1. Introduction

This section is informative.

OIS defines a vendor-neutral data model that links what was required (the intent) to what was produced and to the evidence that it meets the requirement. It works for artifacts made by humans, AI agents and tools alike.

1.1 Problem

AI agents now write requirements, code and tests. Teams can no longer show, in a verifiable way, that a delivered feature does what was asked:

• Requirements, code, tests and results live in separate tools with no verifiable link between them.
• Generated tests often confirm what was built rather than what was required.
• When one model reviews another model’s output, both may share the same blind spots.
• Regulated organizations must increasingly document how AI is used and controlled in software delivery. This evidence is collected by hand or not at all.

1.2 Goals

• Record intents and their acceptance criteria as versioned, immutable records.
• Record who or what produced each artifact, and what context the producer could see.
• Record evidence that links each criterion to a concrete test run, including whether the test was derived from the intent or from the implementation.
• Record human and automated decisions at quality gates.
• Bundle these records into a signed Quality Passport for a release.
• Reuse existing standards wherever possible.

1.3 Non-goals

• OIS does not define how risk, coverage or trust scores are computed. It only defines where such values are recorded. Appendix D offers one informative reference method for intent coverage, so that at least one comparable figure exists.
• OIS does not prescribe tools, agents, workflows or user interfaces.
• OIS does not interpret laws or regulations. Section 13 maps fields to frameworks for orientation only.
• OIS does not store artifacts. Records reference artifacts by location and digest.