Board 1 of 3
Capability
How far a rig gets. Every row is a full configuration — which model, backed by which memory system, at which reasoning effort — ranked in Effective Minutes: quality-adjusted minutes of expert work replaced. Em is unbounded by construction, so harder fixtures raise the ceiling instead of saturating it.
0 ranked
| # | Model | Memory | Effort | Em | Em | Cost | Ctx tok | Runs | Corpus |
|---|---|---|---|---|---|---|---|---|---|
| No rows can be ranked on this view with the current filters. | |||||||||
How this is scored
- Unit
- Effective Minutes (Em) — quality-adjusted minutes of expert work replaced.
- Lift
- Em(rig) minus Em(same model, same effort, vendor-native) within the same surface and corpus. Null when no matched baseline exists — never zero. A positive lift on "None" means the bare model beat having the whole corpus in context, which is a real result at large corpus sizes, not an error.
- Anchors
- T_human anchors are PROVISIONAL expert estimates, not panel-measured.