Skip to content

Floor cost w2: shrink five small emit-module fixtures (778k steps) - #10637

Closed
gunbai-bot[bot] wants to merge 6 commits into
mainfrom
still-newt-660/roster-transition-failure-mode
Closed

gunbai-bot[bot] wants to merge 6 commits into
mainfrom
still-newt-660/roster-transition-failure-mode

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Sep 6, 2026

Copy link
Copy Markdown
Contributor

Auto-opened by session-dashboard for session still-newt-660.
Pushing to still-newt-660/roster-transition-failure-mode advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

  • Title describes the change (not the session id or branch).
  • PR body summarises what and why (replace the TODO below).
  • Tests run: name the command (e.g. npm test, cargo test) and the result.
  • If this closes a work item, the body contains a Closes #N directive.
  • No commits on this branch are surprises (no fork/cherry-pick I did not make).
  • No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

  • TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

gunbc-ci-auto-heal added 6 commits September 5, 2026 15:14
…wt-660/roster-transition-failure-mode

# Conflicts:
#	dag/gunbc/recurring_failure_mode/roster.dag
…-transition-failure-mode

# Conflicts:
#	dag/gunbc/recurring_failure_mode/roster.dag
Ledger-Repair-Judged: docs/design-failure-modes.md
Ledger-Rows-Repaired: docs/design-failure-modes.md claim_edit_changes_execution_policy_outside_reported_cost_delta
Ledger-Repair-Judged: docs/design-rung-drops.md
…-transition-failure-mode

# Conflicts:
#	dag/gunbc/recurring_failure_mode/roster.dag
#	docs/design-failure-modes.md
@briansrls
briansrls marked this pull request as ready for review September 6, 2026 01:38
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Sep 6, 2026 •

Copy link
Copy Markdown

Codex Review Summary

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

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-09-06T01:42:47.201624Z b72d3b9 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.

@gunbai-bot

gunbai-bot Bot commented Sep 6, 2026

Copy link
Copy Markdown
Contributor Author

Closing: this is a stale duplicate of #10579, which already merged.

This PR is on still-newt-660/roster-transition-failure-mode at b72d3b9f0f, which is that branch's pre-heal head — the state before docs/design-failure-modes.md was regenerated onto it. #10579 shipped the same two files from the same branch and landed as 9dbe14e1d8; the row is on main now:

$ git cat-file -e origin/main:dag/gunbc/recurring_failure_mode/claim_edit_changes_execution_policy_outside_reported_cost_delta.dag
YES
$ git log --oneline -1 origin/main -- .../claim_edit_changes_execution_policy_outside_reported_cost_delta.dag
9dbe14e1d8 Record claim-edit policy transitions omitted from cost reports (#10579)

So there is nothing to fix and nothing to push. The three failing checks are explained by what the head is rather than by a defect: heal-generated-artifacts fails because the generated projection is absent from this commit — that is precisely what the heal added before #10579 merged — and the failing run (33981638366) is the same one that sat in action_required earlier because a bot-triggered pull_request run does not dispatch without approval.

Merging this would re-add a row that already exists, duplicating an identity in recurring_failure_mode_roster, whose order is load-bearing.

— sent from deep-wolf-853

@gunbai-bot gunbai-bot Bot closed this Sep 6, 2026

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b72d3b9f0f

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".


"INSTRUMENT AND CURRENT MITIGATION: `claim_executor --required-ci --source-root dag --source-root src/v2 --required-lane witnesses` produces the run receipts through `v1.cli_run` `write_required_floor_claim_cost_tsv` and `write_required_floor_disposition_tsv`, published as `required_floor_claim_cost.tsv` and `required_floor_disposition.tsv`. The wave's reconciliation was a manual identity join of those run products, not a repository-enforced reconciliation producer. RECOGNITION: bind baseline and candidate to immutable run subjects, join every identity in both receipts, list added and dropped executions, retain outcome and changed-witness policy transitions, and reconcile common-row deltas plus additions minus removals to the complete corpus total. Report the intended subset separately. Treat unavailable or censored costs as such, never as exact zeros. This manual reconciliation exposed the cost receipts; the existing changed-witness gate itself detected the route-gap receipt. Restoring an unintended source edit is a scope correction; moving it into a helper merely to avoid selection is not a repair. ",

"RUNG FOUND AT: mitigatable for the cost-report boundary. Complete run products and explicit policy diagnostics exist, and manual reconciliation detected the discrepancy, but no executed evidence here establishes that cost reports must consume that reconciliation. CEILING: mechanically preventable for report completeness, because finite identity sets, dispositions and exact step deltas are decidable at the receipt boundary. Whether a newly executed claim was intended or its additional coverage is worth its cost remains an author/operator judgment, outside that mechanical guarantee; this row does not propose overriding the declared verdict-only policy. NEXT TRIGGER: a cost-change reporting consumer that binds both run subjects, refuses incomplete or unreconciled accounting, and publishes execution and adjudication transitions beside the total. It must discriminate both a newly admitted debt identity and an unchanged-membership route-gap transition, with an unchanged-policy positive control; an added warning alone is insufficient. ",

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Stop describing censored step deltas as exact

When either run contains a deadline-preempted claim, the receipt cannot provide the “exact step deltas” used to justify this ceiling: dag/gunbc/floor_cost_distribution.dag:32-37 states that interrupted rows are sampled only at 1,024-step polling boundaries and therefore carry lower bounds. A reconciliation consumer must preserve an unknown/censored delta or refuse an exact total rather than treating report completeness as mechanically derivable from exact values.

Useful? React with 👍 / 👎.

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.

0 participants