Skip to content

train b35 integration gate (do not merge) - #414

Closed
roodboi wants to merge 7 commits into
nextfrom
train/b35
Closed

roodboi wants to merge 7 commits into
nextfrom
train/b35

Conversation

@roodboi

@roodboi roodboi commented Oct 22, 2026 •

Copy link
Copy Markdown
Contributor

Integration gate for merge batch b35: origin/next + #375 (HACK-1324, HACK-1320, HACK-1321). Tree 29542676. Closed after the batch squashes.

🤖 Generated with Claude Code


View with [code]smith Autofix with [code]smith
Need help on this PR? Tag @codesmith-bot with what you need.

roodboi and others added 7 commits October 21, 2026 17:50
…hase

Native storage carriers took their deadline from HACK_COMPOSE_STARTUP_TIMEOUT_MS,
and guard storage proofs consumed the start deadline of up/restart. On loaded
hosted runners storage preparation alone exceeded a 15 s start budget
(native-config-down-hooks) and a dozen build/spawn guard proofs exhausted the
90 s start budget (native-config-build).

- HACK_NATIVE_STORAGE_TIMEOUT_MS (default 90 s, 1-600 s) bounds each storage
  carrier; it is never derived from the start budget.
- Guard storage proofs during native up/restart extend the start deadline by
  their monotonic elapsed time instead of consuming it.
- A refusal that finds its carrier budget exhausted reports the closed effect
  reason storage-budget, and the summary names the variable.
- Unrecorded command group absence uses a monotonic deadline; a fake-clock test
  pins its probe-budget refusal (kills mutation M6).
- Network acceptance: named 75-minute whole-run budget that gates command
  starts only, so it never kills an in-flight carrier command.
- Build/down-hooks scenarios get measured command budgets and log closed
  per-command storage timings.

Refs HACK-1324, HACK-1320, HACK-1321

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The slowest of 42 hosted passes (2113-3343 s) used 93% of the old 60-minute
budget; 80 minutes keeps 1.44x that pass inside the 150-minute shard job.

Refs HACK-1320

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This PR's own hosted run passed in 3494 s, 97% of the old 60-minute budget
and 73% of the 80-minute one. 90 minutes keeps 1.55x the slowest of 44
measured passes and stays inside the 150-minute shard job.

Refs HACK-1320

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…uard exclusion

Guard storage proofs extend the up/restart start deadline instead of consuming
it, but their sum had no bound. The excluded total is now capped at 16 x
HACK_NATIVE_STORAGE_TIMEOUT_MS or 1 hour, whichever is smaller: the deadline
moves at most that far, a proof that would start past the cap is refused before
it runs, and a proof that ends past it refuses the phase with a closed,
value-free error whose effect reason is storage-budget.

Add a command-level failure-injection test for the start.excludeStorage wiring.
The Docker stand-in reads the CLI's own phase trace and delays the carrier
dependency check only while a guard.storage span is open, so the guard proof
outlasts a 3 s start budget. It fails with the wrap removed. Add unit tests for
the cap, each failing under mutation, and document the per-proof bound, the
phase cap and how the two variables interact.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@roodboi roodboi closed this Oct 22, 2026
@roodboi
roodboi deleted the train/b35 branch October 22, 2026 13:34
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.

1 participant