Skip to content

Milestones

List view

  • v1.0 - Production & Optimization

    GraalVM native image for fast startup, container/deployment optimization, release automation, and Oracle Cloud Free Tier (Ampere ARM64) deployment configurations.

    No due date
    •0/4 issues closed
    0% complete4 open0 closed

  • v0.8 - Self-Play & Book Distillation

    In-engine self-play generation (TurnGenerator-based) to feed training flywheels, and a self-play-distilled opening book to replace the human-game book (measured weaker in arena: booked vs bookless 327-316).

    No due date
    0% complete0 open0 closed

  • v0.7 - Fast Learned Evaluation (NNUE track)

    Engine-side support for a fast learned evaluator: quantized integer inference with an incremental (NNUE-style) accumulator inside make/undo, replacing ONNX Runtime on the search hot path. The designated remaining strength lever once search is squeezed (value-model roadmap step 3c).

    No due date
    •1/1 issues closed
    100% complete0 open1 closed

  • v0.6 - Star Pruning & Search Depth 3

    Star1/Star2 chance-node pruning (Ballard *-minimax), configurable search depth with depth-3 paired-SPRT evaluation, and Ox structured-concurrency parallelism for chance subtrees. Tracked by Epic #56; design in the internal wiki (model/star-pruning-depth-3).

    No due date
    •6/6 issues closed
    100% complete0 open6 closed

  • v0.5 - Search Foundations: Zobrist & TT

    Zobrist hashing (position, active color, castling, en passant, dice pool), a transposition table with bound-typed entries, and search/evaluation hot-path groundwork (KCP optimization). Precondition for search depth 3.

    No due date
    •4/4 issues closed
    100% complete0 open4 closed