Isolated unit tests hide integration-only conservation bugs in tick pipelines
A depletion system passed every unit test but violated conservation in the real loop: the tick that MINTS extraction ran before the tick that DRAWS down the reserve, on mismatched cycle accounting (one used integer completed-cycles, the other fractional days/duration), so a finite deposit minted ~25pct more than it ever held. Invisible because every test exercised the draw function ALONE, never alongside the mint. Two lessons: (1) any invariant that spans two pipeline stages must be asserted by an INTEGRATION test running both in real order, not a fixture; (2) an adversarial reviewer prompted specifically to find what passing tests MISS surfaces exactly this isolated-vs-integrated gap. Fix was ordering (deplete immediately before mint) + matching the exact integer-cycle count + rationing so output never exceeds remaining stock.