Coherent 0.2.22: 3688 ± 52 over 10 anchors
Listen to this post9 min
Created by AI from the text of this post and read by the synthetic voice “Sulafat”; not a word-for-word version, and it may contain errors.
Loading…
Created by AI from the text of this post and read by the synthetic voice “Sulafat”; not a word-for-word version, and it may contain errors.Position and speed stay in this browser and are sent nowhere.
The recording could not be played. Open the audio file
Project measurement, not in the register: an anchoring against foreign engines with a listed rating. The figures come from the data delivery (entry M-NNNN One Elo measurement: a build anchored against foreign engines with a listed rating, on a measuring machine of its own. Numbered in the order delivered; the register run it measured is named with it. The series continues the entries E-0001 to E-0012 and starts at M-0013.); the build measured is the one from run T-0198 of 21 October 2026. The reservations below are those delivered by the day of publication; the full and current account is on the playing-strength page.
Coherent 0.2.22 played 4,000 games against 10 foreign engines with a published rating, avalanche 5.0.0, cinder 0.6.1, halogen 16.0.0, justbot 0.5.0, koivisto 9.0, prune 4.0.1, raphael 4.2.0, renegade 1.3.0, stormphrax 8.0.0 and tcheran 13.0, 400 against each at 120+1. The result is 3688 ± 52, ± 6 of it statistical; anchored to the opponents' CCRL Computer Chess Rating Lists, a public ranking of chess engines from games under standard conditions. blitz ratings, not a CCRL rating. From 25 October 2026 that is the figure at the head of this site; it replaces the anchoring of 0.2.21 at 3701 ± 51.
The runs
Project measurement, not in the register. 10 horizontal intervals, one per opponent, on a scale of anchored playing strength: avalanche 5.0.0 at 3681 with a half-width of 21, cinder 0.6.1 at 3713 with a half-width of 17, halogen 16.0.0 at 3698 with a half-width of 12, justbot 0.5.0 at 3692 with a half-width of 17, koivisto 9.0 at 3686 with a half-width of 15, prune 4.0.1 at 3665 with a half-width of 17, raphael 4.2.0 at 3687 with a half-width of 17, renegade 1.3.0 at 3684 with a half-width of 16, stormphrax 8.0.0 at 3690 with a half-width of 16, tcheran 13.0 at 3684 with a half-width of 16. Below them the published figure for 0.2.22, 3688 with a band of 52, shaded behind every row. The plus-minus values are half-widths of nominal 95 % intervals; the published band also carries the transfer term.
| Opponent | Listed | Games | Score rate | Difference | From it |
|---|---|---|---|---|---|
| avalanche 5.0.0 | 3666 ± 18 | 400 | 52.12 % | +14.80 ± 10.20 | 3681 ± 21 |
| cinder 0.6.1 | 3762 ± 12 | 400 | 43.00 % | −49.00 ± 11.60 | 3713 ± 17 |
| halogen 16.0.0 | 3721 ± 8 | 400 | 46.75 % | −22.60 ± 9.10 | 3698 ± 12 |
| justbot 0.5.0 | 3705 ± 14 | 400 | 48.12 % | −13.00 ± 10.20 | 3692 ± 17 |
| koivisto 9.0 | 3632 ± 7 | 400 | 57.75 % | +54.30 ± 13.40 | 3686 ± 15 |
| prune 4.0.1 | 3645 ± 13 | 400 | 52.88 % | +20.00 ± 10.90 | 3665 ± 17 |
| raphael 4.2.0 | 3731 ± 12 | 400 | 43.75 % | −43.70 ± 11.40 | 3687 ± 17 |
| renegade 1.3.0 | 3694 ± 12 | 400 | 48.62 % | −9.60 ± 10.30 | 3684 ± 16 |
| stormphrax 8.0.0 | 3745 ± 11 | 400 | 42.12 % | −55.20 ± 11.70 | 3690 ± 16 |
| tcheran 13.0 | 3682 ± 13 | 400 | 50.25 % | +1.70 ± 9.30 | 3684 ± 16 |
All half-widths at the 95 % level, as the delivery states them; the column From it is the anchor plus the measured difference. The listed ratings are from the CCRL Blitz Rating List, single-core rows only, as of 2026-10-22.
How the figure is formed
Recalculated from the anchors: the 10 anchored values, averaged with weights inverse to their variance, give 3689.0 with a statistical half-width of ± 5.0; the delivery rounds it to ± 6. With the measuring side's transfer term the band is ± 51.2; the delivery names 3688 ± 52, and this site publishes the delivery's figure. The anchored values contradict one another (chi-square 19.42 at 9 degrees of freedom, bound 16.92; half-widths read as 95 % half-widths).
Against 0.2.21 six anchors were played in both campaigns, cinder 0.6.1, halogen 16.0.0, justbot 0.5.0, raphael 4.2.0, renegade 1.3.0 and stormphrax 8.0.0. Per anchor the difference of the two measured differences is +2.60 ± 16.55, +25.50 ± 14.66, +1.80 ± 14.21, +0.80 ± 16.19, +4.30 ± 14.22 and +0.90 ± 16.76; combined, +6 ± 6. The per-anchor values are compatible with one another (chi-square 8.08 at 5 degrees of freedom, bound 11.07). That is the measured change: the opponent's listed rating and the transfer term cancel in it. The 13 points between the two headline figures are not a measured gain; they carry both bands in full.
Project measurement, not in the register. Six horizontal intervals, one per anchor played in both campaigns, on a scale of Elo difference against 0.2.21: cinder 0.6.1 at +3 with a half-width of 17, halogen 16.0.0 at +26 with a half-width of 15, justbot 0.5.0 at +2 with a half-width of 14, raphael 4.2.0 at +1 with a half-width of 16, renegade 1.3.0 at +4 with a half-width of 14, stormphrax 8.0.0 at +1 with a half-width of 17. Below them the combined value, +6 ± 6. A dashed vertical line marks zero, no change.
What belongs to this figure
- Measured outside the formal registration procedure (project measurement on the project's Arena machine, no oracle finding, Law I).
- The privileges of the measuring agents on the testbench are not yet settled.
- Defaulted opponents do not count (a block not completed or more than 2% of their games aborted, with the opponent or neither as loser): none.
- Abandoned games without a loser (abandoned with * or draw) count as a technical loss for the opponent: none.
- Aborted games with Coherent as loser count as normal (engine finding): none.
- Custom opening book and no endgame tablebases instead of the list conditions; time margin 250 ms, one dedicated core per engine.
- Quotable is ± 52 (95%): statistically 5.0 from games and list errors of the anchors (rounded up to 6), together with the transfer uncertainty 51 as √(statistical² + transfer²). The transfer to the conditions of the list (book, endgame databases, hardware) is not measured; it is a judgement.
Without a provenance record in the delivery: avalanche 5.0.0, cinder 0.6.1, halogen 16.0.0, justbot 0.5.0, koivisto 9.0, prune 4.0.1, raphael 4.2.0, renegade 1.3.0, stormphrax 8.0.0 and tcheran 13.0.
Conditions
120+1, book 8moves_v3 with 34,700 positions, each colour pair with an opening of its own, 256 MB hash, one thread, 7 concurrent games, no adjudication, a fixed number of games without SPRT Sequential probability ratio test. Games are played until the accumulated evidence reaches one of two bounds; then the test stops.; 2 losses on time. The measurement is dated 25 October 2026, 02:10 UTC.
Generated from the data delivery. The recalculation is the site's; every other figure is the measuring side's.
Note of 25 October 2026: the figure is below that of 0.2.21 (3701), although a gain of +6 ± 6 was measured over the six shared anchors. The reason lies in the measurement of 0.2.21: there, 85 games of one opponent were lost on time, which goes back to the measuring environment of the time and raised the value by about 13 points; without those games it gave 3688. Corrected conditions have applied since, a core of its own for each engine and a time margin. In detail in the entry "Arena measurement EV-3: 3701 published, 3688 without hobbes".