Maia 2 Rapid - ChessIQ offline and browser engine Official Maia-2 code: Copyright (c) 2024 CSSLab. MIT License, included in LICENSE.txt. https://github.com/CSSLab/maia2 Source: v0.11.0, commit 230d982b1d5dece7d6c94945a5b9a09b3f4246c9 Paper: https://arxiv.org/abs/2409.20553 Authors: Zhenwei Tang, Difan Jiao, Reid McIlroy-Young, Jon Kleinberg, Siddhartha Sen, Ashton Anderson. Uses the official Rapid checkpoint, converted by ChessIQ on 2026-10-13. The policy and scalar value heads are unchanged. The unused auxiliary training head is omitted from the ONNX graph. No quantization or retraining is applied. The model is split into integrity-checked transport chunks, reassembled locally. Model provenance, input/output hashes and conversion dependencies: model.json. Rebuild procedure and validation: SOURCE_STATUS.txt. The ChessIQ Maia-2 bridge files model.js, uci.js, worker.js and download.js are distributed with this engine's MIT license. They do not include Maia-3 or maia3-js code. This notice does not license the private ChessIQ application. Board encoding follows the official Maia-2 implementation; chess.js provides legal move generation, under its included BSD-2-Clause license. ONNX Runtime Web 1.30.0: Copyright Microsoft Corporation, MIT License. See ort/LICENSE.txt and ort/ThirdPartyNotices.txt. Maia predicts human move choices in a single evaluation. Its rating bins are under 1100, 1100-1199 through 1900-1999, and 2000+. Probabilities are estimated human move likelihoods, not confidence that a move is best. A longer move delay does not deepen the prediction. Positions/sec measures actual root inferences, not the number of possible moves or Stockfish search-tree nodes. Provided without warranty; see LICENSE.txt. Engine notices do not constitute App Store approval or a license exception for other bundled components.