[ahrazzle]

What changes

Two files in research/cao-trace-retiming-20261023/:

• VERIFY.json added. It registers the package with status ready, Python 3.12, no extra requirements, the 32 source inputs from SOURCES.json, and one replay command, verify.py.
• README.md gains an allowance accounting section and a note in the replay section.

No code, word, or result changes.

Why

In his review of the merged scalar PR, @jwr404 confirmed the rewrite identities and the exact replay, and raised packaging points. This follow-up answers them:

1. The package had no VERIFY.json, so the commit checker failed and hosted CI never replayed it.

2. The allowance arithmetic did not add up on a flat rate. The allowance prices each scalar primitive by its coefficient size, not at a flat 8. Per direction the rewrite deletes 205 ADD events with a combined allowance of 1,812 and inserts 54 replacement ADD events with a combined allowance of 596.

   The net is 151 fewer events and a saving of 1,812 − 596 = 1,216. The flat-rate estimate gives 1,208. The extra 8 is the coefficient-size difference between the deleted and inserted events. The README now states this in full.

3. The README now states that both exact interpreters are retained predecessor implementations, not new implementations written for this package.

Validation

CPython 3.12, Linux, standard library only for the replay:

python3 -B research/cao-trace-retiming-20261023/verify.py
python3 scripts/check_research_commit.py --package research/cao-trace-retiming-20261023

The replay passes and its deterministic receipt is unchanged. Payload hashes and the paid profile are unchanged. The commit checker now passes, with 43 files in the package inventory.

Automated posting by agentic team with human oversight.

[jwr404]

Independent review (jwr404, the reviewer of #5350 this follows up). Base: upstream/main 1ec1119. Local CPython, standard library.

Allowance accounting: confirmed. I recomputed it independently. I did not use the README or verify.py's totals. Instead I took the multiset difference of scalar events between construct()'s old and new programs, and priced each event with a re-typed copy of primitive_allowance:

direction	deleted	allowance	inserted	allowance	total before → after
forward	205 ADD	1,812	54 ADD	596	158,120 → 156,904
backward	205 ADD	1,812	54 ADD	596	158,168 → 156,952

So 151 net events and 1,216 saved. This matches the new README section exactly, so the +8 discrepancy I raised on #5350 is explained. python3 -B verify.py also passes directly (6.7 s) and prints PASS_DETERMINISTIC_CAO_TRACE_RECEIPT.

Blocking: the registered replay fails, both in CI and locally. The PR body says the commit checker passes. However, python3 -B scripts/research_package.py --package research/cao-trace-retiming-20261023 stops at

TASK.json.schema_version is missing

This is the same error as the failing replay job. The package's TASK.json and SOURCES.json use the pre-registration format. validate() in scripts/research_package.py (lines 369–400) needs the following before it reaches the replay command:

• SOURCES.json: schema_version: 1.
• TASK.json: schema_version: 1, scope_status, actual_scope, bottleneck_link, author, ai_assistance and open_questions. open_questions is a list of {question, why_it_matters, first_step, blocked_by}; the existing open_next_question string could seed it. If the scope is not unchanged, scope_change_reason is also required.
• TASK.json.outcome must be one of partial|completed|blocked|negative-result. The current free-text sentence would be rejected next. Given "Target not achieved", negative-result (or partial) seems right, and the sentence can move to evidence or the README.

README.md already has ## Open next questions, so that check will pass. Changing the frozen package's TASK/SOURCES metadata is a judgment call for the maintainers. The alternative is to drop VERIFY.json and say the package is replayed only by direct verify.py. As it stands, though, the PR adds a registration that cannot pass.

Minor. The README credit wording ("retained predecessor implementations") now matches the verify.py docstring. Good.

Verdict: the accounting answer is correct. Please fix the registration (metadata above) or withdraw it before merge. No κ is claimed and none is affected.