Add governed agent identity media research batch

[murraylovecode]

What changed

• adds media research batch BATCH-2026-004 for five discovery-approved candidates
• accepts two fully reviewable recordings and holds three candidates without accessible media
• stages 2 source proposals, 9 person candidates, 12 organization references, 29 chapters, and 18 statement candidates
• records exact speaker-at-source-time context, rights treatment, repost groups, and transcript limitations
• preserves all work in staging; no production publication records are created

Why

This follow-on batch verifies canonical source identity and performs transcript-aware analysis for media candidates related to governed identities for AI agents. It prevents event abstracts, slide decks, promotional speaker lists, and inaccessible recordings from being treated as spoken evidence.

Notable review findings

• the SailPoint recording contains one on-record presenter, Amy Shillinglaw; Kirby Fitch is explicitly absent, Paul Patti is not audible, Steve Toole is off-camera Q&A support, and Rachel Leighter is the uploader
• the OpenID, CSA, and Identiverse candidates remain held because no public recording with duration and speaker timing was located
• both accepted sources are third-party and limited to link-and-paraphrase treatment; no transcript or media asset is committed

Checks

• batch-local schema, duration, timestamp, identity, role-history, ownership, transcript-rights, duplicate, abstract, and staging-isolation checks
• canonical data, provenance, attribution, statements, source locators, rights/privacy, review-status, publication, and content validation
• 24 unit tests
• Astro typecheck: 57 files, zero findings
• ESLint
• git diff --check

Review status

Draft for named-human source, attribution, rights, and statement review. Candidate acceptance does not imply publication approval.