Kernel: spent attempts hold the item even while a refused Run is queued #158

Activity

1. don-petry commented on Oct 18, 2026

   

   don-petry

   on Oct 18, 2026

   ContributorAuthor

   More actions

   Working rulings (overturnable in review)

   Solution Architect, delegated. On main at 72f8643 (#162). Builds on #159 (branch kernel/rubric-coverage, record 0029), which must merge first: UNCOVERED_SEEDS is on its branch.

   The cause. In _decide, the wait for a Run in flight returns nothing, working before the brakes are asked. While any Run waits, a count already spent is hidden. No Run in flight can lower a count: attempts and rounds count completed Runs, and only a writer's action renews them. So a Run that waits past a spent count was admitted on a short list, and the item should be held while it waits.

   1. Where the Hold is read. In kernel/decide.py only. Work out the wanted Run before the in-flight wait. With a Run in flight, ask the brakes about the wanted Run against the completed Runs since a writer. Return a Hold if they give one. Anything else, a Pause's deferral included, keeps today's nothing, working. A converged item asks on_act the same way, Holds only. brakes.py is not edited. Why: the issue says any Intent. Taking only Holds keeps the Pause as it is with a Run in flight. Under a Pause the State is not held anyway, so no new move appears.

   2. Completed Runs only. The brakes get [r for r in since if r["status"] == "completed"]. Why: _spent counts Runs in flight toward the Run ceiling. Given the whole list, the tenth Run, admitted with nine completed, would be held at the ceiling it was admitted under.

   3. A superseded Run is not an attempt, before or after: its outcome is superseded, not item (I-23). So r0 counts for nothing. When the lane supersedes r4 at start, the evidence stays r1 to r3, and held stays held. Whether a superseded Run counts toward the Run ceiling is not this story. _spent counts it and the simulator's check does not. File it if wanted.

   4. Boundary tests, written first:

   • Seed 297's facts: verifying / awaiting_review on content C. r0 review-change superseded. r1 to r3 completed, failure, outcome item. r4 queued. Expect held / attempts_exhausted / hold / needs_you, evidence [r1, r2, r3]. Again with r4 in_progress.
   • Without r3 it gives nothing, working: r0 is no attempt.
   • max_runs_per_item - 1 completed and one in flight: nothing, working. max_runs_per_item completed and one queued: held / run_ceiling.
   • Converged, approval_due, max_attempts failed approvals and one queued: held / attempts_exhausted.

   5. Mutants, appended to MUTANTS, each with text that appears once:

   • "a spent count waits behind a Run in flight": the in-flight brake's if becomes if False:. Seed 297's test kills it.
   • "a Run in flight counts toward the ceiling": the completed filter becomes the whole list. The ceiling-minus-one test kills it.

   6. The simulator. UNCOVERED_SEEDS goes back to 1, every seed under faults, and UNCOVERED stays 0.5. Drop the issue-158 sentence from its comment, and say "every seed" in sim/AGENTS.md. sim/world.py is protected. Run with SIM_SEEDS=40 locally; CI's shards run all 300. If a seed then fails for another cause, file it rather than narrow the spread.

   7. Record. No 0031: this makes decision 0003's brakes hold as written. Add a row to 0003's Boundaries, "A count the completed Runs have spent holds the item, whatever Run waits", naming the seed-297 and ceiling tests.

   8. Out of scope. The short list itself. No Kernel can see it, and the re-ask at start (I-38) stops r4. #42's issue decision reads its brakes the same way, in its own module.

   Needs the Maintainer? None.
2. don-petry

   added a commit that references this issue on Oct 18, 2026

   Simulator: under a Pause, a spent count shows while a Run is in flight

   ...

   2d71b32
3. don-petry

   closed this as completedin #170on Oct 18, 2026