[AdaWorldAPI]

D-TGB-4 (EXO-58, under EXO-44). A thought's receipt holds what is needed to recompute its writes after its live slot is gone, and replay reproduces those writes from the receipt alone.

This is lab only, inside the workspace-excluded crates/thought-git-probe. It does not touch the contract, cycle_driver, LanceCycleWriter, the supervisor or the CE64 bits. Claim: EXO-50 session_019fbZ92dFKQ5BJHrA3YcH9A/003, acknowledged in session_01G22yT6htkcdyXsihxxXdrg/036.

What changes

src/receipt.rs:

• Receipt holds:

  • base ref (dataset, branch, version);
  • ProgramId plus template version;
  • ProgramInputs as two lanes, sources and targets, read at the same position;
  • seed and ExecConfig;
  • the live slot and generation, kept as provenance only;
  • a fold over the values read, the provenance roots, a fold over the writes, and the outcome.

• replay(receipt, store, config) takes no slot pool. It refuses on:

  • a missing seed;
  • a changed config;
  • an unknown base;
  • changed sources (checked by the source fold);
  • a different output (checked by the output fold).

  The writes themselves are not stored in the receipt.

• replay_live, and capture / stage, check that the slot still holds the same thought under the same generation. A stale, reused or ABA slot fails closed.

• MetadataTable is keyed by (dataset, data version, thought id), separate from the data version it describes.

  • stage refuses before the data version is sealed (data first, then metadata) and on a key naming another thought.
  • seal writes a fold over every field, last.
  • landed / replay_landed accept only a sealed row whose seal still matches and whose data version exists.

• Replayed reports the receipt's roots as inherited, with zero independent roots, so a replay cannot earn new certification.

• BaseRef and VersionStore are probe-local stand-ins for D-CPG's DurableRef (no contract type yet) and for Lance.

Also in this PR:

• tests/receipt.rs: 10 falsifier tests.
• .github/workflows/thought-git-probe.yml: until now no workflow ran this crate.
• Board: entries/2026-10-21-d-tgb-4-receipt-replay.md, the STATUS_BOARD D-TGB-4 row (In PR), the PREREGISTRATION row, and the regenerated entries index.

Falsifiers and disables

Each disable was applied to the source and the suite rerun. Each one turned the named test red; the tests were green again after restore.

Test	Disable
Released Commit/Plan/Prune thoughts replay cell for cell, even after their slot is reused	replay takes no pool (by signature)
Missing seed refuses	default the seed to 0
Changed seed is an output mismatch (the two seeds are first shown to give different writes)	skip the output check
Changed config refuses	skip the config check
Changed source refuses; a change to a non-source cell does not	skip the source fold
Freed, reused and ABA-reused slots fail closed	ignore the slot generation
Contradictions and roots keep their meaning (equal next_version, exactly one contradiction)	report roots as independent
A replay earns no new certification, while a genuinely new root does (Gadamer)	report roots as independent
Unsealed, torn, missing and data-less metadata rows are not landings	accept unsealed rows / skip the seal comparison / stage before the data version
The same plan in two slots has one identity; a seed change alters it	put the slot into the identity

One disable first came back green: test 8's prior horizon already held the receipt's roots, so mislabelling them as independent could not look new. The prior now predates them, and the disable turns both tests 7 and 8 red.

Local results:

• cargo test: 36 passed (26 existing + 10 new).
• cargo clippy --all-targets -- -D warnings: clean.
• cargo fmt --check: clean.
• Entries index and supersession index are current.

Measured and not measured

Measured: the 10 tests and 10 disables, on the fixture address space (12 rows × 4 lanes) with one program (DERIVE). No timing was taken.

Not measured or not built:

• Lance-backed metadata.
• Atomicity of the two Lance versions (still open in thought-git-basics item 8).
• Programs other than DERIVE.
• Rebase-as-replay.

The folds are 64-bit witnesses, not proofs.

Caveat: replay returns a Thought whose writes field is the fixture's existing Vec<(Cell, u64)>, carried over from D-TGB-0. The comparisons themselves run on folds. Whether a reference may hold tuples internally is still an open operator question (/036 item 1).

Open

• The plan places the receipt, slot and DurableRef in lance-graph-contract at D-TGB-7. This PR does not.
• D-RFT-5 can reference this receipt shape. I will post the shape on EXO-50.

---

Generated by Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: bc5a9cb0-9244-4da7-87d4-52ba56bd395a

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[AdaWorldAPI]

@codex review

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	🔄 Running since 2026-10-21T19:51:51.870694Z	ed54930	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3496242fc7

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[AdaWorldAPI]

@codex review

---

Generated by Claude Code