State

What the register says, as of the last export. Playing strength is only ever a difference to the stage’s opponent, predecessor or reference v0, with an error range where one was reported, never an absolute number.

Key figures

Games in the register: 669,740
Verdicts: 50 passed · 18 failed · 4 rejected
Success rate: 69.4 % 10 of the passed entries played no games (identity, correction, epoch change, adoption); the rate counts them, as the oracle defines it.
Epochs: 4 T-0001 – T-0072
Measured hours: at least 124.0 Durations are recorded for part of the runs only; the sum is a lower bound.
Games per minute: short stage 138.4 · long stage 27.4 Median throughput by stage.

Long stage: Elo per run

Every register entry with a long stage, in register order: estimate and reported 95-percent half-width against the stage's opponent. Square marks measure against the reference v0, round marks against the predecessor or without a named opponent. Hollow marks carry no half-width. Root scale: values near zero are stretched, large values compressed; the tick labels are the true values.

From T-0014 to T-0052 the long stage was played against the reference v0, not the predecessor; there the estimates rise above +1,000 and mostly carry no half-width. Only from epoch 3 on does the long stage measure against the predecessor.

Short stage: Elo per run

Every register entry with a short stage, in register order: estimate and reported 95-percent half-width against the stage's opponent. Square marks measure against the reference v0, round marks against the predecessor or without a named opponent. Hollow marks carry no half-width. Root scale: values near zero are stretched, large values compressed; the tick labels are the true values.

Cumulative long stages, as exported by the oracle

Each point is a passed long stage against its predecessor; the line sums them, the band is the 95-percent interval of the sum. The curve adds up selected predecessor comparisons; it is not a progress meter and not an absolute playing strength.

Run	Date	Delta	Cumulative	95 % interval
T-0064	2026-09-03	+11.72	+11.72	+4.81 … +18.63
T-0065	2026-09-04	+6.74	+18.46	+10.01 … +26.91
T-0066	2026-09-04	+17.47	+35.93	+23.85 … +48.01
T-0067	2026-09-04	+16.22	+52.15	+37.48 … +66.82
T-0072	2026-09-08	+260.44	+312.59	+270.55 … +354.63

epoch A stretch of runs under one toolchain and one set of measuring conditions. Values across epochs are not added.

Epochs separate measurements whose toolchain or conditions are not freely comparable. Values across an epoch boundary are not added.

Epoch	Anchor	From	To	Reason	Measurements	Toolchain
1	T-0054	2026-08-28	2026-09-02	bootstrap	55	`4ea30c1cade2`
2	T-0056	2026-09-03	2026-09-03	toolchain	6	`5c547b3a8043`
3	T-0062	2026-09-03	2026-09-04	measuring conditions	8	`5c547b3a8043`
4	T-0070	2026-09-05	2026-09-08	measuring conditions	3	`5c547b3a8043`

Measuring stages

SPRT Sequential probability ratio test. Games are played until the accumulated evidence reaches one of two bounds; then the test stops. hypotheses in nElo, normalized model. Both stages must pass.

Stage	time control Base time plus increment per move in seconds, e.g. 8+0.08: eight seconds per game plus 0.08 seconds per move.	Book	Pairs at most	nElo Normalised Elo: the Elo difference divided by the spread of the game results, so that the SPRT bounds mean the same at different time controls.	Concurrent
short stage	8+0.08	A	80,000	[0.0, 5.0]	60
long stage	40+0.4	B	40,000	[0.0, 5.0]	60

Books

Name	Role	Stage
A	probe book	short stage
B	secret, long stage only	long stage