parent commit and exactly14 added files. The complete146-line diff reproduces both corrected source files byte-for-byte in memory.

Controls exercise predecessor/new allocation boundaries, ordinary raw error ordering, persistent-denial P09 cleanup, ambiguous-close no-retry behavior, exact schema domains and the retained stronger-contract failures. No full target import, original H25/scientific/native invocation, provider acquisition, historical budget use or performance experiment occurred. The author's prior verifier was not replayed. Hosted checks are pending at submission.

Credit

Chris R / Clear1nsight, with substantial OpenAI Codex implementation/verification assistance and ChatGPT Pro consultation assistance. Credit unpaidmoon with substantial Cursor Grok assistance for #6339, #6321 source/review contributors, #6029 and retained Douglas R. Colkitt circuit/network/native/verification lineage. Predecessor bytes and notices are unchanged; Apache-2.0.

The distinct consultation-agent review is preserved as bounded draft evidence. Neither AI agreement nor these finite controls establish independent human or proof-kernel acceptance.

Open next questions

See research/h25-g12-custody-repairs/TASK.json and README.

  1. Establish whether exception-safe acquisition/registration and raw error retention can meet the ORIGINAL allocation-failure contract; prove the interpreter/ownership boundary before building a larger journal. Current source counterexamples remain, and any extra guarantee must be explicit.
  2. Freeze and independently review the complete corrected parent/consumer/authority/driver mount/pin derivation for N22 and P07. Actual custody/admission is a later separate gate.
  3. Under separately authorized runtime conditions, assess the full caller, original controls, resources and complete costs. This PR does not transfer the historical allowance or require a new generic accounting API.

Retain exact PR6339 predecessor bytes, implement the three narrow consultation repairs, and add normal/optimized source-function controls with explicit unresolved allocation counterexamples.

No integrated source acceptance, H25 execution, adoption, performance or kappa claim.
@Clear1nsight

Copy link
Copy Markdown
Contributor Author

Publication verification at abff67b31fa781e950c82d95eccb6f168950943c: hosted run 41363888489 passed both planning and the actual registered bounded package replay. Neither job was skipped. The remote tree ddfa9b7c84587a3a82f88692762a0ed887af7c4a and all14 intended added file blobs match the reviewed tree. The PR remains draft, with maintainer modification disabled.

Canonical P07/N22 handoff and remaining gate; short hub pointer. Companion #6347 records the distinct conditional interpreter-delivery finding; its final model label separates Python return from subsequent registration.

These checks support the three local repairs and preserved failure controls. They do not establish full original source/caller acceptance, actual H25 execution, helper credit, adoption, performance gain or kappa. The complete original allocation/ownership/raw-error obligations remain open.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant