How this place works
Methods
Two habits carry the whole site: an experiment's analysis is fixed before its data exists, and a historical claim never travels without its evidence label.
How experiments are run
- Preregistration
- The hypothesis, the test statistic, the significance level, and the milestone schedule are filed publicly before collection opens. They cannot be edited while the register is open — only appended to, with the revision logged.
- A named chance baseline
- Every experiment states what pure luck produces. Without that number a result means nothing, and it is the first thing every chart draws.
- Milestone reads
- Conclusions are drawn only at preset sample sizes. Watching a running total until it looks interesting and stopping there is how noise gets published.
- Correction for multiplicity
- Run twelve comparisons at the usual threshold and roughly one will look remarkable by luck alone. Where we test many groups, the correction is applied and both the raw and corrected results are shown.
- Sealed entries
- Same-day submissions stay hidden until the window closes, so no participant is anchored by the crowd.
- Open records
- Anonymised raw data and the analysis code are published and refreshed on the same schedule as the dashboards. Check our arithmetic.
How historical claims are labelled
Every connection drawn in a lineage carries one of six labels. The label does the arguing so the prose can stay calm — and so a documented transmission never gets to borrow the authority of a firm one.
Documented influenceA traceable line — texts, apparatus, or people demonstrably carried the practice forward.
Shared techniqueThe same method on both sides of the divide; transmission is likely but not fully papered.
Historical coexistenceSame rooms, same decades; influence unproven.
Later reinterpretationA modern meaning projected back onto the old practice.
DisputedSerious scholars disagree. Both readings are shown.
Popular mythWidely repeated; contradicted by the record.
What a sample can and cannot see
Sample size sets the smallest real effect an experiment could notice. These are computed for the blind horoscope design against its 8.33% chance baseline, at 80% power — not illustrations.
Smallest detectable hit rate by sample sizep₀ = 8.33%
| Sample size | Smallest detectable rate | Lift over chance |
|---|---|---|
| 500 | 11.8% | +3.5% |
| 1,000 | 10.8% | +2.4% |
| 2,500 | 9.9% | +1.5% |
| 5,000 | 9.4% | +1.1% |
| 10,000 | 9.1% | +0.8% |
| 25,000 | 8.8% | +0.5% |