Skip to content

compass(rules): the reviewer checks gates 1-3 as they apply per task, so approval never waits for the per-wave GPU tier (#369) #371

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jgong5 merged 1 commit into from
Sep 24, 2026
Merged

compass(rules): the reviewer checks gates 1-3 as they apply per task, so approval never waits for the per-wave GPU tier (#369) #371

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions atom/compass/AI_DEV_RULES.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -158,8 +158,8 @@
gets the label. **Holds are landing preconditions, and there are three:**
`need human` on the PR or below it (a label on an issue a PR delivers counts as
on that PR; every escalation rule in this file holds through this label), an
APPROVE covering each head (the reviewer checks gates 1-3 before approving, and
the approval is gate 4), and the tree check below. A
APPROVE covering each head (the reviewer checks gates 1-3 as they apply per task

@jgong5 jgong5 Sep 24, 2026

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

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

Non-blocking (follow-up issue, not a change to this PR): after this clause is narrowed, no sentence in the file says who runs gate 1's per-wave GPU superset, or when.

Rules quoted:

  • L121-122: "the GPU-free tier per task, the GPU superset per wave as a delta."
  • L154-155: "it does not wait for the per-wave GPU superset."
  • Principle 6: "Refuse rather than fall back. A declined answer with a named reason is a result. A guessed one is a defect."

What I checked. grep -n -i -E "gpu|superset|wave|tier" at the head gives only L121-122 and L155 for the superset. L120 still says "Four gates land a task, all required", and L155 exempts landing from the superset. Before this PR, the only reading that gave the superset an owner was the literal one at L161: the reviewer runs it before approving. That reading contradicted L154-155, and this PR removes it, as #369 asked.

Why it does not block. It drops no duty a reviewer has in practice. That reading could never be followed without breaking L154-155, and the PR's own named result is that the reviewer does not wait. The gap was already in the file. This PR only makes it visible.

Suggested follow-up. File an issue asking the owner or planner who runs the per-wave delta, and when it runs, for example the landing agent after a wave's last landing. Until that is answered, a literal reader will find the tier required at L120 with nobody assigned to run it.

before approving, and the approval is gate 4), and the tree check below. A
hold names the one that is unmet. **A violation of any other rule seen in an
approved PR is landed and filed as an issue, not held.** Where a
handoff note contradicts this file, this file wins.
Expand Down