F561 — Balanced Deep Mouth
F561 covers “Balanced Deep Mouth”. The canonical record classifies it as mixed, based on 2 source artifacts and 1 original source figure. The highlighted measurements are Baseline Passed = 15, Candidate Passed = 6, and Candidate Losses = 11.
Purpose and research question
F561 evaluated a balanced Deep Mouth candidate intended to reduce the generic-text dominance observed in the preceding experiment by training only on prompt/assistant rows and balancing language families.
Method
F561 trained a balanced Deep Mouth candidate on 51,394 prompt/assistant rows distributed across nine recorded language families. It compared candidate and current-mouth aggregate scores and 20 paired preference judgments, then checked pseudo-word rate before applying the promotion gate.
Documentation basis
2 source artifacts and 1 phase-specific figure are recorded for this phase.
Evidence highlights
- Training rows: 51,394
- Candidate score: 52
- Current mouth score: 76
- Wins/losses/ties: 0/11/9
- Pseudo-word ratio: 0.0
- Promotion: false
Recorded evidence
| Baseline Passed | 15 |
|---|---|
| Candidate Passed | 6 |
| Candidate Losses | 11 |
| Pseudo Word Ratio | 0 |
| Audits Pass | Yes |
| Model Stats Family Count | 8 |
| Model Stats Sentence Family Count | 46875 |
| Model Stats Adam Used | No |
Public evidence and source figures
Original source figure · 1980 × 984 px
These are original source figures from the existing public evidence set, not editorial HTML visualizations.
Result and interpretation
The balanced candidate scored 52 against the current mouth's 76, with 0 wins, 11 losses, and 9 ties. Its pseudo-word ratio was 0.0, but it did not beat the current mouth and remained RESEARCH_ONLY_NOT_PROMOTED.
Key results
- Baseline Passed
- 15
- Candidate Passed
- 6
- Candidate Losses
- 11
- Pseudo Word Ratio
- 0
- Audits Pass
- Yes
- Model Stats Family Count
- 8
- Model Stats Sentence Family Count
- 46875
- Model Stats Adam Used
- No
Evidence sources
Values on this page are reproduced from the archived phase record and the source artifacts below. No new benchmark was run to create this page.
- BALANCED DEEP MOUTH 561 (JSON)
- BALANCED DEEP MOUTH 561 (MD)