SAKUWA CONTROL · VOLUME 3
Revising with Hallucination
A protection list and revision ledger — eleven measured revisions and what they invalidated
What breaks after a manuscript is finished and revised? This Japanese-language field log holds the manuscript and anchor constant, then changes only the revision instruction across eleven runs.
ASIN B0HHJGY74F · Public listing checked 2026-09-06
MEASURED IN THE BOOK
What the book actually counted
The book fixes the completed manuscript and anchor, then changes only the revision instruction. “Improvement” here means detection, loss, new violations and ledger invalidation—not literary value.
- 11
- controlled revision conditions
- 3
- defects found only when manuscript and ledger were combined
- 10
- statements invalidated by chapter reordering alone
- 7
- ledger source locations shifted by chapter reordering
- 10/12
- post-revision self-reports matching the measured record
Separate three jobs before asking for a rewrite
1. Read the manuscript
Find continuity breaks, repetition and explanatory imbalance.
2. Read the ledger
Use a separate evidence path to test fixed facts and source locations.
3. Protect, revise, recheck
Freeze facts that must survive, then test invalidation and new violations after revision.
The public evidence is longer than the video
Five YouTube episodes demonstrate the method on separate material. Each ledger page preserves fixed facts, condition-A/B outputs, a comparison table and limits. They are not the book’s eleven experiments.
ハルシネーションで改稿する — 作話コントロール 第3巻
Kindle B0HHJGY74F · Paperback B0HHJVT2V6