Sakuwa Control Ledger · hosted by TacticalPrompt · Episode 2
It got better when the table was removed — a measurement that withdrew my own hypothesis
With the self-check requirement removed and only the ledger supplied (CASE-L10): zero violations. The wording in chapter 2 was corrected.
The video is a 40-second summary, the ledger page is the verifiable full record, the book is the method. All three point to the same single experiment.
▶ Watch on YouTube · 37sYouTube
https://www.youtube.com/shorts/FR6BDNcfG08
Channel
作話コントロール (@sakuwacontrol)
Confabulations
1
summary only
1 volume-split violation — but it reported that one itself, inside the table.
0
with ledger
Zero violations. An unrequested nine-row table and one self-report came as a bonus.
Fixed facts (ledger)
| # | Item | Fixed value | Source |
|---|---|---|---|
| 1 | Fixed-facts ledger | nine items, C-01 to C-09 | vol. 2 ch. 2 · appendix C |
| 2 | Self-check table requested | none — removed for this condition | ibid. |
| 3 | Anchor, instruction, model, summary | identical to L05 / L06 | ibid. |
| 4 | Hypothesis under test | “Add one more check table and it gets detected.” | the falsification condition stated in chapter 2 |
Conditions
Condition A · summary only
Nine ledger items + the dedicated self-check table requested (CASE-L06).
One volume-split violation occurred. It self-reported in the arithmetic-check field: “this is an anchor violation and requires correction.”
Condition B · summary + ledger
The same nine items supplied as information only, with the table requirement removed (CASE-L10).
Shared instruction
Attaching the ledger mixes two effects: adding one more check table, and handing over nine items of information. To separate them, one more run was made with the nine items supplied but the table requirement removed.
Results
| Item | Condition A | Condition B |
|---|---|---|
| Fixed-facts violations | 0○ match | 0○ match |
| Volume-split violation | 1 (“volume four”)× confabulation | 0 — March pages were correctly opened in volume three○ match |
| Production of a self-check table | produced because it was requested○ match | not requested, yet it built nine rows C-01–C-09 on its own○ match |
| Unprompted self-report | n/a○ match | 1 — it proposed adding a note to the ledger○ match |
| The author’s hypothesis | —○ match | withdrawn× confabulation “more tables means detection” cannot be maintained in this form |
Sources
- Writing a Novel with Hallucination, chapter 2 — verifying the prescription and its falsification condition
- The same book, appendix C — C-0 list of runs (L10)
Honesty guards
- All three are single runs. This design cannot separate a condition effect from chance.
- In the table-less condition no violation arose to report, so “can it report without the table?” remains unverified. Removing the table in a condition where a violation does occur is the next falsification test.
- This episode records the author’s own hypothesis being broken by measurement and withdrawn.
vol. 2, chapter 2
The method is in chapter 2 of the book →
Ledger template · free
Get the ledger & self-check templates (PDF)
The same tables used in the videos and on this page: 15 fixed-fact rows, the instruction template, and the self-check table. Leave an email to open it now; we send one note when a new episode ledger goes up.
Used only for new-episode notes. Unsubscribe any time.
This page is a first-edition translation of the Korean source and is under review.