Everything about the Fault-Line Radar in one place — what it forecasts, how it scores, how evidence gets in, and where the pressure is building right now.
The radar forecasts where legal AI is headed by tracking fault lines — the places a new AI capability rubs against an existing legal duty. It does not predict dates. It measures pressure, weighted by who is doing the pushing.
Five meters each, 0–10. A fault line "fires" when it crosses 7. Sorted by how much pressure is building.
| Fault line | The fix (control) | Cap | Rule | Adopt | Enable | Soft |
|---|---|---|---|---|---|---|
| verification | Documented verification + trace logs | 8.4 | 10.0 | 8.6 | 9.0 | 5.6 |
| competence | Firm-wide training + governance program | 6.6 | 9.9 | 8.0 | 8.4 | 5.0 |
| disclosure | Verification-by-default with an audit trail | 5.5 | 9.9 | 7.7 | 7.4 | 5.5 |
| convergence | One operating model, strictest standard | 2.5 | 9.8 | 7.7 | 4.0 | 5.5 |
| insurance | Governance as an insurable artifact | 2.5 | 8.7 | 8.4 | 6.9 | 4.0 |
| confidentiality | Data-flow mapping + vendor attestation | 4.5 | 9.9 | 7.0 | 6.7 | 5.5 |
| benchmark | Tool-certification benchmark (NERVE) | 9.0 | 9.1 | 7.6 | 8.9 | 7.5 |
| agentic | Human-decides gates + scope-limiting | 6.9 | 9.4 | 6.2 | 6.5 | 8.3 |
| vendor liability | Tool-provenance documentation | 2.5 | 9.1 | 2.5 | 3.0 | 3.8 |
| judge analytics | Guardrailed strategy sim, no actor prediction | 4.0 | 6.4 | 3.0 | 3.0 | 3.0 |
| fees | Value-delivered measurement | 2.5 | 5.8 | 3.0 | 3.5 | 3.0 |
Authority over volume. Every piece of evidence is stamped with a tier when it enters, and the tier decides the weight — not how many people repeated it.
| Tier | What it is | Weight |
|---|---|---|
| T1 | Court ruling, statute | 1.00 |
| T2 | Bar opinion, regulatory guidance | 0.80 |
| T3 | Empirical study, institution | 0.60 |
| T4 | Commentary, trade press | 0.35 |
| T5 | Vendor marketing | 0.008 |
A vendor blog carries about a hundredth of a bar opinion. Ten vendor posts cannot outweigh one ruling. Two things keep the scoring honest: the L2 ruling meter is two-sided (an item can argue against a line and drag it down, not just up), and every capital event is flagged continuation / origination / direction-shift — the momentum signal only predicts continuation, so a first-of-kind move is labeled rather than scored as if it were predictable.
A fact does not enter the feed because someone found it. It enters because it survived being argued against.
This is the step that caught real mistakes: a court rule only one source mentioned, a "disqualification" that was actually a reprimand, a claim about two vendors no source could hold up.
Seventy-six dated, cited items flow from four buckets into eleven fault-line predictions. The number beside each prediction is how many streams corroborate it — more streams, more confident call. Full interactive version: the streams page.
No insurer has yet made a certified AI tool a condition of malpractice coverage, and no bar has adopted a named benchmark. When one does, that is the event that flips the method layer from voluntary to enforceable. That absence — the thing that has not happened yet — is the single most valuable signal in the whole system.