Per-issue resolve bound is per call, not per land #181

New issue

Closed

opened 2026-08-16 01:18:55 +00:00 by john · 1 comment

john commented

2026-08-16 01:18:55 +00:00

Owner

From the phase 2 risk_surface review (medium, adjudicated survivor).

issue-check.mjs stops the resolve loop only when a call comes back timedOut. A tea that answers slowly and exits nonzero is never marked timed out, so each of the capped resolves can burn nearly the full call timeout. With the default bound and five unanswered references that is roughly five timeouts of latency on the land path, which is the multiplication the loop break was written to prevent.

Fix direction: bound the resolve phase on total elapsed wall clock, not on a per-call flag.

Found in: cadence-core/bin/issue-check.mjs (the resolve loop after partitionIssues)

From the phase 2 risk_surface review (medium, adjudicated survivor). `issue-check.mjs` stops the resolve loop only when a call comes back `timedOut`. A `tea` that answers slowly and exits nonzero is never marked timed out, so each of the capped resolves can burn nearly the full call timeout. With the default bound and five unanswered references that is roughly five timeouts of latency on the land path, which is the multiplication the loop break was written to prevent. Fix direction: bound the resolve phase on total elapsed wall clock, not on a per-call flag. Found in: cadence-core/bin/issue-check.mjs (the resolve loop after partitionIssues)