#4023 fixed the syntax corruption #3998's own patch introduced and moved
repair_attempt recording to execution time, but left
REGATE_REPAIR_MAX_ATTEMPTS_PER_SHA at 2 -- #3998's stated intent was to
raise it to 5 for resilience against transient failures (a single AI
provider timeout no longer exhausts the budget). Bumps the constant and
updates both orb-retry-storm regression tests' attempt counts to match;
one of them was also still asserting the pre-#3998 dispatch-time
recording behavior, never executing the fanned per-PR job.
Summary
repair_attemptrecording to execution time, but leftREGATE_REPAIR_MAX_ATTEMPTS_PER_SHAat 2 — fix(queue): prevent regate repair exhaustion from premature attempt counting #3998's own stated intent (point 3 of its PR description) was to raise it to 5, for resilience against transient failures (e.g. a single AI-provider timeout no longer exhausts the whole repair budget).#orb-retry-stormregression tests' hardcoded attempt counts to match the new cap.Supersedes #4027 (same underlying gap, but that branch's diff conflicts with #4023 which already landed the syntax fix independently).
Scope
Validation
npx tsc --noEmit— clean.npx vitest run test/unit/queue.test.ts— full file green (both#orb-retry-stormtests pass).npm run test:coverage— clean, changed line confirmed hit.Safety
site//CNAME/lovablechanges.