fix(runtime): name the group pin a node timeout pin falls back to (#675)

[aviggiano]

Follow-up to #1256 (issue #675).

Problem

• If a node pins timeout_seconds inside a group that also pins, the TOPOLOGY_TIMEOUT_SHADOWS_DEFAULT warning says "Raise or remove the pin to use the longer default." Removing the node pin falls back to the group pin, not to the profile or run default. Example: group pin 7200, node pin 3000, run default 3600. The warning compares the node pin against 3600, but removing the node pin gives 7200.
• The docs and CHANGELOG say validate, plan and run report the warning. There is no user-facing plan command. doctor does show it, because it reports the validation diagnostics.
• The CHANGELOG says goal nodes kept timing out at two hours because of the 7200 s pins. The maintainer's root-cause comment on goals/strategies group timeout_seconds: 7200 shadows the profile default and kills high-reasoning nodes at 2h with total loss of node work #675 says the timeouts came from the oversized goal statement (Goal planner inlines entire JSON records into the goal prompt, making each goal statement a multi-thousand-token wall #680/fix(goals): shrink goal prompts from inlined JSON to path references, and stop killed goal lanes being attested as verified empty results #690). The part that still holds is the resolution order: a pin shadows a longer default.
• No test covered a pin equal to the default (the <= boundary). No test covered a node fanned out to several model profiles with different timeouts.

Change

• timeout-shadowing.ts: for a node pin whose group also pins, the message now ends with "Raise the pin, or remove it to fall back to groups.<g>.defaults.timeout_seconds=." Other pins keep the old text. The comparison and the set of reported pins do not change.
• The doc comment in timeout-shadowing.ts and the CHANGELOG now keep only the resolution-order claim.
• docs/reference/configuration.md and the CHANGELOG now say validate, doctor and run.

Tests

• New: a run default equal to the packaged 7200 s pins reports no warning.
• New: a node fanned out to claude (5400), deepseek (10800) and default (run default 3600) with a 1800 pin is reported once, against model profile deepseek (10800).
• Extended: the plan test has a node pinned at 3000 inside a group pinned at 7200. Its warning names the group fallback. The other warnings keep the old remediation text.
• To check that the tests catch bugs, I ran them against broken versions of the source:
  • Changing <= to < fails the boundary test.
  • Breaking the largest-default selection fails the fan-out test.
  • The code as merged in feat(runtime): warn when a topology timeout pin is below the default it overrides (#675) #1256 fails the extended plan test.
• Ran: runtime tests matching timeout|doctor and ^validate|^plan , eslint and prettier on changed files, pnpm --filter @ultrafuzz/runtime typecheck, ESLINT_PLUGIN_DIFF_COMMIT=origin/unstable pnpm -w lint:strict:ci, pnpm -w docs:check.

🤖 Generated with Claude Code

Confidence Score: 5/5

The PR appears safe to merge; no actionable issues were identified.

Summary

The PR corrects the timeout-warning remediation for node pins inside pinned groups, updates the documented commands and root-cause wording, and adds equality and model fan-out tests.

_{Reviews (1) · Last reviewed commit: "fix(runtime): name the group pin a node ..."}