Week 9 Session: Verifying You Didn't Just Move the Problem

How do you know an intervention actually resolved the cause, rather than relocating or masking the symptom? This session's fix looks like it worked — closer verification is what the week is actually testing.

26 April 2027

Before the session#

This session is the second half of Week 8’s problem: a scoped intervention is only meaningful once you can show it actually did what it was meant to.

The incident#

Last week’s fix to fulfil — restoring reserved stock earlier in the cancellation path — is in place. The negative-stock reports have stopped. An existing smoke test that cancels and immediately resubmits an order now passes where it used to fail. Both of those are real observations. Neither is the claim the diagnosis actually made.

What the smoke test proves, and doesn’t#

The diagnosis’s causal claim was specific: the race between cancellation and resubmission — restoration running too late relative to re-allocation — has been closed. The smoke test proves something narrower: that one particular cancel-then-resubmit sequence, run once, at the timing the test happens to use, no longer shows the symptom. A race condition can pass a single deterministic test run while remaining open under a slightly different timing or a higher-concurrency resubmission than the test exercises. “The report is clean” and “the race no longer exists” are different claims, and only one of them is what the diagnosis promised.

In the session#

You write a verification check tied to your diagnosis rather than to the symptom’s disappearance, apply the intervention, and show that your check would have flagged it had the intervention merely masked the original failure instead of resolving it.

Afterwards#

Assessment 3 is due around now, and shares this week’s central demand: verification tied to a diagnosed cause, not to the absence of an error message.