Black defendants were seventy-seven percent more likely to be labeled higher risk for future violent crime. The algorithm made this distinction across thousands of cases.
F3
no take
—
Machine bias convergence probe
machine-bias-convergence-probe
live · 0 of 6 readywriting the scriptAwaiting approval
The board
Six blocks, one per beat. Each card shows the provider that actually ran — a fallback nobody can see in the record is a substitution, not a fallback.
Courts use risk-scoring algorithms to predict future crimes. These scores shape bail, sentencing, and parole decisions. When they fail, lives hang in the balance.
F5
no take
—
Brisha Borden took a child's bicycle and scooter. The algorithm rated her high risk. She has not been charged with any new crimes since.
F1
no take
—
ProPublica analyzed risk scores for more than seven thousand people arrested in Broward County, Florida. The algorithm was nearly twice as likely to falsely flag black defendants.
F2F4
no take
—
In twenty fourteen, Attorney General Eric Holder warned that risk scores might be injecting bias into the courts. Yet these systems remain in use today.
F5
no take
—
The numbers have not changed. The systems that produce them have not changed. The people they harm keep waiting for justice to catch up.
no claims mapped
no take
—
Run log
Every submission, rejection, retry and take measurement, newest first. This is literally the audit trail rendered.
- 11:56:50script6 blocks written
- 11:56:50decision.scriptpass — 6 blocks, every claim traced to a fact after 1 repair pass(es)
- 11:56:06decision.scriptreject — block 3: POL-5 — 22 words (need 23-27). Estimated ~8.5s. | block 4: POL-5 — 31 words (need 23-27). Estimated ~11.9s. (after 8 attempts — surfacing rather than retrying further) What to do: retry — refusals are free and the model redrafts.