v0.1.0

First public release. Alpha-stage research kernel.

Kernel

• Reusable substrate primitives: EchoStateSubstrate, DelayLineSubstrate,
  LinearMemorySubstrate, OscillatoryMemorySubstrate, MixedMemorySubstrate,
  HierarchicalSubstrate, plus the factories config-driven dispatch.
• Memory primitives: ExactContextMemory, NgramMemory,
  StatisticalBackoffMemory, OnlineCausalMemory, plus ExactContextCache
  and StatisticalBackoffCache view layers.
• Control surfaces: ControllerSummary, PredictiveController,
  LearnedSegmenter, PathwayGateController, SummaryRouter,
  HormoneModulator, PredictiveSurpriseController.
• Feature views: ByteLatentFeatureView, HierarchicalFeatureView,
  LinearMemoryFeatureView, SampledMultiscaleReadout,
  ProbabilityDiagnostics, bridge_feature_arrays,
  BidirectionalContextProbe.
• Readouts and experts: RidgeReadout, FrozenReadoutExpert.
• Adapters: CausalPredictiveAdapter, NoncausalReconstructiveAdapter,
  OracleAnalysisAdapter, BridgeExportAdapter, TeacherExportAdapter,
  ByteLatentPredictiveCoder.
• Causal-bank family: CausalBankConfig, build_linear_bank,
  validate_config, scale_config, learnable_substrate_keys. Substrate
  modes frozen, learnable_decays, learnable_mixing, learned_recurrence,
  gated_retention. Optional memory attachment, stacked blocks, selective
  scan, banded readouts, byte-to-patch encoding, fast/slow hemispheres,
  polynomial expansion, training noise, adaptive regularization.
• Runtime: CausalTrace, FitReport, evaluate_rollout, score_next_step,
  evaluate_dataset, evaluate_rollout_curve, evaluate_transfer_probe,
  TrainModeConfig, ArtifactAccounting, artifact audits.

Backends

• decepticons.models.causal_bank_torch.CausalBankModel — PyTorch
  implementation with frozen substrate, selective scan augment, banded readout.
  Optional installs: pip install -e ".[torch]".
• decepticons.models.causal_bank_mlx.CausalBankModel — MLX equivalent.
  Optional installs: pip install -e ".[metal]".

CLI

• decepticons fit — single-command fit + sample over a UTF-8 corpus.

Verification

• tests/test_causality.py verifies every substrate mode is causal under
  perturbation. CI fails if any future-leak is detected.
• tests/test_dependency_firewall.py AST-scans the kernel to ensure
  decepticons never imports its descendants.