Skip to content

fix(dispatch-runner): give the predicate verdict to Hypatia, and record no_op and unverified (E0-02) - #683

Merged
hyperpolymath merged 1 commit into
mainfrom
fix/e0-02-verdict-pass-through
Nov 15, 2026
Merged

hyperpolymath merged 1 commit into
mainfrom
fix/e0-02-verdict-pass-through

Conversation

@hyperpolymath

Copy link
Copy Markdown
Owner

What

scripts/dispatch-runner.sh gives its predicate verdict to mix hypatia.record_outcome --verdict, and the call for no_op and unverified is no longer skipped. The automation
quarantine and the licence refusal gate do not change.

Work order: E0-02 (hyperpolymath/hypatia, docs/operations/e0-closeout/work-orders/).
Depends on hypatia E0-01 (merged). Related: gitbot-fleet #647; hypatia #984, acceptance A3, A5.

Why

The runner collapsed each verdict to an outcome and made no call for no_op and unverified, so
Hypatia never saw an attempt that fixed nothing. Hypatia now takes --verdict and stores those
two as audit rows that carry no credit.

The change

  • hypatia_detect_verdict_mode: learns once per run whether Hypatia knows --verdict, from
    mix help hypatia.record_outcome and never from an exit code. A Hypatia from before E0-01
    ignores an unknown option, so a no_op call would be recorded as a success.
    • verdict: each of the 5 verdicts is passed with the outcome of the pair table (success for
      no_op and unverified), and those two are recorded. A refusal passes no verdict.
    • legacy: no --verdict, and no call for no_op or unverified: the calls the runner made
      before.
    • off: no calls (reporting off, no checkout, or no mix).
    • help fails: the run stops with a visible error. A Hypatia that does not answer is not an
      older Hypatia.
  • One line per run says which mode it uses.
  • The run learns the mode with one call, hypatia_detect_verdict_mode || exit 1, placed after
    record_outcome and after exit 78, so the quarantine still ends the run before the manifest.

Evidence

  • tests/outcome-verdict-test.sh: 60 checks, 0 failures (it had 43; before the change the 17 new
    ones failed because the function did not exist). A stub mix on PATH records its arguments and
    answers help as each case needs: knows --verdict (all 5 verdicts, outcome and verdict checked),
    an older Hypatia, a failing help, and reporting off.
  • tests/automation-quarantine-test.sh passes with no change to that file;
    tests/hypatia-record-outcome-smoke.sh passes; exit 78 is still at line 152, before the new call.
  • Legacy is today's behaviour exactly: I ran the same 6 verdicts through the previous script and
    the new one in legacy mode, and the 4 mix calls are identical.
  • End to end against the real Hypatia (main of hypatia, MIX_ENV=test, HOME redirected so
    nothing touched ~/Documents): detection gave verdict; no_op, unverified and failed were
    stored as verification no_op, unverified and failed; the real gate counted ignored: 2, successes: 0, failures: 1. I removed the rows I created.
  • Mutations (each restored): --verdict passed in every mode (1 check fails); no_op/unverified
    recorded for an older Hypatia (2); the option never learned from the help (8); the run does not
    stop when help fails (1); a failing help treated as an older Hypatia (3).

Not done, and limits

  • Nothing here runs in production: the runner exits 78 before it reads the manifest, so this
    takes effect only when the quarantine is lifted. The tests extract the runner's functions, as the
    existing ones do.
  • The runner passes no --pattern-id and no --match-kind to Hypatia. Hypatia's E0-18 can now
    record both; wiring them from the manifest entry is a separate change.
  • Observed, not caused here: overriding Hypatia's store with HYPATIA_VERISIM_DATA_PATH is
    refused by Mix's compile_env check unless it equals the compile-time path. Anyone who redirects
    the store for a test will hit it.
  • I did not run shellcheck (not installed here); CI may.

Process notes

  • The first write to this repository from this session; the push worked with the same fine-grained
    token as hypatia.
  • The work order names Sonnet at medium effort (cap USD 4). This session was switched to Opus 5.5
    at high effort; I did not choose that. Cost: unknown to me; please read it in the Anthropic Console.
  • Draft. Not merged, approved or marked ready by me.

…rd no_op and unverified (E0-02)

The runner collapsed each verdict to an outcome and made no call for no_op and
unverified, so Hypatia never saw an attempt that fixed nothing. Hypatia's
`mix hypatia.record_outcome` now takes --verdict (hypatia E0-01), and stores
no_op and unverified as audit rows that carry no credit.

The runner learns once per run whether Hypatia knows the option, from the help
text of the task and never from an exit code: a Hypatia from before E0-01
ignores an option that it does not know, so a no_op call would be recorded as
a success.

* Hypatia names --verdict: each of the 5 verdicts is passed with the outcome of
  the pair table (success for no_op and unverified), and no_op and unverified
  are recorded. A refusal passes no verdict.
* An older Hypatia: no --verdict and no call for no_op or unverified, exactly
  the calls that the runner made before (checked against the previous script).
* The help command fails: the run stops with a visible error. A Hypatia that
  does not answer is not an older Hypatia.
* One line per run says which mode it uses.

The automation quarantine (exit 78, before the manifest) and the licence
refusal gate are not changed. tests/outcome-verdict-test.sh gets a stub mix on
PATH for the 4 cases: 60 checks.

Co-Authored-By: Claude <noreply@anthropic.com>
@coderabbitai

coderabbitai Bot commented Nov 15, 2026 •

Copy link
Copy Markdown
Contributor

Explain Change →

Caution

Review failed

The pull request is closed.

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

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

@sonarqubecloud

Copy link
Copy Markdown

@hyperpolymath
hyperpolymath marked this pull request as ready for review November 15, 2026 18:26
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 15, 2026 •

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

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

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-15T18:28:17.142551Z ba569ec Draft marked ready
🔒 Security Review ✅ Completed 2026-11-15T18:29:02.319454Z ba569ec Draft marked ready
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.

@hyperpolymath
hyperpolymath merged commit 7cc3144 into main Nov 15, 2026
66 checks passed
@hyperpolymath
hyperpolymath deleted the fix/e0-02-verdict-pass-through branch November 15, 2026 18:27
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.

3 participants