Quality Mechanisms

Quality Mechanisms

Epistemic quality at the commitment boundary — each mechanism stress-tests or improves the read before bind (distinct from legitimacy). Entries link to their canonical pattern or term.

Index collection — each row links to the canonical pattern or term entry; nothing is duplicated here.

13 of 13 entries

Blind Peer Review

Independent second read before bind; override rate as quality signal

AI Stress Testing

Adversarial challenge to corners of the read; full disclosure required

Disclosed Model Disagreement

Preserve disagreement signal — do not pre-aggregate

Experiment Architecture

Pre-committed design, separate streams, reproducible proof of claims

Logical Inference

D-F execution — compute or replay from designated evidence, do not vote

Committed Provenance Chain

Pin inputs, code, environment — claims must be replayable

Calibration Audit

Track record, Brier scores, override rates — honest standing for the bind

Manual Recency Sampling

Human audit of recent auto-binds at volume

Seeded Review Cases

Known cases injected to detect drift and rubber-stamp approval

Recorded Delegation

Auto-bind only with pre-committed calibration evidence and policy ref

Closed-Loop Control

Outcome feedback to comparators — MOE not MOP theatre

Refutation Propagation

Demotion travels to dependents — quality feedback at scale

Processor Fidelity

Bind carries honest epistemic tier; no tier collapse at commit