Every investigation cause carries a posterior between 0 and 1, which maps to a tier:
| Tier | Posterior | | --- | --- | | Ruled out | below 0.1, with a contradicting test | | Possible contributor | below 0.5 | | Likely contributor | 0.5 to below 0.75 | | Strong contributor | 0.75 to below 0.9 | | Confirmed | 0.9 and above |
The word "confirmed" is reserved for the top tier, and only when the calibration ledger supports it: the 0.9 to 1.0 bin needs at least 20 resolved cases and an observed hit rate of at least 0.85.
Ceilings
Some kinds of cause can never reach the top tier because the mechanism is not directly observed. A Google update is capped at "likely" and seasonality at "strong".
Estimate tiers
Gain ranges and forecasts use a separate scheme from backtests. With fewer than 15 backtested cases the range is unrated. A range whose realised coverage is below 0.6 is weak. Strong needs coverage of at least 0.8 and at least 30 cases.
This page reads its numbers from the same constants the product uses.