feedback
On a phone, the recovery instructions hide the next step below the keyboard.
demo:report:1The Spire · Decision Theater · practice case 01
A better number. A worse experience.
Recovery rose from 58% to 73.5%. Would you launch? Follow the evidence through three possible paths, find the missing denominator, and see what a fair trial changes.
Hypothetical case · no real outcome attached
Your decision
Should a product team launch its new account-recovery screen?
Practice days 1–7 · Synthetic recovery team
feedback
On desktop, the clearer recovery instructions helped me choose the right identity method.
demo:report:2measurement
Previous screen: 580 successful recoveries in 1,000 attempts.
demo:aggregate:beforemeasurement
New screen: 735 successful recoveries in 1,000 attempts.
demo:aggregate:afterChange a premise, inspect the consequence.
Changing a premise starts a fresh comparison. It cannot edit your earlier signed choice.
Deterministic suggestion · test
Aggregate improvement does not identify a treatment effect when the device mix changes.
Cohort denominators and allocation quality must be checked. Two feedback reports cannot estimate prevalence.
Compare the simple baseline and inspect sources
Simple aggregate triage: choose the larger recovery rate. Ignores allocation, cohort mix and unequal harm.
demo:aggregate:before · demo:aggregate:after · demo:report:1
Launch to everyone
Fastest reach; the overall improvement may conceal a worse mobile experience.
Based on demo:aggregate:before · demo:aggregate:after
Run a stratified trial
Compare like with like before expansion. Requires additional attempts and a named outcome owner.
Based on demo:report:1 · demo:report:2
Pause and instrument
Keep the current screen while collecting device denominators and failure reasons.
Based on demo:report:1
Blind Spot Expedition
Ask for evidence that can change your mind.
- Recovery successes AND attempts by device in both windows.
- Allocation method, exclusions and identical recovery definitions.
Two rival explanations
- The screen improved recovery.
- More desktop attempts raised the aggregate while each cohort worsened.
Frozen before the trial
Agree on success before looking.
At least 200 attempts per arm in each device cohort; a recovery gain of at least 5 percentage points in every cohort; no cohort may regress. These are an exercise decision threshold, not a statistical significance claim.
Owner: Synthetic recovery team · Practice days 8–14
Choose before the replay
What would you do next?
No account, model call, points or purchase. The receipt is in its signed address, with no new participant record in the database. It expires in 30 days. Anyone you give the address can read this choice.
Keep the future out of the first decision.
The blind packet has no trial outcome. Make a practice choice above to reveal the authored later result beside the original rule. This separation helps you inspect hindsight; it does not make a public exercise an independent evaluation.
Evidence scope and agent contract
One authored public-safe case. No external integration, private source records or real outcome attached.
Accepted recommendations, completed real experiments, failure recurrence and model quality are unmeasured. Read the same projection at the expedition API; use view=investigate for the cohort reveal. POST a JSON choice, premise and source_hash to this page to mint a practice receipt; no actor identifier is accepted as authority.
Version 2026-09-30.1 · source commitment 7799ca7378e80b9b0fe9925f664c7992f3422843f6805692b8f676975fb75448. Receipt validity requires this exact commitment. The Time Lens uses this case only; it never substitutes live state for the past.