Week 2 Session: A Hypothesis That Could Be Wrong

What makes a hypothesis useful, as opposed to merely plausible? A reproducible failure with two similar-looking causes forces a real choice, and the choice has to be committed to before it's tested.

1 March 2027

Before the session#

Bring the observation log from Week 1’s system-reading habit — this session reuses it as the starting material for hypothesis formation, on a slightly larger system.

The incident#

Northfield Mercantile’s nightly batch reconciler — reconcile-batch, three files, one entry point — compares a night’s processed orders against the accounting ledger. Out of roughly 500 orders in last night’s run, three show a total that doesn’t match the ledger’s figure for the same order. All three mismatched orders happened to be placed within the same ten-minute window, and all three used a discount code. Two candidate explanations are already circulating among the team before you arrive: the reconciler is reading a currency-conversion rate that hadn’t refreshed yet, or it is double-counting a line item on orders that were partially refunded. Both are consistent with what’s known so far. Neither has been tested.

Two guesses, one test#

Both circulating explanations would produce a small numeric discrepancy, consistent with the mismatch count, the shared time window, and the shared use of discount codes. Consistency is why neither can be preferred yet: it’s the bar both already clear. What’s missing is a test result that only one of the two explanations could survive — and nobody has proposed one before now.

In the session#

You submit a hypothesis and the specific, named test that would disconfirm it — before you’re allowed to run that test — then report what actually happened. A guess that survives contact with its own disconfirming test is treated differently from one that doesn’t, and so is a guess that gets quietly reworded after the result comes in.

Afterwards#

Revising a hypothesis on genuine disconfirming evidence is graded as successful investigation, not as having gotten it wrong the first time. You leave this session able to commit to what would prove a hypothesis wrong before you know the answer — the habit next week’s structural reading has nothing to fall back on without.