A Defensible, Tamper-Evident Record of Every AI-Assisted Review

Objective

Defensibility in review and e-discovery means a consistent, demonstrable process — not an ad-hoc one that varies by reviewer. This recipe records every AI-assisted review in a tamper-evident immutable table, so you can show exactly what was flagged, by which model, and how the reviewer decided.

Step 1: An immutable review-audit table

CREATE IMMUTABLE TABLE review_audit (
    doc_id        TEXT,
    model_version TEXT,
    flags         TEXT,     -- risky clauses / responsive hits, cited to source
    reviewer      TEXT,
    decision      TEXT
) WITH (ENCRYPTION = 'AES256GCM');

Omit WITH (ENCRYPTION = 'AES256GCM') if AIDB_IMMUTABLE_MASTER_KEY is not set; the table is still append-only and hash-chained.

Step 2: Record a review

INSERT INTO review_audit VALUES (
    'DOC-MSA-2026-14',
    'clause_risk_v3',
    'uncapped liability (clause 8.2); auto-renewal (clause 12.1)',
    'M. Okafor, counsel',
    'flagged for negotiation'
);

Step 3: Prove the process was not altered

VERIFY TABLE review_audit;

is_valid | message
---------+----------------------------------------------------------------
true     | chain verified: 1 record(s) ... per-record lineage anchored
         | (tamper-evident); ...

Why it matters

Every review — the document, the model version, the flags, the reviewer's call — is anchored in an append-only chain you can verify with one query. That is the defensible, consistent process e-discovery demands, and the audit layer under clause-risk review and the obligation graph.