Falandays core benchmark, version 2
This next benchmark profiles the canonical Falandays node on Tracking, Pong, and Wall. It does not compare Falandays with another node or assign a cross-task rank. No version 2 run record exists yet.
Protocol
Each task uses 20 independent blocks with one freshly constructed trial per block.
| Task | Horizon | Warm-up | Scored ticks | Root seed |
|---|---|---|---|---|
| Tracking | 7,200 | 1,200 | 6,000 | 61,001 |
| Pong | 7,200 | 1,200 | 6,000 | 62,001 |
| Wall | 1,000 | 800 | 200 | 63,001 |
Pong produces about one ball event per 305 ticks. Its scored interval therefore contains about 20 events, rather than the two or three available in version 1. This is enough to make the per-run hit rate more than a two- or three-event fraction, while the 20 blocks remain the independent units for its interval.
Wall provides the third task coordinate. Its scope-setting measurement was 0.9933 raw and
0.9703 normalised. This near-ceiling result makes Wall a competence floor-check with poor
upward discrimination.
The scope-setting values are planning measurements, not a benchmark record or interval.
The evidence state remains planned.
Frontier outside the aggregate
cartpole_plank_easy is registered with the :frontier tag but is not a case in this
protocol. Canonical Falandays measured 10.0 raw and 0.0007 normalised. It remains
unsolved by the canonical node and cannot discriminate designs until one clears the floor.
The repository contains the planned protocol.