Why Did the Tests Pass?

Where the evidence lived: the running world. From docs/unedited/2026-07-30-100641-local-command-caveatcaveat-the-messages-below-w.txt, around line 1384.

The agent had just shipped a fix for a stuck harvester, behind green checks:

Three witnesses. All passed.

• paradox-check: Spec is consistent, Z3 signed off
• typecheck: Shell compiles clean against generated types
• fsm-behavior: All 11 assertions pass

The fix: one line in the spec (collecting + spawnFull → upgrading), one line reorder in the shell (spawnFull checked before storeFull), and a behavioral test that proves both the bug and the fix. Redeploy when you’re ready.

I was watching the game.

upgrader-2216 only carries 2 energy at a time

i always start clean, stop second guessing me

i started clean

My bad. The event priority reorder was too blunt — it makes spawnFull fire for the upgrader on every tick once the spawn fills up, so it transitions to upgrading with just 2 energy instead of filling up first.

why did the tests pass?

The tests passed because the FSM behavioral test only tests the transition functions — it verified collecting + spawnFull → upgrading, which is correct and still needed. It doesn’t test emitEvent, which is where the bug was. The test was incomplete — it proved the spec fix but didn’t catch the shell-level event emission problem.

Three green checks, and the creeps misbehaving on screen anyway. The world had the last word, and the question that opened it cost four words.