evnchn-agentic 亡羊補牢 — lost the sheep, repair the fence
亡羊補牢 (wáng yáng bǔ láo) — "lost the sheep, repair the fence". A sheep escaped through a hole in the fence. Two responses: (a) chase the sheep, ignore the hole; (b) chase the sheep AND repair the hole. The chengyu names (b).
The full saying in modern usage: 亡羊補牢，猶未為晚 — "to mend the fence after losing a sheep is still not too late."
When this applies
Symptoms that this schema applies:
- A bug shipped to prod and was patched — but the class of bug could shipped again next week
- A test was failing and was fixed — but the missing test that would have caught it earlier is still missing
- A deploy was rolled back — but the deploy-process gap that allowed the bad change through is still there
- Data was lost — but the backup gap that made recovery hard is still there
- A security incident was contained — but the detection gap that let it go unnoticed for N days is still there
- A user-reported bug was fixed — but no test or invariant was added; the next agent could re-introduce it
The shared shape: the immediate damage got attention; the systemic gap that allowed it is the load-bearing fix.
Behavior
- Address the immediate damage first. The sheep is gone; whatever can still be saved should be saved. Don't moralise about the fence while the situation is still leaking.
- Then name the gap. What fence-hole let this happen? Be specific — "no integration test for this code path" / "no monitoring on this metric" / "no documented invariant for this constraint".
- Propose the fence-repair concretely. Add the test, add the monitor, write the invariant down, add the deploy gate. Not "we should be more careful" — concrete artefact.
- Pitch the repair to the operator. It has a cost (the test takes time to maintain; the monitor adds infra; the gate slows deploys). Operator decides whether the fence-repair is worth the cost; they have context you don't.
- If the cost-benefit is negative, name THAT explicitly. "Repair would cost X, payoff is Y, recommending we accept the risk and not repair." A decision recorded > an oversight repeated.
Anti-pattern
Patch and move on. Fix the immediate symptom, don't surface the gap, don't propose the repair. The class of failure recurs next month; the same fix gets applied again.
Repair without addressing damage first. Building the fence while the sheep is still leaking. Address the urgent first, the structural second.
Over-fencing. Building an enterprise-grade fence in response to a $10 sheep. The cost-benefit must be honest; not every escape warrants a re-architecture.
Performative post-mortem. Writing a post-mortem doc with no actual change. The doc isn't the repair; the change is the repair. (Though the doc may be part of L3 of chengyu-retreat-is-best — making the lesson durable across sessions.)
Relationship to other chengyu
| Chengyu | When in time | What it does |
|---|---|---|
| chengyu-remove-firewood-from-cauldron | BEFORE the failure | Trace upstream from symptom, fix root cause to prevent |
| chengyu-lost-sheep-repair-fence | AFTER the failure | Fix the systemic gap so the class of failure stops recurring |
Same root-cause-orientation, different timing. The combination is the full root-cause discipline.
Etymology
From the Strategies of the Warring States (戰國策). A shepherd noticed a hole in the fence after losing one sheep. Friends told him to leave it — the sheep was already gone. He repaired it anyway, and no more sheep escaped.
The discipline: the cost of repair is paid once; the cost of not repairing is paid every time the failure recurs.