Add governed agent identity evidence research batch

[murraylovecode]

What changed

Adds BATCH-2026-005, a staging-only evidence-research package for governed identities for AI agents.

• 9 verified source records representing 8 intellectual works
• 22 versioned, source-located statement candidates
• 7 statistics reviews with denominator, sample, scope, uncertainty, and locator context
• complete methodology, funding/conflict, rights, duplicate, correction/retraction, version, DOI, author, and Executive AI Research crosswalk records
• analytical abstracts and separate evidence-quality dimensions for every source

Why

This converts a balanced subset of accepted discovery candidates into methodology-aware evidence records without treating normative standards, preprints, and synthetic benchmarks as equivalent forms of evidence.

Important review notes

• All records remain staging-only and pending human review.
• The OpenID PDF and arXiv preprint are retained as related renditions and must not be double-counted.
• AgentDojo v3 is used because its authors identify a bug fix and travel-suite update.
• Two initially considered reports were excluded because their full downloads were access-gated or permission-blocked.
• No source PDFs, full text, figures, tables, or direct quotations are staged.

Validation

• Canonical source and statement schemas: pass
• Evidence-specific completeness checks: pass
• DOI, version, duplicate, methodology, sample, statistics, locator, funding, correction/retraction, OFF crosswalk, rights, and staging-isolation checks: pass
• Canonical repository validators and audits: pass
• Tests: 24 passed across 4 files
• git diff --check: pass