Goal Hive resume should detect stale budget state after interrupted runs

[HamsteRider-m]

Summary

During a long Goal Hive run, the master/worker processes were interrupted while a late dispatch was still in progress. When resuming the same Hive board, the stored goal_state.json still contained the original time budget boundary. The resumed master therefore treated the run as effectively budget-exhausted and could stop/close prematurely unless the operator manually inspected and refreshed the resume budget.

Observed behavior

• A Goal Hive run was started with multiple workers and a fixed multi-hour budget.
• The run progressed through several dispatches, then the master/workers were externally stopped during a late dispatch.
• On resume, the BBS state and artifacts still existed, but goal_state.json carried the stale start_time / budget window.
• Relaunching the master without correcting that state risked immediate budget exhaustion instead of continuing remaining work.
• Manual recovery required:
  • checking goal_state.json for objective / done prompt integrity,
  • resetting the resume budget boundary after explicit user authorization,
  • watching BBS posts plus report/evidence artifacts until the final master closure post.

Expected behavior

Goal Hive resume should make stale budget state explicit and safe. Ideally, the master or resume helper should detect that a persisted start_time is from an earlier interrupted session and ask/require an explicit resume budget reset instead of silently treating the resumed run as out of budget.

Suggested hardening

• Add a resume preflight for Goal Hive that validates:
  • objective and fixed done_prompt are present and unchanged,
  • persisted budget fields are not stale relative to the requested resume,
  • BBS latest posts and artifact paths are still consistent before dispatching new work.
• Make budget-expired-on-resume a distinct diagnostic from normal budget exhaustion.
• Consider a supported resume command/flag that refreshes only the budget boundary while preserving the board, objective, and dispatch history.
• Document that final completion should be determined from master closure / BBS / report artifacts, not process liveness alone.

Related branch

I pushed a small documentation note for this operator-side pitfall to my fork so it can be rebased later:

• HamsteRider-m/GenericAgent@maygo/private
• commit ef227f1 (docs: document Goal Hive resume checks)