Research: discover governed AI-agent identity sources

[murraylovecode]

Summary

• Pre-registers protocol-BATCH-2026-001 for governed identities for AI agents.
• Screens 120 candidate sources: 91 accepted for human selection review, 19 held, and 10 rejected.
• Records 38 query families plus failed and low-yield supplementary searches.
• Keeps quality dimensions separate; no composite authority score is created.
• Adds coverage, concentration, gap, limitation, duplicate, and next-batch analyses.
• Creates no production source records, statements, propositions, dossiers, or publication artifacts.
• Adds a batch-local candidate schema and discovery validator.
• Fixes the protocol validator to compare dates with the actual current UTC date rather than a hard-coded date.

Coverage and limitations

Strongest coverage is in public policy and standards, security research, workload and delegated identity, implementation documentation, and CISO/CIO/CTO perspectives. The batch retains explicit warnings for India, Japan, English-only intake, thin legal and board evidence, sparse healthcare and critical-infrastructure operator evidence, preprint concentration, and vendor-sponsored surveys.

Candidate acceptance is not source approval or publication approval. All 91 accepted candidates remain staging-only and require named-human selection review.

Validation

• Canonical protocol schema and provenance: PASS
• Candidate schema and decision partition: PASS
• Candidate ID, URL, title, source identity, rights, ownership, duplicate, and query traceability checks: PASS
• Coverage and concentration checks: PASS with documented India, Japan, and language warnings
• Staging-leak check: PASS
• TypeScript and targeted ESLint: PASS
• Unit tests: 24 passed
• Repository content, rights, review-status, publication, duplicate, concentration, and coverage gates: PASS

Dependency

This is a stacked draft PR based on build/foundation-2026-08-14 and depends on draft PR #1. It should be rebased or retargeted to main after the foundation merges.

Human review requested

• Confirm protocol scope and the 91 candidate selection judgments.
• Resolve the 19-item hold queue.
• Review five duplicate-rendition resolutions.
• Decide the balanced subset, if any, to advance in a separate canonical-source batch.