Hypothesis log

A template for running continuous discovery: capturing assumptions as falsifiable hypotheses, choosing the cheapest test that can settle each one, and recording the verdict — validated, killed or inconclusive.

The log is created during team mobilization, empty, alongside the backlog. It fills as soon as the team starts forming opinions about what to build — and gets its first real harvest during strategic event storming, where a fresh event map exposes a room full of assumptions at once. The product manager and product designer own it; the tech lead joins for feasibility spikes.

Writing a hypothesis

A hypothesis is not a feature request or a to-do. It’s a claim about the world that could turn out to be false, written so the team knows in advance what evidence would change its mind.

FieldWhat goes here
HypothesisThe falsifiable claim, in the form above. One claim per entry — compound hypotheses can’t be settled.
SourceWhere it came from: an event-storming hotspot, a customer request, production telemetry, a support theme, a stakeholder’s conviction. Sourcing matters — patterns in where wrong assumptions originate are worth noticing.
RiskWhich of the four product risks this addresses: value, usability, feasibility or viability. Names the owner implicitly — value and viability sit with the product manager, usability with the product designer, feasibility with the tech lead.
TestThe cheapest activity that can settle it: customer interview, clickable prototype, fake door, concierge test, feasibility spike, telemetry query. If the cheapest available test is “build it and see,” say so explicitly — that’s a signal, not a plan.
CostTime-box for the test. Most entries should read in hours or days. An entry measured in weeks needs a cheaper test or a smaller claim.
VerdictValidated, Killed, Inconclusive or Testing — plus the date and a one-line summary of the evidence.
OutcomeWhat changed because of the verdict: a candidate feature entering increment selection, a value map update, a replacement hypothesis, or nothing at all (which is a legitimate outcome for a killed idea).

Verdicts

VerdictWhat it means, and what happens next
TestingThe test is running. Entries shouldn’t sit here long — if an entry has been “testing” for more than a couple of weeks, the test was too expensive.
ValidatedThe evidence supports the claim. The candidate feature carries that evidence into value mapping and increment selection. Move the entry to Hypothesis history once the work is selected.
KilledThe evidence contradicts the claim. Remove the candidate from consideration, note what you learned and — where the test surfaced a better idea — write the replacement hypothesis immediately. Move to Hypothesis history.
InconclusiveThe test didn’t settle it. Decide deliberately: sharpen the claim and re-test, or accept the risk and note that the decision is being made without evidence. Don’t let inconclusive entries drift — an unexamined “maybe” behaves exactly like an unexamined “yes.”

Hypothesis log

VerdictHypothesis
Testing
2026-03-04

We believe brokers lose significant time to one-at-a-time document upload because the intake tool has no batch capability. If we add multi-file upload, we expect median intake completion time to drop by a third. We’ll know we’re wrong if watching brokers work shows the time is going somewhere other than the upload step.

  • Source: broker feature request, raised three times in two months
  • Risk: value
  • Test: watch three brokers complete a live intake — 90 minutes each
  • Cost: 1 day
Validated
2026-02-19

We believe first-time applicants abandon the application at income verification because they don’t have documents on hand. If we let them save and resume from a link, we expect completion rate for started applications to rise above 60%. We’ll know we’re wrong if abandonment stays flat or applicants don’t return to the link.

  • Source: production telemetry — 41% drop-off at that step
  • Risk: value
  • Test: fake door — a “save and finish later” link that emails the applicant
  • Cost: 2 days

Evidence: 68% of applicants who used the link returned within 48 hours. Outcome: save-and-resume selected into the next increment.

Hypothesis history

Settled hypotheses are retained here. The history is worth keeping: it shows the team what it used to believe, and a run of kills from the same source is a useful signal about where assumptions come from.

VerdictHypothesis
Killed
2026-01-30

We believe claims handlers want a configurable dashboard because they’ve asked for “more visibility.” If we ship a widget-based dashboard, we expect daily active use above 70%. We’ll know we’re wrong if handlers can’t articulate what they’d put on it.

  • Source: stakeholder request (operations director)
  • Risk: value
  • Test: five interviews, prototype walkthrough
  • Cost: 3 days

Evidence: none of the five could name a widget they’d add. All five described the same single missing number — current queue age. Outcome: dashboard dropped. Replacement hypothesis written and validated the same week; the number shipped in the following increment as a header field.