// AI governance · Core distinction

Explainability versus replayability

Explainability addresses a model. Replayability addresses the institutional decision made around it.

Explainability asks why a model produced or tends to produce an output. Replayability asks whether a specific decision can be reconstructed: what was authorised, which data and rules were in force, where judgement entered, what executed, what outcome followed and where any material divergence occurred.

// 01

Four different records

Monitoring

Establishes what a system did, while it was doing it.

Explainability

Establishes why a model produced a particular output.

Audit trails

Establish which events occurred, and when.

Replayability

Establishes whether the decision the institution authorised is the decision that was executed.

// 02

A concrete example

Knight Capital: a technical explanation could show that dormant Power Peg code generated unintended orders. Replayability asks the institutional question: after the authorised client order was complete, why did the next order retain execution power, and where was the evidence that could have stopped it? The difference is mechanism versus authority.

Replayability does not require an identical model output to be reproduced. It requires enough connected evidence to establish whether the decision executed was the decision authorised.

// 03

Why boards need both

Explainability can support validation, challenge and recourse. Replayability connects that explanation to policy, authority, rules, judgement, execution, outcome and correction. An institution may therefore have an explainable model inside an unreplayable decision chain.

Read the canonical definition of decision replayability.

// Questions people ask

Common questions

Is replayability another name for an audit trail?

No. An audit trail records events. Replayability connects those events to mandate, authority, state, judgement and outcome.

Can a generative model be replayable if outputs vary?

Yes. The governance standard concerns sufficient evidence to reconstruct the decision chain; it does not depend on deterministic reproduction.

Does replayability replace explainability?

No. Explainability remains useful. Replayability connects it to the wider institutional decision.

// The practical test

Test one institutional decision

The ten-day Decision Drift Audit™ maps one material decision across all eight layers, assesses replayability and authority boundaries, and delivers one prioritised board finding.

Test one decision →Canonical definitions →Decision Integrity Chain™ →