Pure-MoonBit port of doubleml-for-py: Double / Debiased Machine Learning for partial linear, IRM, IV, DID, SSM, APO(S), PQ, QTE, LPQ, CVaR, RDD, BLP, and policy-tree models

PDF backend for asciidoctor.mbt: a port of Asciidoctor PDF 2.3.27 with its themes and fonts bundled

Prawn's layout behaviour for MoonBit on moonbitlang/pagelayout: the document cursor and bounds, formatted text, font metrics, prawn-svg and prawn-table sizing

functional Web UI framework for MoonBit

Bit-exact MoonBit port of SpikingNeuralNetworks.jl, extended with modern training machinery. NEURONS AND PLASTICITY: IF, AdEx, Izhikevich (IZ), HH, MorrisLecar, Poisson (including InhomogeneousPoisson), WilsonCowan; Markram STP, Gerstner / MexicanHat / AntiSymmetric / vSTDP / Confavreux2025 STDP; AMPA / NMDA / GABAa / GABAb receptors; multicompartment dendritic neurons (BallAndStick, Tripod, Multipod); IF with population synaptic conductance; spike-time and current stimuli. CORE DL: Tensor struct with broadcasting + softmax / log-softmax / cross-entropy, Conv2d / MaxPool2d / GlobalAvgPool2d / ReLU / Flatten / Linear forward AND backward, ResNet foundations, generic Chain / Sequential, pre-made MLP / SimpleCNN / LeLen5 bottles, image adapter (BMP/QOI/TGA/PNG/GIF/JPEG/ICO/TIFF) via riantr/moonbit_image, 5D [T,B,C,H,W] STImage. OPTIMIZERS AND SCHEDULING: SGD, SGD-with-momentum, Adam with bias correction, AdamW, RMSprop, AdaGrad, Adafactor; StepLR, ExponentialLR, CosineAnnealingLR, ReduceLROnPlateau; BatchNorm2d with running stats and LayerNorm per-sample, both with backward. SURROGATE GRADIENTS AND ATTENTION: fast-sigmoid (Zenke & Ganguli 2018) for BPTT through IF/LIF Heaviside; GELU, exact sparse GELU via libm erff; multi-head self-attention with mask utilities; sinusoidal and learnable positional encoding; Pre-Norm transformer block; inverted dropout; SAttention unifying three spiking attention variants; LayerScale; T5-style relative position bias; SpikingAttention with full BPTT backward; SpikeFormer demo. RL: SequenceReplayBuffer for recurrent RL, DDPG, TD3, SAC, GRU/LSTM/GTrXL recurrent actors and critics, Decision Transformer, Trajectory Transformer with beam search. META-LEARNING AND CAUSAL: MAML, FOMAML, Reptile, TaskSampler; LinearGaussianSCM, DoCalculus, CounterfactualReasoning, CausalEffectEstimator. GENERATIVE: RealNVP / Glow / NormalizingFlow, VAE, IWAE, ELBO, DDPM diffusion, EnergyBasedModel with Langevin sampling, DCGAN, WGAN-GP. OTHER: Vision Transformer with trainer, EmbeddingNet / SiameseNet / SiameseTrainer, CapsuleNetwork. GRAPH NEURAL NETWORKS: Graph with COO edge list and GCN adjacency normalisation; GCN, MPNN, GIN, PNA, GAT (multi-head), Set2Set readout, EdgeGNN, H2GCN, GraphClassifier -- and FULL BACKWARD PASSES for all of them, dispatched behind a single TrainableGraphNet enum, with node-level and graph-level training steps. VERIFICATION: two runnable harnesses, since `moon test` cannot link this package on Windows (998 .mbt files exceed the 32K command-line limit). `verify/verify_gnn_grads.ps1` runs a finite-difference gradient gate over 8 dL/dh configurations, 8 dL/dW configurations, the multi-head GAT stack, a deterministic ELU-mask row, and a cross-entropy loss-vs-gradient consistency section -- judged by relative tolerance, by a kink-versus-bug component count, and against negative controls. `verify/run_gnn_train_demo.ps1` runs an end-to-end transductive two-community training demo over all five backbones with a held-out split, a shuffled-label control and a no-message-passing baseline. These harnesses found 13 real defects that the type checker and code review had both passed, including a cross-entropy loss whose sign was inverted for six releases. Float32 end to end with libm FFI (expf, tanhf, logf, sqrtf, cosf, sinf, erff).

The Starlark configuration language, implemented in Moonbit

WHATWG-compliant URL and URLPattern library for MoonBit

LLM Agent framework with hexagonal (ports-and-adapters) architecture. Defines extension ports + Agent loop. Depends on moonbitlang/async and colmugx/fuwaroid.

MoonDBC: CAN / CAN FD DBC parsing, validation, multiplexed frame codec and CLI

MoonBit-native IPP codec, capability validation and network printing client

A date and time library for MoonBit

A port of the Typst typesetting engine to MoonBit.

MoonDBC: CAN / CAN FD DBC parsing, validation, multiplexed frame codec and CLI

A date and time library for MoonBit

MoonBit-native IPP codec, capability validation and network printing client

Local incremental DAG task execution with cache and recovery

Layered configuration merge, interpolation and provenance audit

Typed rule evaluation, decision tables and explainable decisions

Local incremental snapshots, verified restore and recovery

Property graph import, indexing, query and impact analysis

API contract compatibility classification and response verification

Persistent full-text indexing and ranked search

Offline event normalization, query and request correlation

Typed tabular import, validation and transformation

experimental packages for moonbitlang/core

Asynchronous programming library for MoonBit

Cross-platform MoonBit compatibility facades with native delegation and accelerated JS implementations for async IO, bytes, regexp, and JSON.

Pure MoonBit zlib/deflate implementation.

Automatic testing of MoonBit programs

AST and parsers for the MoonBit programming language

Workflow engine core with Starlark-subset parser, DAG execution, and cache planning primitives.

Temporary file and directory management for MoonBit

Pure MoonBit LLM client library

A generic lexer library for handwritten lexers in MoonBit

A TOML parser implementation in MoonBit

A LR(1) parser generator for MoonBit programming language.