Publish frozen comparative baselines

[aronchick]

What changed

• add four deterministic repository-owned generation paths with three seeded runs each
• freeze 12 editable decks, complete local reports, cohort provenance, hashes, and one-command verification
• add a GitHub-forward comparative chart and rebuild the 21-second launch film from the same summary JSON
• keep every score self-reported and every path free of model attribution

Verification

• 218 grader tests, 86.70% coverage
• 115 service tests, 85.04% coverage
• corpus validation PASS
• comparative-v1 verifier: 12/12 runs, 240 slides
• launch verifier: 1080x1080, 30 fps, 21 seconds, silent
• Linux/amd64 image rebuilt twice at identical digest sha256:29f5fe733d762d2633f28824a18807c3273d02c040db56c0cda12affd246732d
• desktop/mobile browser canary: no console errors or document overflow

Truth boundary

These are reproducible workflow baselines, not model rankings. Reports are labeled local artifact score; self-reported and remain ineligible for an official leaderboard.