Did the working model survive reality?
Founding testers provide the initial before/after baseline. Purchased reports now add private 7-, 14-, and 30-day checks that record action, observed outcome, usefulness, and whether the model was supported, weakened, or disconfirmed. Only manually reviewed records count toward public claims.
Collecting 20 verified outcomes; no claim yet
We will not publish averages, recommendation scores, or selected quotes before 20 founding-cohort outcomes have been manually reviewed. Longitudinal paid-report results use a separate, stricter gate of 50 verified follow-ups per product and checkpoint. Verified means machine-checked, not adjudicated: the answer arrived through a valid private evidence link for a report that already existed, on or after the scheduled checkpoint date. It establishes that the prediction was on record before the outcome was reported — it does not mean anyone confirmed the account of what happened. That part remains self-report, and no amount of it becomes a causal claim.
50 verified follow-ups per product and checkpoint
Day 7, 14, and 30 are evaluated separately for Clarity and MindScan. Empty or sub-threshold cells remain unpublished; we do not combine products or checkpoints to manufacture a stronger-looking sample.
Every cell reports how many could have answered, not only how many did. A failure rate over volunteers is uninterpretable on its own, and the people whose model failed worst are the least likely to come back — so the non-response is shown rather than left out. All four verdicts are published, including the intermediate one: counting only outright disconfirmation would understate failure by construction. “Not enough evidence” is an honest non-answer and is excluded from any denominator rather than folded into either side. The eligible count is a rolling cohort; attempts leave it when their 90-day retention expires.
No longitudinal cell has reached its publication threshold. No outcome rate is claimed.
What this can show
Whether a self-selected cohort reports action and continued usefulness, and whether the original working model is sometimes weakened or rejected after real observations.
What it cannot show
Causation, clinical efficacy, population norms, decision correctness, or how people who never applied would respond.
Pre-declared integrity rules
- 1. Show absolute sample sizes beside every average.
- 2. Include neutral and negative outcomes; do not publish only favorable responses.
- 3. Verify eligibility before an outcome enters the aggregate.
- 4. Treat quote permission separately from participation.
- 5. Record material product changes before comparing cohorts.
- 6. Read every count as a rolling window, not a running total. Follow-ups are deleted with their scan 90 days after unlock, so a published number can fall as older records expire — including after it has been cited. A figure quoted from this page is only accurate on the day it was read.
Interested in testing the product honestly? Apply for the founding cohort.