How it answers — by matching, not guessing
It checks your case againstForcing authority (Martinez, Trevino, Coleman, Lee v. Kemna) + the ground courts actually reversed on
It gives you backthe closest real example that fits — and how close it is
If nothing fitsit says “no match” — it will not invent one
reference=material-injustice-lead-detector · receipt=c88ce7fe4872c12c · the same question always gives the same answer · no AI guessing in the answer
It never bluffs. If your case doesn’t line up with
something real and known, this engine tells you “nothing fits” and stops. A clear
“no match” is a real answer — often the most valuable one, because it’s
honest. That’s the whole point.
Why matching beats guessing. We have trusted machines to match — fingerprints, DNA, x-rays, radar — in the decisions that touch people’s liberty and lives, far longer than we have trusted them to make something up and hope. Material Injustice Lead Detector stands in that older, safer tradition.
How a person uses it
1. Enter a casepaste the opinion or record text — it runs on the box, nothing is uploaded
2. The gate scores itclaim detectors fire only on the record’s own words; the holding window is read for the ground the court acted on
3. Verdict cardEMIT (a lead + its forcing authority) · ABSTAIN (nothing earns a call) · REQUIRE (declare the posture) · FLAG (declared disposition contradicts the record)
4. Receiptsha256 of the engine and of the input — the same text always gives the same leads
Validation — reproduced by running, not quoted
Post-conviction remedy detectorscored against the ground the court actually reversed on (1,435 full-text opinions; resolved reversed set N=50): top pick 74% · top-2 96% · top-3 100%. Per-claim precision reported where ≥5 gold exemplars; ABSTAIN where fewer.
SCOTUS forcing-gate93.8% precision / 8.7% coverage on the validation set (re-run 2026-10-11: 134/143 committed calls correct, N=1,453). About 1 in 16 committed calls is wrong — which is why every call is a LEAD to verify, never a verdict.
Method — the anchored method, applied
Anchorfixed primary sources only: the record, the rule text, the U.S. Reports holding
External sensorverbatim-quote verification with a negative control — a self-report is never the sensor
Correction loopverify-or-abstain on every element; a correct “no” is part of the product
Rule-face onlyoutput names rules and doctrines (Martinez, Trevino, Coleman, Lee v. Kemna), never people; case-face output stays with the litigant
Deterministicno model in the read path
Intellectual property on file (USPTO provisionals, micro entity, pro se)
Binary Collapse Cascade / Hilbert-Odù detector63/958,936 · Zenodo DOI 10.5281/zenodo.18250410
Sterling-Ratio cost-externalization64/061,831
Guardian distributed missing-persons (zero retention)64/066,209
Forensic Signal Validation (reflectance + XRF)63/967,246
assignees: Black Swan Technologies 501(c)(3) / EJ Audit Hub · ten provisionals on file; titles shown are the method-adjacent ones