Skip to content

Math-#335: c2-torus omission controls require intended MISMATCH reason (Draft; merge HOLD for Dylan workflow OK) - #562

Draft
d6g8k5htny-coder wants to merge 1 commit into
mainfrom
cursor/c2-torus-omission-reason-2920
Draft

d6g8k5htny-coder wants to merge 1 commit into
mainfrom
cursor/c2-torus-omission-reason-2920

Conversation

@d6g8k5htny-coder

Copy link
Copy Markdown
Owner

Math-#335 — omission controls must carry the MISMATCH reason

Draft. Merge HOLD until Dylan OKs this .github/workflows/ change. Do not mark Ready. Engineering only. This does not change checkers, proofs, RESULTS.json, manifests, or transfer.py, and it does not claim scientific acceptance.

Base: dee08f4f4c984bfac0ee4ff289c87c677efb5ec9
Head: 6bd4fae46575c068af58b80a3ba13ba2105ffae4
Workflow blob: e8c11afdc19eba9e9090aa7dca8ab130e411c833 → 9886f17ef75729f4ffafc48c10bcc50839e32ff7
Unchanged transfer.py blob: 05dc3fba1d1f1399519569f113911ed043a67fba

Characterization

Local interpreter: CPython 3.12.3 (/usr/bin/python3). The workflow pins hosted CPython 3.11.16; that pin is not installed here. Commands match the job: PYTHONDONTWRITEBYTECODE=1, packet directory, -B -S (normal) and -B -O -S (optimized), stdout and stderr kept separate. The workflow's manifest/pin/copy step printed manifest, pins and copies verified (rc 0, empty stderr).

No mutant failed through require() / AssertionError. Each genuine mutant exited 1 with stdout exactly MISMATCH against RESULTS.json\n (30 bytes, SHA-256 800a127a1a5fc450d536fddc71c7c47884d283a798f08c885f782544f37452d9) and empty stderr (SHA-256 e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855). Normal and optimized streams were byte-identical. The fixed literal in the workflow is that driver line.

child mode rc stdout reason stderr empty seconds
transfer.py --check normal 0 check passed yes 79.3
transfer.py --check optimized 0 check passed yes 81.5
controls.py normal 0 ends controls: 0 failures yes 8.7
controls.py optimized 0 ends controls: 0 failures yes 8.9
--mutant no-image normal 1 MISMATCH against RESULTS.json yes 7.7
--mutant no-image optimized 1 MISMATCH against RESULTS.json yes 7.6
--mutant no-theta normal 1 MISMATCH against RESULTS.json yes 79.5
--mutant no-theta optimized 1 MISMATCH against RESULTS.json yes 80.1
--mutant drop-ratio normal 1 MISMATCH against RESULTS.json yes 79.9
--mutant drop-ratio optimized 1 MISMATCH against RESULTS.json yes 80.2
--mutant eta-half normal 1 MISMATCH against RESULTS.json yes 79.1
--mutant eta-half optimized 1 MISMATCH against RESULTS.json yes 78.1
--mutant fp-const normal 1 MISMATCH against RESULTS.json yes 80.2
--mutant fp-const optimized 1 MISMATCH against RESULTS.json yes 79.2

controls.py stdout was 1789 bytes, SHA-256 90247407c99f551e4d2c7eda56602b408b4fe11fbacde3f1f92b215f5de0eeed, same in both modes. --check stdout SHA-256 c6b9ef18c4dcb1476b486467d448ef6967aa414837cfa1c6c512598965191045.

Repair

.github/workflows/c2-torus-transfer.yml only. Each omission control is accepted only when rc is 1, stdout contains the fixed literal MISMATCH against RESULTS.json, the stderr file exists, and stderr is empty. Any other exit 1, exit 0, or other status fails the job and prints that mutant's stdout and stderr. Stdout and stderr are separate files (/tmp/mutant_$m.stdout, /tmp/mutant_$m.stderr).

Matcher evidence (VM, not committed)

The acceptance loop was extracted from this workflow and run against the captured genuine streams and against real short-lived Python processes that wrote the adverse streams. 27/27 rows matched. Genuine normal and genuine optimized accepted all five mutants (exit 0). Each adverse case was applied to each of the five mutant slots while the other four stayed genuine; every one exited 1 and printed that slot's logs.

case slots want got
genuine captured streams, normal 5 together 0 0
genuine captured streams, optimized 5 together 0 0
unrelated RuntimeError: unrelated crash, rc 1, traceback on stderr each of 5 1 1
silent exit 1, empty stdout and stderr each of 5 1 1
wrong reason WRONG REASON, rc 1, empty stderr each of 5 1 1
exit 0 whose stdout is the MISMATCH line and stderr empty each of 5 1 1
rc 2 whose stdout is the MISMATCH line and stderr empty each of 5 1 1

Sample failure text for no-image:

  • runtime: mutant no-image not rejected for the intended reason (rc=1) then stderr RuntimeError: unrelated crash
  • silent: same heading (rc=1) with empty stdout and stderr sections
  • wrong reason: (rc=1) and stdout WRONG REASON
  • exit 0: (rc=0) and stdout MISMATCH against RESULTS.json
  • other rc: (rc=2) and stdout MISMATCH against RESULTS.json

Limits

Scientific effect NONE. No theorem, proof, axiom, or RESULTS.json disposition. Hosted 3.11.16 was not the interpreter for this characterization. The probe feeds streams to the acceptance loop; it is not a hosted Actions run.

Grok Bot agent 7 (Grok Bot support agent; non-Claude, nonauthor lane) · credit 0 · sci NONE · OBL OPEN · engineering only, math fail-closed

Open in Web Open in Cursor 

The replay job treated any exit code 1 as a passed omission control.
Each mutant now passes only when it exits 1, stdout contains
"MISMATCH against RESULTS.json", and stderr is empty. Any other
status prints that mutant's separate stdout and stderr and fails the job.

Co-authored-by: Dylan Roy <d6g8k5htny-coder@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