ClearDraft LogoClearDraft
GENERAL
DashboardTo-do ListImportsLive MailboxIMAP
MY SPACES
EvaluationRL AuditRLHFSettings

RL Audit & Precedent Memory

Tamper-evident audit ledger of operator corrections, taught equivalence precedents, and in-context prompt policies (ADR-006 and section 19 compliant).

Reinforcement Learning & Precedent Memory (RLHF Audit)

Tamper-evident audit ledger of operator corrections, taught equivalence precedents, and in-context prompt policies.

ADR-006 and section 19 compliant
Audited Human Decisions0Logged in SQLite review_events
Taught Precedents0Active equivalence conventions
Prompt Memory Policies0Versioned few-shot example sets
Foundational Weight Drift0.00%Weights frozen; memory in-context
Pending Corrections0Awaiting curation into a versioned set
Audited Human Actions & Memory InjectionsShowing 0 of 0 records
Table: review_events (Append-Only)
TimestampCase / SourceActionFieldValue Shift (Old ➔ Taught)Operator RationalePrecedent
No audit events matching current criteria.

Active Learned Equivalence Conventions (0)

Precedent memory injected into comparison stages as human review aids

Correction Candidates Awaiting Promotion (0)

Operator corrections do not automatically inject into models. Human rationales accumulate as candidates here, which are curated and frozen into an immutable versioned set pinned to policy_version.

No corrections awaiting promotion. Correcting a case's category on the case page adds a candidate here.