Lead README with execution evidence and publish reproducible evaluation

[alloevil]

Summary

• Reorganize both READMEs around execution evidence, a reproducible example, quick start and explicit limitations; preserve detailed usage and all 33 API rows in bilingual guides.
• Include the referenced synthetic pilot and prospective-study source/protocols, without private logs, replacement maps, local configuration or output artifacts.
• Keep the null pilot result visible: no demonstrated advantage over mechanical context. Remote smoke runs are infrastructure checks, not real-world productivity evidence.
• Exclude ignored output from Biome discovery so locally captured nested configurations do not break lint.

Verification

• 332 product tests and 29 prospective-study tests pass locally.
• 16 automatic claims pass; 96 local links/anchors and both asset audits pass.
• Eight local Chromium previews cover both languages, light/dark themes, ~900px content and 360px viewport; no page overflow or missing local images.
• Six synthetic remote runs audited; independent grades agree and resume adds zero model calls.
• actionlint passes. Gitleaks reports two reviewed synthetic redaction fixture values (one already in repository history); no real credentials identified and no private artifacts staged.
• npm lint passes with existing warnings. No package version bump or release.

Limits

Real prospective admitted/evaluated tasks remain zero. Local GitHub-style rendering is not a published GitHub or physical-device test. Full private receipts remain in ignored local output directories.