The trial of the trial

The
calibration
ledger.

A verdict is worthless until someone grades it. This is where we grade ours, in public, on a schedule set before the first company was admitted.

What the ledger records

Every verdict, followed to month eighteen, whatever it did in between.

Verdict issued

GO, GROW, PARK or KILL, with the registered thresholds that produced it.

Predictor version

Which version of the buyer-access predictor admitted the company.

Month 9 verdict

Durability. Did the customers stay and did the motion hold.

Month 18 state

Operating, acquired, raised, wound down, or dormant.

Grade

Whether the verdict was right, and if not, in which direction it was wrong.

Interpretation rules

Published before
the first company
is admitted.

01
Both error directions are published

A GO that died is a false positive. A KILL that thrived is a false negative. Reporting only the first makes a strict grader look accurate and an honest one look bad.

02
The sample size is stated with every claim

Fifty companies can reject a strong effect and cannot establish a subtle one. We report ranges and claim nothing the sample can't carry.

03
Thresholds cannot move after the run

Registered criteria are frozen at day zero. If a threshold was wrong, the next version changes it, and the ledger keeps the old one visible.

04
Declined applicants are tracked with consent

The comparison group is self-selected, which biases it. We name the bias rather than correct it away.

05
The ledger publishes annually, on a fixed date

Not when the results are flattering. A publication schedule chosen after the data arrives is not a schedule.

The other half

Calibration says
how well we read.
Kill conditions say
when to stop.