decision model · v1 set (earlier version)

Clef

Earlier version

Results f24d1f41, updated 16 November 2026, 22:20 UTC; scored by GNU Backgammon 1.08, 2-ply.

How we called it

Model ID
clef
Provider or endpoint
our GPUs
Date run
Missing
Price
no price per token; own hardware; price date not recorded
Model type
decision model
Weights
open
Runs on
our GPUs
Reasoning setting
none (decision model)
Confidence source
decision-model probability
Rules text
short (the v1 rules text; the same stored request text for every model)
Option groups
groups of up to 110 options; no decision in this data was split
Sampling settings
not sent (decision request)
Notes
Missing

Missing B7 fields: date_run

Results in v1 set (earlier version)

Earlier version: plays were listed in the runner's order, which tends to put the best play early. When Gemini 4 Argon answered the same 1,000 positions with the plays shuffled, its error rate changed by +1.9 (95% interval −0.2 to +4.2), so the effect is not proven.

Error rate (PR)
34.129.9 to 38.4
Decisions
1,000 (1,000 scored, 0 invalid)
Engine's best play
48.9%
Blunder rate
24.8%
P95 equity loss
0.317
Mean loss, worst 5%
0.527
Median / P90 s per decision
0.94 / 1.27
$ per 1,000 decisions
Not measured
Risk judgement, margin 0.001
60.2 (95% interval 57.0 to 63.6); risks found 43.9% (156 / 355), right safety 76.4% (282 / 369)
Risk judgement, margin 0.02
60.0 (95% interval 56.5 to 63.4); risks found 45.0% (143 / 318), right safety 75.1% (305 / 406)
No trade-off
276 (left out of every risk metric)

JSON

Risk call card

60.2risk judgement, margin 0.001

right risks156
wrong risks62
needless risks87
missed risks137
right safety282

Hit risk only: being hit. Gammon risk and priming are not in it.

By game phase

PhaseError rate (PR)Decisions
OpeningToo few decisions143
Contact42.8641
Holding gameToo few decisions20
RaceToo few decisions136
Bear-offToo few decisions60