Runs

Every register entry, newest first. Short and long stage are labelled separately; the quoted value is the long stage when there is one and is never added to or averaged with the short stage. Verdicts without games name their evidence. This table is not a highlight reel.

Register export 2026-09-08 · Register T-0001 – T-0072 · 72 entries · Run feed (RSS)

How each run ended

How each run ended 39 runs passed, 17 failed, 4 were rejected; 10 of the passed runs played no games. passed failed or rejected passed without games Epoch 2 Epoch 3 Epoch 4 +2.94 0 −2.94 upper bound lower bound T-0013 2026-08-29 T-0028 2026-08-29 T-0043 2026-08-30 T-0057 2026-09-03 T-0072 2026-09-08
One mark per run in register order, final log-likelihood ratio of the quoted stage. Every SPRT run starts at 0; it passes at the upper bound and fails at the lower one. Hollow marks passed without games (identity, correction, epoch change, adoption) and sit outside the bounds, not as an SPRT exit.

Register T-0072 – T-0013

Run Verdict Elo Strength difference to the opponent, estimated from the games, with a 95-percent half-width where the oracle reported one. Never an absolute rating. Date Games LLR Log-likelihood ratio, the running evidence of an SPRT. It starts at 0; a stage passes at the upper bound (about +2.94) and fails at the lower one (about −2.94). Hypothesis
T-0072 passed
long stage +260.44 ± 39.40
short stage +222.52 ± 36.23
378 +2.96 Together with the int16 evaluation core, the third network generation replaces the hand evaluation and gains significantly in playing strength compared with the bound predecessor.
T-0071 rejected
long stage +1.05 ± 1.65
short stage +2.71 ± 2.11
76,396 −2.97 If negamax clamps its alpha-beta window to the mate values reachable at ply [-Mate + ply, Mate - ply - 1], the fixed-depth result and mate distance are preserved, while impossible subtrees are eliminated.
T-0070 passed epoch change, no games Epoch change of the same predecessor source
T-0069 rejected
short stage +0.71 ± 2.51
44,794 −2.96 A session-wide correction of systematic raw evaluation errors, indexed by pawn structure and side to move, can make RFP, null move, Futility and QSearch stand-pat more reliable.
T-0068 rejected
short stage −7.07 ± 6.81
6,196 −2.95 If the main search at depth at most 3 omits materially losing captures, weakly contributing subtrees can be saved without touching the explicitly protected cases.
T-0067 passed
long stage +16.22 ± 8.31
short stage +7.60 ± 5.35
3,730 +2.97 If a sufficiently deep non-PV Principal variation: the line of best moves the search currently expects. node without a TT Transposition table: a cache of already searched positions, keyed by a hash of the position. move is reduced by exactly one level, the worse-ordered search costs less and the saved budget can produce additional depth at more important nodes.
T-0066 passed
long stage +17.47 ± 8.64
short stage +24.65 ± 10.49
3,364 +2.95 Using the already maintained occupancy bitboards in scan_board and see preserves search behavior and reduces computational effort.
T-0065 passed
long stage +6.74 ± 4.86
short stage +5.82 ± 4.43
11,088 +2.97 Omitting the lower Zobrist hashing A way of computing a position hash incrementally: each piece on each square has a random number, and the hash is their combination. bits already determined by the table index enables 16-byte TT entries and doubles the number of slots with the same hash budget; this could reduce TT collisions and search effort.
T-0064 passed
long stage +11.72 ± 6.91
short stage +3.13 ± 2.44
5,428 +2.95 A generation-aware, depth-preferred replacement preserves more valuable Transposition table entries and brings measurable Elo compared with unconditional replacement.
T-0063 passed byte-identical, no games A pure comment line under `engine/src/` leaves the binary unchanged; the identity path can record the candidate as passed without games.
T-0062 passed epoch change, no games Epoch change of the same predecessor source
T-0061 passed correctness proof, no games Correction K7: counters saturate at INT_MAX, movetime capped; behaviorally identical to the predecessor (manifest empty)
T-0060 passed correctness proof, no games Correction K2: FEN Forsyth–Edwards Notation, a one-line text description of a chess position. parsing checks the plausibility of the en-passant square; behaviorally identical to the predecessor (manifest empty)
T-0059 passed correctness proof, no games Correction K1: draw.inc counts knights correctly; behaviorally identical except for three manifest positions
T-0058 passed correctness proof, no games Correction K3: evaluate() clamps the side-relative value to [-28999,28999]; behaviorally identical to the predecessor (manifest empty).
T-0057 passed byte-identical, no games A pure comment line under engine/src/ leaves the binary unchanged; the identity path then records the verdict without games.
T-0056 passed epoch change, no games Epoch change of the same predecessor source
T-0055 rejected
short stage −0.17 ± 2.99
25,880 −2.94 Infrastructure test on the compute machine: predecessor source against itself (live path cell → SFTP → build → oracle).
T-0054 passed adoption, no games Takeover of the running predecessor
T-0053 failed
short stage +0.38 ± 2.52
36,362 −2.95 Reporting forced mates as signed UCI Universal Chess Interface, the text protocol between a chess engine and the program that runs it. move counts should make Coherent's analysis output correct for protocol clients without changing the search tree or bench depth.
T-0052 passed
long stage –
short stage +17.17 ± 8.88
486 +2.95 Giving a side with at least two bishops a tapered bonus of 30 centipawns in the middlegame and 50 in the endgame should strengthen Coherent by valuing their complementary color control and greater reach as the board clears.
T-0051 failed
short stage −13.95 ± 9.25
3,664 −2.95 Tapered pawn-structure scores should improve endgame move choices by increasing doubled- and isolated-pawn penalties with the endgame pawn value and doubling passed-pawn bonuses as non-pawn material disappears.
T-0050 failed
short stage −9.19 ± 7.57
5,028 −2.95 Persisting history and countermove heuristics across searches in a game should improve quiet-move ordering by avoiding a cold start on every move.
T-0049 passed
long stage +1,194.89
short stage +4.79 ± 3.80
486 +2.95 Probing current-generation transposition bounds in quiescence and prioritizing a matching tactical table move should reduce repeated evaluation, move-generation, and SEE Static exchange evaluation: an estimate of the material outcome of a capture sequence on one square, without searching. work without replacing deeper principal-search entries.
T-0048 failed
short stage −2.91 ± 5.01
11,566 −2.96 Pruning non-promoting, non-checking captures with negative SEE at non-PV nodes through depth two should strengthen Coherent by avoiding full searches of exchanges already known to lose material.
T-0047 failed
short stage −12.54 ± 8.63
3,882 −2.95 Preserving castling rights should improve move choices by assigning a phase-tapered value to each retained castling option that the king table and pawn shield do not represent.
T-0046 failed
short stage −4.43 ± 5.30
8,152 −2.97 Configurable 50 ms move overhead and a per-move cap of half the remaining clock should reduce time forfeits by preserving transmission reserve and preventing single searches from exhausting the clock.
T-0045 passed
long stage +1,075.01
short stage +44.77 ± 14.83
488 +2.96 Deferring quiescence legality checks until a move is tried and inlining debug-checked board access should strengthen Coherent by avoiding duplicate make, check, and unmake work, especially for full check-evasion lists.
T-0044 passed
long stage +1,194.89
short stage +2.75 ± 2.13
486 +2.95 Coherent should search more positions in the allotted time because pseudo-legal move generation enumerates only the moving side's occupied squares from incrementally maintained masks while preserving move order.
T-0043 passed
long stage +1,194.89
short stage +7.83 ± 5.35
486 +2.95 Preallocating the pseudo-legal move vector should strengthen Coherent because it removes the dominant repeated growth allocations while preserving move order and retaining automatic vector growth as a safe fallback.
T-0042 failed
short stage −9.79 ± 7.47
4,542 −2.96 Restricting quiescence stalemate checks to low-material positions without an unblocked pawn should correct drawn endgame scores without paying for legal move generation on common stand-pat exits.
T-0041 failed
short stage −11.28 ± 8.15
4,190 −2.96 Limiting each line to two check extensions should improve move selection by preserving the first two tactical extensions while preventing long checking sequences from monopolizing the search budget.
T-0040 passed
long stage +1,194.89
short stage +12.98 ± 7.54
486 +2.95 Retaining transposition moves across searches in an explicit game session should strengthen play by improving move ordering on successive positions while generation-scoped bounds avoid reusing clock- or repetition-dependent scores.
T-0039 failed
short stage −4.22 ± 5.66
9,136 −2.96 Increasing the maximum quiescence depth from six to eight plies should strengthen Coherent by resolving longer forcing capture sequences now that tactical move generation and pruning keep the added search cost modest.
T-0038 failed
short stage −0.32 ± 1.85
21,706 −2.97 Correctly treating bishop-and-knight and two-knight endings as playable rather than automatic draws should strengthen endgame decisions because checkmate remains legally possible in both material configurations.
T-0037 failed
short stage −10.98 ± 6.75
3,292 −2.97 Scoring stalemate as a draw in quiescence should prevent static material evaluations from favoring drawing continuations in otherwise winning endgames.
T-0036 passed
long stage +1,075.01
short stage +3.22 ± 2.43
488 +2.96 Coherent should search more nodes within its time limit because child and PVS re-searches reuse the parent's already computed check status instead of scanning attacks again.
T-0035 passed
long stage +1,075.01
short stage +3.11 ± 2.35
488 +2.96 Searching a valid transposition-table move before ordering the remaining moves should make the engine stronger by avoiding unnecessary SEE evaluations and sorting when that move causes an immediate cutoff.
T-0034 failed
short stage −2.47 ± 3.06
8,440 −2.97 Coherent should play more strongly because replacing full-board Undo snapshots with reversible move metadata reduces copy traffic in every legality check and searched move.
T-0033 passed
long stage +1,075.01
short stage +10.82 ± 5.83
488 +2.96 Incrementally cached king squares should improve playing strength by eliminating repeated board scans from check detection while preserving the search tree.
T-0032 passed
long stage +1,075.01
short stage +40.25 ± 13.25
488 +2.96 Ordering pseudo-legal moves and rejecting self-check only for each attempted move should strengthen search by avoiding full-list legalization at nodes that cut off after examining only a few moves.
T-0031 passed
long stage +1,075.01
short stage +48.65 ± 15.12
488 +2.96 Generating only legal captures, en passant moves, and promotions outside check should strengthen play by avoiding make/unmake work for quiet moves that quiescence discards while preserving the search tree.
T-0030 failed
short stage −18.71 ± 10.88
2,974 −2.95 Counting distinct pieces and attacked squares around the enemy king should improve tactical threat recognition by rewarding coordinated attacks rather than general mobility alone.
T-0029 passed
long stage +1,075.01
short stage +5.75 ± 4.38
488 +2.96 Search strength should improve because deeper or clearly favorable nodes use larger null-move reductions while high-depth verification rejects unsafe cutoffs in zugzwang-prone positions.
T-0028 passed
long stage –
short stage +22.28 ± 10.29
486 +2.95 Depth- and move-count-dependent reductions of up to three plies for late quiet moves, softened at principal-variation nodes and for killer, countermove, or high-history moves and followed by a full-depth re-search after an alpha raise, should make Coherent stronger by spending fewer nodes on unlikely late alternatives.
T-0027 passed
long stage –
short stage +13.24 ± 7.72
486 +2.95 Search should become stronger because penalizing quiet moves that fail before a quiet beta cutoff keeps history scores discriminative, while a countermove table prioritizes context-specific cutoff replies.
T-0026 failed
short stage −11.11 ± 8.42
4,474 −2.95 Verified razoring should strengthen Coherent by replacing full searches of shallow fail-low nodes with cheaper quiescence searches while preserving tactically forcing replies.
T-0025 passed
long stage +1,075.01
short stage +60.64 ± 17.22
488 +2.96 Pruning non-checking, non-promotion quiescence captures that cannot reach alpha by a material delta bound or have a negative static exchange evaluation should strengthen Coherent by spending fewer nodes on unpromising leaf exchanges.
T-0024 passed
long stage +1,075.01
short stage +21.14 ± 10.17
488 +2.96 Late move pruning should strengthen Coherent by spending shallow non-PV search effort on earlier ordered and forcing moves instead of unlikely late quiet moves.
T-0023 failed
short stage −2.06 ± 4.58
14,006 −2.95 Using non-cutting transposition-table bounds to narrow alpha-beta windows should reduce redundant search because previously proven score limits can trigger earlier cutoffs.
T-0022 passed
long stage +1,194.89
short stage +153.43 ± 28.10
486 +2.95 Coherent should make stronger moves under tournament time controls because phase-aware soft time limits save clock in stable positions and let unstable positions search longer within a reserved hard limit.
T-0021 passed
long stage +1,004.40
short stage +19.28 ± 9.17
488 +2.95 Searching forced check evasions beyond the quiescence depth limit and including quiet promotions should strengthen tactical play by avoiding static evaluations of positions where a legal forcing move remains.
T-0020 failed
short stage −20.67 ± 10.40
2,390 −2.98 The engine should play more strongly because it will preserve winnable minor-piece endings, recognize repetitions despite irrelevant en-passant fields, and keep halfmove- and path-dependent draw scores from contaminating the transposition table.
T-0019 failed
short stage −0.58 ± 2.76
20,294 −2.95 Depth-preferred replacement with generation aging should improve move selection in longer games by preserving deeper transposition results from shallower collisions while allowing stale entries to yield immediately.
T-0018 passed
long stage +1,194.89
short stage +54.24 ± 16.96
486 +2.95 Scoring pseudo-legal destination squares for knights, bishops, rooks, and queens should improve move selection by preferring active pieces without duplicating the existing pawn-structure and king-safety terms.
T-0017 passed
long stage +915.30 ± 230.58
short stage +11.00 ± 7.01
488 +2.95 Splitting the piece values and the piece-square table into separate middlegame and endgame columns and interpolating between them by the non-pawn material still on the board should play endgames better, above all because the single table previously told the king to centralise in every phase, including the middlegame where it belongs behind its own pawns.
T-0016 passed
long stage –
short stage +67.30 ± 18.50
488 +2.96 Futility pruning at shallow depths — cutting a node outright when its static evaluation stands a pawn per remaining ply above beta, and skipping quiet moves when it stands that far below alpha — should shrink the tree enough to reach greater depth in the same time, with both cuts disabled in check and in pawn-only endgames where the static evaluation cannot be trusted.
T-0015 passed
long stage +954.24 ± 369.98
short stage +43.66 ± 11.71
488 +2.95 The search now scores threefold repetition, the fifty-move rule and insufficient material as draws, checked before the transposition-table cut because two of the three depend on the path to the position, so the engine should stop repeating won positions into draws and stop valuing material it can never mate with.
T-0014 passed
long stage +622.70 ± 72.31
short stage +541.17 ± 75.80
500 +2.95 In quiescence search a side to move that is in check no longer stands pat but searches every legal evasion, quiet moves and promotions included, and reports mate when there is none, so the value returned at the horizon comes from a continuation that actually exists.
T-0013 passed
long stage +593.01 ± 62.45
short stage +528.89 ± 74.59
502 +2.95 Extending the main search by one ply whenever the side to move is in check should improve play in forcing lines, above all at the horizon, where a checked position was previously handed to a quiescence search that only sees captures and therefore evaluated it statically with the king under attack.