Experiment ·
DEEPROUTE
Descriptive (0 calls): where does the deep tier (Qwen3.6-35B-A3B engine, night-10-01) beat the 4B engine item by item on v0.1 EN 57, and what would a simulated hybrid router capture?
From the record
- What was tested
- DEEPROUTE
- Result
- 43 %/34 % capture out-of-sample
- Verdict
- parity: an equivalence (same quality, cheaper or transferable), not a gain
- Detail
- closed (descriptive, no verdict rule; routing family closed in CLOSED R28)
- Set
- v0.1 EN 57
- Judge
- Claude Opus
- n
- 57
- Seeds
- deep s42/s43; 4B s42/s43/s44 (existing rows)
- Pre-registered
- yes, sha256-frozen
- Reopen if
- a pre-generation signal capturing >=70 % of the gain at <=40 % deep (none tested reached it)
Linked facts · 1
- R28 Per-item routing 4B ↔ deep: 0 callsR28 Per-item routing 4B ↔ deep: 0 calls43 %/34 % capture out-of-sampleKinda record, not a measurement (set, status, judge, n, seeds, date: not recorded)Caveatwhy closed: chance-level; needs 12 GB deviceSourceresearch record
reports/CLOSED.md:47Open the data file