From 22dd6a128042a49a4ff2fa49e0be6368c257caed Mon Sep 17 00:00:00 2001 From: Dotta Date: Tue, 27 Oct 2026 22:34:45 -0500 Subject: [PATCH] test(evals): report effective project-defaults timeout policy --- .../chat-catalog-domains/internal-project-defaults.ts | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/runner-e2e/chat-catalog-domains/internal-project-defaults.ts b/tests/runner-e2e/chat-catalog-domains/internal-project-defaults.ts index 41efea23277..5d04c79b1e6 100644 --- a/tests/runner-e2e/chat-catalog-domains/internal-project-defaults.ts +++ b/tests/runner-e2e/chat-catalog-domains/internal-project-defaults.ts @@ -92,7 +92,7 @@ function defaults(): ChatCatalogCase { const held = await pollUntil({ label: "Real prior active Launch checkout", deadlineAt: Date.now() + 150000, load: async () => ({ task: await ctx.input.api.get(`/api/issues/${t.id}`), runs: await ctx.allRuns() }), accept: row => gate!.calls.length > 0 && row.runs.length === 1 && row.runs[0]?.status === "running" && row.task.executionRunId === row.runs[0]?.id }); - await ctx.input.evidence("chat-catalog-project-defaults-prerequisite-timeout.json", { heldWorkerId: worker.id, heldWorkerTimeoutSec: 420, canonicalTimeoutSec: 120, futureWorkerTimeoutSec: 120, canonicalStageBoundMs: 150000, futureProbeBoundMs: 180000, attemptBoundMs: 660000 }); + await ctx.input.evidence("chat-catalog-project-defaults-prerequisite-timeout.json", { heldWorkerId: worker.id, heldWorkerTimeoutSec: 420, canonicalTimeoutSec: CHAT_NATIVE_TIMEOUT_POLICY.savedAgentTimeoutSec, futureWorkerTimeoutSec: CHAT_NATIVE_TIMEOUT_POLICY.savedAgentTimeoutSec, canonicalStageBoundMs: 150000, futureProbeBoundMs: 180000, attemptBoundMs: 660000 }); return { project: p, goal: g, checklist, task: held.task, worker, futureWorker, activeRunId: held.runs[0]!.id, originalEnv: (p as Entity & { env?: unknown }).env ?? null }; }, gradeProjectDefaults); return { ...base, expectedRunCount: 3, minimumRunCount: 3, attemptTimeoutMs: 660000, runPolicy: { kind: "source-settled-with-active-work", maximumActiveWorkerRuns: 1 },