chat model, answer-letter probabilities · v1 set (earlier version)

GLM-5.3-Flash NVFP4

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
Missing
Provider or endpoint
Missing
Date run
Missing
Price
Missing
Model type
chat model, answer-letter probabilities
Weights
Missing
Runs on
Missing
Reasoning setting
Missing
Confidence source
letter probability
Rules text
short (the v1 rules text; the same stored request text for every model)
Option groups
not recorded; no decision in this data was split
Sampling settings
Missing
Notes
Missing

Missing B7 fields: model_id, provider, weights, runs_on, reasoning, option_groups, sampling, price, 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)
85.477.6 to 92.6
Decisions
1,000 (1,000 scored, 0 invalid)
Engine's best play
29.1%
Blunder rate
47.8%
P95 equity loss
0.690
Mean loss, worst 5%
0.933
Median / P90 s per decision
0.79 / 1.60
$ per 1,000 decisions
Not measured
Risk judgement, margin 0.001
34.9 (95% interval 31.7 to 38.2); risks found 22.8% (81 / 355), right safety 46.9% (173 / 369)
Risk judgement, margin 0.02
33.0 (95% interval 29.9 to 36.3); risks found 20.4% (65 / 318), right safety 45.6% (185 / 406)
No trade-off
276 (left out of every risk metric)

JSON

Risk call card

34.9risk judgement, margin 0.001

right risks81
wrong risks192
needless risks196
missed risks82
right safety173

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

By game phase

PhaseError rate (PR)Decisions
OpeningToo few decisions143
Contact106.4641
Holding gameToo few decisions20
RaceToo few decisions136
Bear-offToo few decisions60