Playing strength

Published figure: Coherent 0.2.24

Coherent 0.2.24: 3683 ± 52, ± 5 of it statistical; anchored to the opponents' CCRLCCRL Computer Chess Rating Lists, a public ranking of chess engines from games under standard conditions. blitz ratings, not a CCRL rating. Project measurement, not in the register. Published here since 29 October 2026; it replaces the anchoring of 0.2.22 at 3688 ± 52.

All measurements against foreign engines so far

The runs

All half-widths at the 95 % level, as the delivery states them; From it is the anchor plus the measured difference. Listed ratings from the CCRL Blitz Rating List, single-core rows only, as of 2026-10-27.

How the figure is formed

Recalculated from the anchors: the 10 anchored values, averaged with weights inverse to their variance, give 3683.7 with a statistical half-width of ± 5.0; the delivery rounds it to ± 5. With the measuring side's transfer term the band is ± 51.2; the delivery names 3683 ± 52, and this page publishes the delivery's figure. The anchored values are compatible with one another (chi-square 5.69 at 9 degrees of freedom, bound 16.92; half-widths read as 95 % half-widths).

Against 0.2.22 the anchors played in both campaigns are avalanche 5.0.0, halogen 16.0.0, justbot 0.5.0, raphael 4.2.0 and renegade 1.3.0. Per anchor the difference of the two measured differences is −4.40 ± 13.94, −6.10 ± 13.45, −7.00 ± 13.87, +4.40 ± 15.84 and −0.80 ± 13.61; combined, −3 ± 6. The per-anchor values are compatible with one another (chi-square 1.50 at 4 degrees of freedom, bound 9.49). That is the measured change: the opponent's listed rating and the transfer term cancel in it. The 5 points between the two headline figures are not a measured gain; they carry both bands in full.

What belongs to this figure

1. Measured outside the formal registration procedure (project measurement on the Arena machine, no oracle finding, Law I).
2. The measuring agents' privileges on the machine are not yet settled.
3. Defaulted opponents do not count (a block not played to the end or more than 2 % of their games abandoned, with the opponent or neither as loser): none.
4. Games abandoned without a loser (abandoned with * or drawn) count as a technical loss for the opponent: none.
5. Aborted games with Coherent as loser count as normal (engine finding): none.
6. Own book (8moves_v3) and no endgame databases instead of the list's conditions; time margin 250 ms, one dedicated core per engine.
7. Quotable is ± 52 (95%): statistically 5.0 from games and list errors of the anchors (rounded up to 5), together with the transfer 51 as √(statistical² + transfer²). The transfer uncertainty to the conditions of the list (book, endgame tablebases, hardware) is not measured; it is a judgement.

Without a provenance record in the delivery: avalanche 5.0.0, halogen 16.0.0, justbot 0.5.0, minke 7.0.0, pzchessbot 7.1, quanticade_cronus 3.0, raphael 4.2.0, renegade 1.3.0, tcheran 14.0 and zangdar 7.

Unregistered: measured on the arena measuring machine under the export rule EV-5 fixed beforehand, outside the registration procedure. The reservations of the delivery are listed below.

Delivered as entry M-0021M-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. of the data delivery; the build measured is the one from run T-0202, the measurement is dated 29 October 2026. T-0202 · The post on this measurement

Earlier measurements against foreign engines

The delivery’s values as delivered. Each measurement is read, as it was at the time, in its post.

Unregistered: measured outside the registration, on a machine whose chain of custody was not secured throughout (order, storage and access with one account). Such a value is listed but never given as the playing strength above.

0.1.91: from the measuring side's report, not in the delivery; never the published figure, replaced by 0.1.93 on the same anchors.

0.1.61: as delivered, after a 600-game repeat run against Leorik 2.1 replaced the aborted one; the post gives 2570 ± 55 from the aborted run.

0.1.58: the first transfer, over two anchors that disagree with each other; not in the delivery.

The measuring conditions of the anchoring of Coherent 0.2.21 departed from the standard of the rating lists. Both engines of a game shared one processor core, and the tournament interface had no time buffer. As a result, some opponents lost games on time for reasons that lie in the measuring environment. In addition, Coherent was measured from a separate arena build, not from the binary checked by the oracle. The stated error band of ±5 is purely statistical. The citable total band, including the transfer to the rating list, is considerably wider (about ±50). In which direction the bias works is not determined. The value is therefore of limited reliability. From the next measurement on, corrected conditions apply: one core per engine, a uniform time buffer and measurement of the oracle binary.

The engines 4ku 2.0, Black Marlin 7.0, Black Marlin 9.0, Chal 1.3.2, Cinder 0.6.1, Combusken 2.0.0, Dumb 2.1, Frozenight 6.0.0, Koivisto 3.0, Leorik 2.1, Mess 0.3.0, Minke 4.0.0, Minke 6.0.0, PZChessBot 7.1, PlentyChess 0.3.0, Princhess 0.22.0, Reckless 0.9.0, Renegade 0.12.0, Stormphrax 8.0.0, Velvet 1.1.0, Velvet 3.1.0, Velvet 3.3.0, Winter 4.0, Xiphos 0.6, Zahak 9.0, akimbo 0.7.0, akimbo 1.0.0, avalanche 5.0.0, cinder 0.6.1, halogen 16.0.0, hobbes 3.0, justbot 0.5.0, koivisto 9.0, minke 7.0.0, prune 4.0.1, pzchessbot 7.1, quanticade_cronus 3.0, raphael 4.2.0, reckless 0.9.0, renegade 1.3.0, stormphrax 8.0.0, tcheran 13.0, tcheran 14.0, viridithas 20.0.0 and zangdar 7 are named by word mark and version because the measurement conditions require it; no connection to their authors and no endorsement by them exists.