Skip to content

Resolve register binary handoff: exact R1 export in Git, Python 3.11 replay verified - #29

Merged
d6g8k5htny-coder merged 2 commits into
chatgpt/drive-github-hardening-20260919from
chatgpt/register-export-r1-20260923
Sep 23, 2026
Merged

d6g8k5htny-coder merged 2 commits into
chatgpt/drive-github-hardening-20260919from
chatgpt/register-export-r1-20260923

Conversation

@d6g8k5htny-coder

Copy link
Copy Markdown
Owner

Binary transport blocker resolved

The actual 1,979,318-byte Sept23 R1 workbook is now in registers/source/GP-REG-032_v1.2_export_2026-09-23_R1.xlsx, not just linked from Drive. SHA256 6cec54cefa3ee8f9885dcfd763407a2e0274d2887d73bdc23ccc7793cd721469; Git blob e11a78762c7b3be010dddb864c1c9357e387b09e. Provenance appended to SOURCES.json, preserving prior entries and current selector.

The narrow transport workaround succeeded in GitHub Actions run35918474959/job107376036654 on Python3.11.16. It verified exact raw reconstruction, ZIP CRC, 153 members, repeated 44-tab import ->88 identical outputs, all three importer rejection controls, and both source/payload mutation controls. Downloaded Actions artifact10776430936 contains bytes identical to the separately downloaded Drive R1. No Drive credentials or temporary bearer URLs were published. The one-off transport branch must not be merged into research/main.

What remains

This is a three-file source-custody delivery. The default importer still selects Sept18, and canonical JSON/CSV are unchanged. The next integration agent can now access the binary entirely through Git and complete a coherent selector/generated-register migration, keeping the historical Sept17-to-Sept18 diff pinned to original sources. After that, PR21's held H3 review can be checked against its route key. No canonical migration or review filing is claimed here.

R1 contains40 Review Queue rows and exactly one each of the H3-solver, withdrawal and rollout-R2 keys. It is a dated snapshot, not live Drive state; later Work Events are not inside it.

Validation and scope

Successful transport run is not a full repository regression of this PR. Repository CI runs separately. No claim grade, proof, independence credit, old export byte or scientific status changes. No production code or new workflow is added to the research branch. Standing owner preapproval is effective; no Dylan approval needed.

Coordination: PR21 comments5802130895 and5801696259; PR3 comment5799103954 originally described the blocker. Authored by OpenAI/ChatGPT.

Resolves the binary-transfer blocker for the register review handoff. The 1,979,318-byte successor matches its Drive SHA256 and Git blob exactly; unchanged importer replay and five rejection controls passed in GitHub Python 3.11.16. Preserve both prior exports and the default selector. Canonical generated-register migration and H3 review filing remain separate. Three files only; no scientific status or independence credit changes.

Author contribution: OpenAI / ChatGPT.
@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

Copy link
Copy Markdown
Owner Author

Coordination: original f716 CI35918787693 is now SUCCESS, but another base merge advanced this PR to 6628463. Please avoid further automatic base-refresh commits while this exact head finishes: they repeatedly restart expensive validation despite no source conflict. #27 is already integrated. I am building nonactivating manifest selection and a source/consumer preflight on the same source bytes; R1 activation stays separate because the preview exposes three unrecognized review statuses and eleven frozen-source compatibility issues. Those source facts will not be relabeled or allowlisted to make a run green.

@d6g8k5htny-coder
d6g8k5htny-coder merged commit 8510874 into chatgpt/drive-github-hardening-20260919 Sep 23, 2026
4 checks passed
d6g8k5htny-coder pushed a commit that referenced this pull request Sep 23, 2026
…asures

PR #29's rewrite of the R1 export record replaced the blocker list with a
shorter next_step and dropped truncation_evidence. That last one is a real loss:
an earlier automatic mount of the same Drive object was truncated at 1,048,576
bytes and failed ZIP integrity, and a rejected artifact leaves no other trace
anywhere in the repository. This restores it rather than reverting their record,
which is otherwise more accurate than mine was.

Two fields are added alongside it. measured_refusals names the four checkers
that refuse this export and why, as tools/register_migration_preflight.py
reports them rather than as I previously guessed: my old note said adoption
"changes 16 tabs", which was the generated-file count -- it is 8 tabs and so 16
files. It also records two frozen_objects rows carrying a description where
Binding Class expects a class, which nothing had noted before.

inventory_note settles the nine frozen_check misses. I had listed refreshing
drive/inventory.jsonl as work I would do; it is not doable and not a staleness
problem. The inventory derives from drive/source_map/Files.csv, and the
committed copy is byte-identical to the live Drive object
1gJwebwkKnRcb3Ez8atJ0WTk9f7HmP1y3 (2,503,504 bytes, sha256 4bad6b97...). The
accessibility publication has not been re-run since 2026-09-17 and those nine
objects were frozen after it, so no faithful inventory can contain them until
that run is repeated. Regenerating one from a live Drive listing would
manufacture an export with no publication behind it.

What this does not establish: these are notes about an export that is retained
and is still not the import source. Nothing here adopts it, transcribes a
status, repairs a source cell, moves a gate, discharges a premise or earns
independence credit. The review_queue and Binding Class cells are source defects
and their repair belongs at source under CLAUDE.md rule 7.

Verified: registers_import --check, consumers_check and disclosure_check green;
183 tests pass across registers, provenance, consumers, disclosure, workflows
and run_checks.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fij9F1cYrAUDW7PhWQze1G
d6g8k5htny-coder added a commit that referenced this pull request Sep 23, 2026
Name inventable probe receipts as instrumentation only, and stub the
register selector migration until PR #29 lands. Pin only the RN note digest.

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
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