F885 — Hard Blind Ablation
F885 covers “Hard Blind Ablation”. The canonical record classifies it as validated, based on 2 source artifacts and 1 original source figure. The highlighted measurements are Valid reasoning pass = 0.833, No-memory pass = 0.083, and Empty-memory pass = 0.083.
Purpose and research question
This phase tested whether the F884 result depended on learned reasoning memory by comparing the valid path against no-memory and empty-memory ablations.
Method
F885 is documented by 1 structured JSON record and 1 Markdown report and 1 original source figure. The reported table retains safe scalar measurements found in that phase package.
Documentation basis
2 source artifacts and 1 phase-specific figure are recorded for this phase.
Evidence highlights
- Valid reasoning pass: 0.833
- No-memory pass: 0.083
- Empty-memory pass: 0.083
- No-memory drop: 0.750
- Empty-memory drop: 0.750
- Reasonmem use rate: 0.833
Recorded evidence
| Valid reasoning pass | 0.833 |
|---|---|
| No-memory pass | 0.083 |
| Empty-memory pass | 0.083 |
| No-memory drop | 0.750 |
| Empty-memory drop | 0.750 |
| Reasonmem use rate | 0.833 |
Public evidence and source figures
Original source figure · 1400 × 672 px
These are original source figures from the existing public evidence set, not editorial HTML visualizations.
Result and interpretation
The valid path reached 0.833 while both blind ablations reached 0.083. The recorded 0.750 drop supports the report's conclusion that the measured result was not explained by a hidden fallback or an overly easy benchmark.
Key results
- Valid reasoning pass
- 0.833
- No-memory pass
- 0.083
- Empty-memory pass
- 0.083
- No-memory drop
- 0.750
- Empty-memory drop
- 0.750
- Reasonmem use rate
- 0.833
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.
- HARD BLIND ABLATION 885 (JSON)
- HARD BLIND ABLATION 885 (MD)