# Audit Replay: The Forgotten Half of Reproducibility

DevFeed: [Audit Replay: The Forgotten Half of Reproducibility](<https://devfeed.tech/articles/audit-replay-the-forgotten-half-of-reproducibility-93922.md>)

Original publisher: [Read original article](<https://hackernoon.com/audit-replay-the-forgotten-half-of-reproducibility?source=rss>)

Author: Karan Sehgal, MSc

Published: 2026-10-21T19:46:20Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [audit](<https://devfeed.tech/topics/audit.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>), [reproducibility](<https://devfeed.tech/topics/reproducibility.md>), [recording](<https://devfeed.tech/topics/recording.md>), [event-sourcing](<https://devfeed.tech/topics/event-sourcing.md>), [software-architecture](<https://devfeed.tech/topics/software-architecture.md>), [versioning](<https://devfeed.tech/topics/versioning.md>), [ML model versioning](<https://devfeed.tech/topics/ml-model-versioning.md>)

Tags: [22](<https://devfeed.tech/tags/22.md>), [about](<https://devfeed.tech/tags/about.md>), [access-controls](<https://devfeed.tech/tags/access-controls.md>), [account](<https://devfeed.tech/tags/account.md>), [afterthought](<https://devfeed.tech/tags/afterthought.md>), [aggregate](<https://devfeed.tech/tags/aggregate.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-assisted](<https://devfeed.tech/tags/ai-assisted.md>), [ai-audits](<https://devfeed.tech/tags/ai-audits.md>), [ai-deployment](<https://devfeed.tech/tags/ai-deployment.md>), [ai-governance](<https://devfeed.tech/tags/ai-governance.md>), [ai-regulation](<https://devfeed.tech/tags/ai-regulation.md>), [ai-systems](<https://devfeed.tech/tags/ai-systems.md>), [all](<https://devfeed.tech/tags/all.md>), [audit](<https://devfeed.tech/tags/audit.md>), [audit-replay](<https://devfeed.tech/tags/audit-replay.md>), [but](<https://devfeed.tech/tags/but.md>), [climate-modeling](<https://devfeed.tech/tags/climate-modeling.md>), [enterprise-ai](<https://devfeed.tech/tags/enterprise-ai.md>), [gdpr-implementation](<https://devfeed.tech/tags/gdpr-implementation.md>), [half](<https://devfeed.tech/tags/half.md>), [mlops](<https://devfeed.tech/tags/mlops.md>), [model](<https://devfeed.tech/tags/model.md>), [regulators](<https://devfeed.tech/tags/regulators.md>), [reproducibility](<https://devfeed.tech/tags/reproducibility.md>), [with](<https://devfeed.tech/tags/with.md>)

## AI overview

Audit replay answers what state supported a specific past decision; model replay only regenerates an output from a current or rerun system. The article argues that defensible audit replay requires three commitments: persist the complete decision-time state as an immutable event, preserve historical versions through event sourcing or bi-temporal indexing, and verify reconstructions with signed cryptographic hashes. It describes a mid-market option using PostgreSQL JSON records in an append-only table and a cloud KMS with separated key access. These records cost more to store than scores alone, but the article characterizes the added infrastructure cost as a few percent in its cited pipeline. Audit replay can establish the state seen and evidence recorded, but does not reproduce opaque or stochastic model reasoning; decision-trace persistence can partly address that where traces are available.

## Source excerpt

Model replay only regenerates outputs, but audit replay reconstructs specific past decisions with proof of fidelity--and only the latter answers regulators.