It Passes Locally

Where the evidence lived: CI. From docs/unedited/2026-07-30-100641-local-command-caveatcaveat-the-messages-below-w.txt, around lines 1717 and 2066.

The compound-condition fix from why-did-the-tests-pass.md went in, with a new integration test behind it:

Two layers of tests:

1. fsm-behavior (pure Node, no runtime) — proves the transition functions produce correct states for all events.
2. itest (VM + real Screeps server) — new subtest polls Memory.stats.creeps until a harvester has event: “spawnFull” and fsm: “harvesting”. This proves the compound event logic works in the actual game runtime, not just in theory.

we have a problem upstream a failed itest from their gitlab runner. this test passes locally

The job log ended:

Exception: timed out waiting for harvester spawnFull event

The log tells the whole story. …

1. Harvester delivers to spawn. Spawn had ~280/300 energy, creep had 50. transfer() moves 20 energy (all the spawn can hold), creep retains 30.
2. Spawn is now 300/300 (full). Creep has 30/50 (partial).
3. emitEvent: storeFull = false. The compound condition storeFull && spawnFull is false.
4. Returns “tick”. FSM: delivering + tick → delivering.
5. transfer() → ERR_FULL. Nothing happens. Stuck forever.

“Works in the actual game runtime” had been checked in one world — mine. CI ran a world with different timing, reached the partial transfer my runs missed, and the deadlock surfaced. The claim fell to evidence neither of us was holding.