Auditable AI delivery · the gate pack

Five controls your audit function can read — and challenge.

Work produced by AI agents passes five automated gates before it ships. Each gate is a plain-language document, not a comment buried in a script — so a client's audit, risk or procurement function reads the rule, disagrees with the threshold if they wish, and has that argument with a person rather than discovering the rule by failing it.

Spec existsEvery deliverable has an executable specification — and it runs.

Glossary cleanNo undefined project shorthand anywhere in the artefact.

Evidence traceableEvery claim traces to a spec clause, a query, or a logged human decision.

Human legibilityThe artefact is legible to a human outside the pipeline.

Audit log completeThe audit log can reconstruct the run, end to end.

Portable by design

Copy one directory into a repository and it works. No dependencies, no install step, no configuration file — Python 3.9 or later is the only requirement. One command runs every gate, locally and in CI:

python3 qa-gates/checks/run_gates.py --project path/to/project

The pack also demonstrates its own controls firing: --specimen runs it against a worked example before and after remediation, and fails if the gates don't fail the first and pass the second. A control that has never fired has not been tested.

See it working

The invoice-injection demo shows a naive agent walking into a poisoned invoice — and the same task run behind these gates, with every control decision logged.

Watch the demo → Book a review

STORYC LTD (company no. 17293375) · AI-Augmented Consulting™ (UK TM application UK00004409146) and the AI Augmentation Line™ are trade marks of StoryC Ltd. The Augmentation Line® is a registered trade mark of Story-C Ltd (UK00004411978, classes 35 and 41), of which the AI Augmentation Line™ is the form in use. The gate pack ships with client engagements; the documents above are the actual controls, not marketing copies of them. Privacy · storycltd.co.uk