BlackboxBook/.github/review-cache/scope-log.md

Scope Log

Use one row per reviewed scope so future runs know what was already checked and which topic caches were used.

Scope Kinds

• full-book: whole-manuscript audits.
• chapter: one chapter file.
• section: one section-bounded slice of a chapter.
• cross-book-topic: terminology, naming, or other cross-chapter review.
• follow-up-fix: incremental correction after a previous review.
• structure: rename, move, delete, or navigation repair work.

Column Guide

• Scope ID: stable identifier for the reviewed slice.
• Scope kind: one of the kinds above.
• Files: affected files or chapters.
• Topic IDs: reusable topic caches consulted or updated.
• Last reviewed: date of the latest pass.
• Review mode: full, incremental, research-only, edit-follow-up, structure, or similar.
• Cache result: short status such as reused, partial-refresh, new-topic, stale-pending.
• Next action: what still needs a refresh or follow-up.
• Notes: concise summary of why this row exists.

Scope ID	Scope kind	Files	Topic IDs	Last reviewed	Review mode	Cache result	Next action	Notes
fbr-2026-04-11	full-book	book/00–25, readme.md	anthropic-claude-family, openai-gpt5-family, google-gemini-gemma, alibaba-qwen, deepseek-family, xai-grok, meta-llama, zhipu-glm, minimax-m27, mcp-ecosystem, slm-edge, hybrid-architectures	2026-04-11	full	new-topic (all 12 topics created)	Refresh fast-class topics by 2026-05-11	First full-book review. 8 research passes, 5 fact-check passes, 4 synthesis briefs, 4 editor passes, 1 structure pass. ~64 actionable findings, ~56 resolved. Validation clean.
ta-muse-spark-2026-04-11	section	book/24_landscape_2026.md	meta-llama	2026-04-11	topic-addition	partial-refresh	Revisit when Meta publishes Muse Spark architecture/benchmarks	Added Meta Muse Spark (closed multimodal LLM, MSL, April 2026) to Ch.24 text. No table entry — no quantitative data available. Source: about.fb.com/news/2026/04/introducing-muse-spark-meta-superintelligence-labs/
fbr-2026-04-11-b	full-book	book/00–25	anthropic-claude-family, openai-gpt5-family, google-gemini-gemma, alibaba-qwen, deepseek-family, xai-grok, meta-llama, zhipu-glm, minimax-m27, mcp-ecosystem, slm-edge, hybrid-architectures	2026-04-11	full	reuse_cache (all 12 topics)	Refresh fast-class topics by 2026-05-11	Second full-book review. 6 FC passes, 3 CA passes, 2 synthesis, 16 editor passes, 1 structure pass. 55 actionable findings after dedup (16 major, 39 minor), all resolved. Key fixes: Gemini audio pricing recalc (Ch18), guardrails dedup (Ch15↔Ch13), stale URLs (Ch22), DSA expansion (Ch24), CoT dedup (Ch03↔Ch09), ~100+ hyperlink conversions, glossary expansion (Ch25), Jamba 2 param hedging (Ch02/Ch05). Validation clean.
rfu-2026-04-15-llm-mechanics	follow-up-fix	book/02, book/03, book/12, book/19, book/21	interpretability, rag-evals, post-training, serving-runtime	2026-04-15	edit-follow-up	new-topic	Consider cross-linking into Ch13 / Ch24 on next pass	Targeted research refresh on LLM mechanics: circuit tracing added to Ch02/03, CRAG + RAGChecker + ClashEval integrated into Ch12/03, ORPO/KTO/SimPO + online RLHF added to Ch19, disaggregated prefill metrics clarified in Ch21. Pending: optional multimodal follow-up in Ch18.