Skip to content

Reconcile the V02 imported-decoder receipts on digest 7001e2ba - #284

Draft
unpaidmoon wants to merge 1 commit into
cursor/pinned-upstream-e545235-6cf3from
cursor/v02-parent-decoder-receipt-6cf3
Draft

unpaidmoon wants to merge 1 commit into
cursor/pinned-upstream-e545235-6cf3from
cursor/v02-parent-decoder-receipt-6cf3

Conversation

@unpaidmoon

Copy link
Copy Markdown
Owner

Change and mathematical scope

Static reconciliation of two published receipts for one V02 child. On child SHA-256 7001e2ba395ebb7e943816301be7830e6a9bb699acc182e2bbeb786148459048, an identity replacement of a snapshotted json.loads binding during the first fd0 read raises before _document and cannot assign phase = 'SEALED'. The final ingress case imported_decoder_alias_after_read records that refusal and one modeled read. The intermediate receipt's three foreign callbacks and SEALED are not an execution of that identity replacement on these bytes.

This is a scoped source argument plus a receipt comparison. It is not a harness replay, an exhaustive mutation census, a parent-origin join, or a κ change. Selected κ stays 363005817/100000000000.

The success path after source_recheck(True) calls _document twice in _receive_v02_preseal and once in _construct_v02_issued_input. Three is that later count. Imported function descriptors are checked only when include_import_functions is true, which is the post-frame source_recheck(True), matching the ingress split of 1 modeled read for an alias change and 1111 for an imported code change. The package records 1111 from the receipt and does not derive it as a frame length.

Remaining obligations: the ingress harness named in REVIEW-final.json is unpublished, and the prebirth-to-child FIFO join is still unspecified. Original omission cases remain 0/8.

Task and overlap

Base commit: e54523517d0f03e58352e647c403a5af3e525a57. Audited head: PR CrocSwap#7137 b5587eef1dd66d477a2e4d7b6f83616e07a01386, which is not an ancestor of this base. The four inputs are vendored under research/v02-imported-decoder-receipt-split/inputs/ because that commit is not in this repository's history.

Task card: portable-replay, scope refined. The card allows one canonical comparison audited for an omitted proof-relevant field. The omitted field is the mutation class (binding identity versus a change the identity check does not see). This is not a gzip-container repair.

Handoff: Discussion CrocSwap#1166 comment 19530677. Nearest open PR is CrocSwap#7137 itself, which exports the conflict and does not select a finding. A direct open-PR search for "imported decoder" and "parent custody" returned only CrocSwap#7137. scripts/find_overlap.py --offline had no usable remote snapshot; its only hits were this package. No fork PR already covers the split.

Fork main (f9701fe9b662162ab4048be76b19053a6acfc52a) is not an ancestor of the pinned upstream commit, so this draft's base is cursor/pinned-upstream-e545235-6cf3 at that commit. The scientific base remains CrocSwap main at e54523517d0f03e58352e647c403a5af3e525a57.

Validation

CPython 3.12.3, standard library only. The child is parsed with ast and is not imported or executed.

python3 -B research/v02-imported-decoder-receipt-split/verify.py
python3 -O -B research/v02-imported-decoder-receipt-split/verify.py
python3 -B scripts/research_package.py --package research/v02-imported-decoder-receipt-split --check-only
python3 -B scripts/research_package.py --package research/v02-imported-decoder-receipt-split
python3 -B scripts/check_research_commit.py --package research/v02-imported-decoder-receipt-split --check-source-history

All five commands passed. The verifier reports identity_replacement_during_first_read_can_seal=false, five mutation controls, and upstream_blobs_absent=4 because commit b5587eef is not in this clone. python3 -B scripts/research_selected_baseline.py printed PASS_RESEARCH_BASELINE_BINDINGS_ONLY with κ 363005817/100000000000. make frontier7000-verify was not run.

The certificate does not cover replacing the module __dict__ object, a mutation after the pre-decode source_recheck(True), or an arbitrary interpreter mutation.

Credit

The child and receipts are Clear1nsight / Chris R, PR CrocSwap#7137, building on PRs CrocSwap#6867, CrocSwap#6873, CrocSwap#6885 and CrocSwap#6910. That archive's NOTICE records OpenAI assistance. Prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0.

Open next questions

  1. Publish the harness that produced three callbacks and SEALED, the object it replaced, and the child digest it actually ran. test_ingress_independent.py is named at SHA-256 52d6b54ee6c465a00d6bc4d0a5ccf4271376651a061246cf293d7a5dce0a429b (4450 bytes) and was absent from the review directory of b5587eef.
  2. Specify the FIFO join from parent prebirth to prepare_selected_child / install_before_capture / _receive_v02_preseal only after an exported observation shows a FIFO fd0 with an authenticated writer. This reconciliation does not authenticate protected origin.

Workflow feedback (optional)

docs/research/current-priorities.md revision 8 still quotes the earlier κ. certificates/selected-result.json is the record this package followed. A one-line pointer there would have saved a re-read.

Open in Web Open in Cursor 

An identity replacement of snapshotted json.loads during the first fd0 read
cannot reach three callbacks and SEALED on SHA-256 7001e2ba. The final ingress
receipt matches that refusal. Selected kappa is unchanged.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.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.

2 participants