The Lawyer Analysis Metric: documenting human supervision in-tenant
Why the audit trail needs to record who approved a finding, who overruled it, and under which supervision policy — and what that record is for.
"A lawyer reviewed the output" is not a supervision record. It is an assertion made after the fact, by the person whose diligence is in question, about an event nobody logged. When it matters — a professional conduct question, a client audit, an insurer — it is worth approximately nothing.
The Lawyer Analysis Metric is the question of how much human review counts as *sufficient* supervision for a given task. LexCore implements it as Review Level: an explicit dial, set before the work, recorded with the work.
Supervision is a spectrum, and both ends are wrong
Review everything and the tool saves no time; review nothing and you are somewhere between negligent and sanctionable. The right point is task- and risk-dependent, which is exactly why it should be a setting rather than a habit.
| Review Level | Typical task | Effect |
|---|---|---|
| Low | Routine extraction, defined-term pulls | High auto-pass; few escalations |
| Moderate | Standard contract review | Balanced; the default |
| High | Unusual clauses, liability, novel risk | Low auto-pass; more findings escalate |
| Highest | Deal-critical, strategy | Almost everything reaches a lawyer |
Mechanically, the level sets the confidence bar a finding must clear to auto-pass. Raise it and more findings escalate to a Human Gate. The review records the level applied, so the judgment stops being hand-waved and becomes a fact about the matter.
What the trail has to contain to be worth having
- The policy in force — which Review Level, and which playbook version, applied to this run.
- The disposition of each finding — auto-passed, escalated, approved, overruled.
- The person — who cleared the gate, not merely that a gate was cleared.
- The grounding — the CitationGate verdict attached to the finding, so a later reader can re-check the claim against the source rather than trusting the approval.
The last one is what makes the record self-defending. An approval that points at a verified span can be audited by someone who was not in the room. An approval that points at nothing can only be believed.
Why in-tenant matters
A supervision record held by the vendor is a record you can lose access to at renewal. The audit trail belongs in your tenant, it is append-only, and it survives account purge on purpose — an erasure that also erased the record of the erasure would defeat the point of both.
Keep reading
Why we reject findings that fail the exact character offset test
Most legal AI products work on summaries of chunks. Here is the failure mode that introduces into an indemnities review, and how we anchored CitationGate to verbatim spans instead.
NDPA, NDPR and cross-border legal documents: what the regulation actually requires
A reading of the Nigerian data-protection position that matters to legal ops teams — which instrument is the primary statute, what a cross-border transfer is, and where an AI review sits in it.