v0.6.3

Added

• Added a release-orientation documentation spine covering use cases, market
  value, onboarding routes, tutorial selection, notebook production paths, and
  API-by-job navigation.
• Added JAX-backed information-geometry and multiverse rollout paths, expanded
  multiverse scenario coverage, added strange-loop drift scenario gates, and
  added review-only sheaf obstruction control proposals with deterministic
  evidence surfaces.
• Added public Studio facade exports for the integrated-information,
  strange-loop, and morphogenetic-field passive review panels.
• Added passive Studio surfacing for multiverse counterfactual rollout
  manifests and branch-risk reports.
• Added explicit local NumPy statevector and density-matrix simulator backend
  contracts plus passive Studio surfacing for hybrid classical-quantum
  entanglement order-parameter evidence.
• Added passive Studio surfacing for information-geometry control proposal
  audit records and deterministic domain scenario evidence.
• Added passive Studio surfacing for Topos semantic-binding validation reports
  and deterministic domain obligation examples.
• Added passive Studio surfacing for evolutionary supervisor policy-search
  reports, enriched domain examples, and policy-DSL mutation evidence.
• Added a standalone Studio product manifest that registers passive physics
  review panels, carries a deterministic manifest hash, and exposes them to the
  packaged Streamlit shell without enabling execution, transport, hardware, or
  actuation paths.
• Added explicit Pareto objective constraints to the value-alignment supervisor
  guard, including fail-closed missing-objective evidence, counterfactual
  Pareto violation audit records, and fallback preservation.
• Added a lightweight Python API layer with Orchestrator.from_yaml(...) and
  Orchestrator.run(...), plus the scpn import alias for onboarding scripts
  that should not depend on CLI invocation.
• Added a real-data review demo for PhysioNet heart-rate-belt data through
  spo demo --dataset heartbeat.csv --target coherence --steps N, including
  dataset citation output, review-only auto-binding, and dashboard guidance.

Changed

• Changed runtime request limiting from fixed minute windows to token-bucket
  semantics so callers cannot consume a full per-minute allowance in a single
  burst without bucket capacity.
• Made action-projector slew limits binding-configurable through
  actuators[].rate_limit_per_step while preserving conservative defaults for
  omitted binding metadata.
• Documented Modbus/TLS as requiring server certificate verification for both
  deployment CA bundles and the operating-system trust store path.

Fixed

• Hardened UPDE, Hodge decomposition, E/I balance, chimera, Lyapunov,
  plasticity, coupling-lag, K_nm geometry, spectral-coupling, SSGF, NPE,
  ITPC, embedding, order-parameter, swarmalator, splitting, simplicial,
  reduction, PAC, and market numerical boundaries around real-valued inputs,
  zero self-coupling, exact backend-output preservation, typed direct
  Go/Julia/Mojo validation, and object-dtype complex alias rejection.
• Hardened auto-binding YAML generation by emitting structured YAML instead of
  interpolating raw source column names into text.
• Hardened symbolic binding compilation with intent sanitisation, schema
  validation, and explicit review-gate metadata before any generated binding is
  treated as operational evidence.
• Hardened demo domainpack loading against path traversal outside
  domainpacks/.
• Seeded audit hash chains with the header record content so replay integrity
  includes the stream header.
• Hardened QueueWaves WebSocket ingress to accept only keepalive messages from
  clients and reject arbitrary inbound payloads.
• Hardened Petri-net guard evaluation by rejecting unknown guard context keys
  and non-finite guard values before transition evaluation.
• Hardened bifurcation continuation and critical-coupling fast paths by
  rejecting non-zero self-coupling diagonals, invalid composite Rust
  R(K) arrays, and invalid Rust K_c results.
• Hardened typed-array maintenance by replacing residual runtime-only
  np.ndarray annotations in monitor and direct accelerator validation helpers
  with parameterised NumPy payload aliases, and strengthened the source hygiene
  test to block regressions.
• Hardened multiverse counterfactual rollouts by rejecting object-dtype boolean
  and complex aliases before phase, frequency, coupling, phase-lag, or topology
  mask payloads are accepted for non-actuating branch simulation.
• Hardened information-geometry control proposals by rejecting complex
  distribution and gradient aliases before simplex normalisation or
  natural-gradient proposal generation.
• Hardened value-alignment policy boundaries by rejecting NumPy boolean
  aliases in direct constraint, score-threshold, and Pareto objective scalars.
• Hardened information-geometry scenario fixtures by rejecting complex
  distribution and record-gradient aliases before deterministic audit hashing.
• Hardened multiverse domain scenario fixtures by rejecting boolean and complex
  phase, frequency, and branch-knob aliases before deterministic review hashing.
• Hardened the FusionCore bridge boundary by rejecting non-positive q-profile
  bounds, negative fusion observables, negative stability ratios, and empty
  phase-feedback vectors before phase encoding or order-parameter export.
• Hardened the SCPNControl bridge boundary by rejecting empty, boolean-alias,
  complex-alias, and non-zero self-coupled K_nm imports plus empty or
  non-real natural-frequency imports before orchestration state conversion.
• Hardened the SNN controller bridge by bounding UPDE order-parameter
  magnitudes, rejecting negative spike rates, and rejecting boolean or complex
  aliases before LIF current, rate, and neuromorphic schedule conversion.
• Hardened the Neurocore bridge by rejecting negative deterministic seeds and
  boolean-alias rate vectors before action projection or Rust backend output
  acceptance.
• Hardened the LSL BCI bridge by rejecting boolean EEG samples and malformed
  sample timestamps before buffered Hilbert phase extraction.
• Hardened the hardware I/O boundary by rejecting boolean and complex
  sample/frequency aliases before buffering or synthetic EEG generation.
• Hardened the Prometheus adapter decoded sample-pair boundary by rejecting
  boolean, negative, and non-real sample timestamps before returning range or
  instant metric values.
• Closed the named Gaian/Prometheus JSON deserialisation roadmap scope and
  moved future adjacent JSON reviews into public boundary maintenance.
• Expanded the artifacts API reference with QPU data artifact schema, safety,
  hash, JSON, failure-mode, and file I/O contracts plus a dedicated docs guard.
• Expanded the scaffold API reference with review-only proposal, provider,
  validation, audit-record, CLI, and operator-review contracts plus a dedicated
  docs guard.
• Expanded the Python facade API reference with import, binding-validation,
  deterministic-run, state-record, and safety-boundary contracts plus a
  dedicated docs guard.
• Expanded the distributed sync API reference with phase-gossip wire-message,
  digest, sequence-watermark, bounded-correction, replay, and audit contracts
  plus a dedicated docs guard.
• Expanded the coupling inference API reference with transfer-entropy,
  orientation, thresholding, backend-failure, audit-record, CLI, and
  review-boundary contracts plus a dedicated docs guard.
• Expanded the visualization API reference with JSON-encoder, streamer,
  validation, deterministic-rounding, pipeline-wiring, typed-array, and
  presentation-boundary contracts plus a dedicated docs guard.
• Expanded the Poincare monitor API reference with recurrence, result-record,
  backend-validation, accelerator-parity, fallback, sample-time, and
  observational-boundary contracts plus a dedicated docs guard.
• Aligned Gaian mesh peer and local phase validation with circular phase
  semantics by wrapping finite negative psi values modulo 2*pi.
• Hardened the PlasmaControl bridge boundary by rejecting boolean numeric
  aliases, non-zero layer self-coupling, empty phase snapshots, negative plasma
  ratios, and non-positive safety-factor minima before coupling expansion or
  phase-state export.
• Aligned the Python Lyapunov reference QR routine with the Rust/Go/Mojo
  two-pass modified Gram-Schmidt tangent-basis contract so finite-horizon
  spectra remain reproducible across built backend artifacts.