Index of everything on this site: https://ai.hiina.space/index.md

# Topic: llm-evaluation

- [Agent comic studios exist; the reader loop doesn't](https://ai.hiina.space/reports/agent-comic-and-manga-studios.md) — Agent comic studios already exist; the missing piece is a calibrated, cross-family cold-reading VLM that critiques lettered pages, and a six-page pilot costing tens of dollars can test it. (v1, 2026-10-27)
- [AI enfeeblement, idle supervisors and the Peter principle in agents](https://ai.hiina.space/reports/ai-enfeeblement-and-the-peter-principle.md) — None of the four questions has been measured directly; the cheapest real result is a tool-off audit of released tool-RL checkpoints, and the most novel is an idle-supervisor tick loop. (v1, 2026-10-27)
- [Score provenance as evidence, never as permission: AI moderators and provenance signals](https://ai.hiina.space/reports/ai-moderators-and-provenance-signals.md) — Let an AI moderator lean on classifiers, reputation and signed-agent provenance only through a calibrated scoring pipeline it never sees raw; treat provenance as a capped, measured, positive-only prior. (v1, 2026-10-27)
- [Story decompilers exist, but nobody closes the loop](https://ai.hiina.space/reports/story-decompiling-and-remixing.md) — Story decompilers and plan-conditioned generators both exist, but nobody closes the extract-regenerate-edit loop; it is cheap to start on text, provided decoys control for memorised plots. (v1, 2026-10-27)
- [Two-stage styled translation exists only in pieces](https://ai.hiina.space/reports/two-stage-user-styled-translation.md) — No system yet joins an offline per-line meaning record to a cheap user-styled renderer; build it with closed-class style flags in code, the Japanese kept in the prompt, and style-keyed render caches. (v1, 2026-10-27)