Skip to content

Raise the Repair and base-move budgets to 5 - #95

Merged
JacobStephens2 merged 2 commits into
mainfrom
issue-90
Sep 27, 2026
Merged

JacobStephens2 merged 2 commits into
mainfrom
issue-90

Conversation

@JacobStephens2

Copy link
Copy Markdown
Owner

Summary

Every Run now allows up to 5 Repairs and up to 5 rounds after the Base branch moves while CI runs, instead of 3.

 // src/run.rs
-const MAX_REPAIRS: usize = 3;
+const MAX_REPAIRS: usize = 5;
-const MAX_BASE_MOVES: usize = 3;
+const MAX_BASE_MOVES: usize = 5;

The README and the site (unit 4 text and the illustrative Failed run) now say 5. The tests that assert the caps now expect 5 and have names that don't include the number.

Evidence

  • Before: with the tests updated to 5 and the code still at 3, cargo test --test ci failed:

    • red_ci_after_the_last_repair_is_a_failed_run expected 6 sessions (implement + 5 Repairs) but got 4.
    • conflict_and_ci_fix_repairs_share_one_cap expected a conflict Repair plus 4 CI-fix Repairs before "repairs exhausted: conflict" but got fewer.

    After: cargo test passes in full. It includes:

    • fails_when_the_base_branch_keeps_moving_during_the_ci_wait, which asserts 5 merges and then "merged it again 5 times".
    • cargo clippy --all-targets is clean.

Merge Danger

Door: two-way

Revert the two constants.

Blast Radius: small

A Failed run can now start up to 2 more agent sessions before it gives up, which means more time and API cost on runs that were going to fail anyway.

Unaddressed findings

Standards

  • Magic numbers tied to the cap in tests (tests/ci.rs, tests/mergeable.rs): not changed. The caps are private to src/run.rs and the integration tests are black-box, so spelling out the expected counts is intended.
  • Shotgun Surgery (README/site/test doc comments restate the number): not changed. The issue asks the docs to state 5, so the number is supposed to appear there.

Spec

  • docs/research/thirdshift-site-prototype-handback.md:42 still quotes "starting Repair n of 3": not changed. It is a research handback recording what was checked at that time, so it's historical like docs/grilling.md.

Closes #90

🤖 Generated with Claude Code

JacobStephens2 and others added 2 commits September 27, 2026 12:38
A flaky CI or a busy Base branch is less likely to fail a Run that could
have succeeded. Tests, README and site now state 5.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@JacobStephens2
JacobStephens2 merged commit eb1bff9 into main Sep 27, 2026
13 checks passed
@JacobStephens2
JacobStephens2 deleted the issue-90 branch September 27, 2026 12:45
@JacobStephens2 JacobStephens2 mentioned this pull request Sep 27, 2026
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.

Raise the Repair and base-move budgets to 5

1 participant