feat(hooks): Full IntelligenceEngine with MCP tools, trajectory tracking, and attention

  Built from commit 23fc783

  Platforms updated:
  - linux-x64-gnu
  - linux-arm64-gnu
  - darwin-x64
  - darwin-arm64
  - win32-x64-msvc

  🤖 Generated by GitHub Actions

Add complete integration examples for all 20+ @neural-trader npm packages
with the RuVector platform:

Core Integration:
- basic-integration.js: HNSW vector indexing with trading operations
- hnsw-vector-search.js: Pattern matching with 150x faster native search
- technical-indicators.js: 150+ indicators (RSI, MACD, Bollinger, etc.)

Strategy & Portfolio:
- backtesting.js: Walk-forward optimization, Monte Carlo simulation
- optimization.js: Markowitz, Risk Parity, Black-Litterman portfolios

Neural Networks:
- training.js: LSTM training for price prediction with RuVector storage

Risk Management:
- risk-metrics.js: VaR, CVaR, stress testing, position limits

MCP Integration:
- mcp-server.js: 87+ trading tools via Model Context Protocol

Accounting:
- crypto-tax.js: FIFO/LIFO/HIFO cost basis with native Rust bindings

Specialized Markets:
- sports-betting.js: Arbitrage detection, Kelly criterion sizing
- prediction-markets.js: Polymarket/Kalshi expected value analysis
- news-trading.js: Sentiment-driven event trading

Full Platform:
- platform.js: Complete trading system integration demo

Packages integrated:
- neural-trader@2.7.1 (core engine, 178 NAPI functions)
- @neural-trader/core, strategies, execution, portfolio, risk
- @neural-trader/neural, features, backtesting, market-data
- @neural-trader/mcp, brokers, predictor, backend
- @neural-trader/agentic-accounting-rust-core
- @neural-trader/sports-betting, prediction-markets, news-trading
- @ruvector/core for HNSW vector database

Advanced examples (production-grade):
- live-broker-alpaca.js: Production broker integration with smart order routing
- order-book-microstructure.js: VPIN, Kyle's Lambda, spread decomposition
- conformal-prediction.js: Distribution-free guaranteed prediction intervals

Exotic examples (cutting-edge techniques):
- multi-agent-swarm.js: Distributed trading with consensus mechanisms
- gnn-correlation-network.js: Graph neural network correlation analysis
- attention-regime-detection.js: Transformer attention for regime detection
- reinforcement-learning-agent.js: Deep Q-Learning trading agent
- quantum-portfolio-optimization.js: QAOA and quantum annealing
- hyperbolic-embeddings.js: Poincaré disk market embeddings
- atomic-arbitrage.js: Cross-exchange atomic arbitrage with MEV protection

Updated package.json with npm scripts for all new examples.
Updated README.md with documentation for advanced/exotic techniques.

Key fixes across exotic neural-trader examples:

- reinforcement-learning-agent.js: Fixed broken backpropagation that only
  updated output layer. Now stores activations and flows gradients through
  all hidden layers properly.

- quantum-portfolio-optimization.js: Fixed QAOA mixer Hamiltonian that was
  incorrectly accumulating all qubit operations. Now applies Rx rotations
  sequentially per-qubit with proper normalization.

- hyperbolic-embeddings.js: Fixed Math.acosh/atanh domain errors and
  implemented proper Riemannian gradient descent using expMap in Poincaré
  ball model.

- multi-agent-swarm.js: Added division-by-zero guards for linear regression,
  z-score calculation, and iterator type fixes. Added memory bounds.

- gnn-correlation-network.js: Added guards for betweenness normalization
  (n<3), density (n<2), and clustering/degree calculations (n=0).

- attention-regime-detection.js: Added empty array handling for softmax and
  matrix validation for transpose operations.

- atomic-arbitrage.js: Added guard for flash loan spread calculation.

Generated during deep review of exotic neural-trader examples.

The download logic would immediately fail if model.onnx wasn't at the
repo root, never trying the onnx/ subfolder where most sentence-transformer
models store their ONNX files.

Now tries both locations:
1. Root: {repo}/model.onnx
2. Subfolder: {repo}/onnx/model.onnx

Also applies fallback logic to auxiliary files (tokenizer.json, config.json).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Changed invalid category "machine-learning" to "algorithms".

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

New optional companion package using Tract for inference:
- Runs in browsers, Cloudflare Workers, Deno, edge environments
- Same API as native crate
- JavaScript bindings via wasm-bindgen
- Supports all pooling strategies (Mean, Cls, Max, etc.)

Uses Tract instead of ONNX Runtime for WASM compatibility.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…ples

- gnn-correlation-network.js: Added RollingStats class for O(1) incremental
  updates and correlation caching with TTL to avoid redundant O(n²) calculations

- attention-regime-detection.js: Optimized matmul with cache-friendly i-k-j
  loop order and added empty matrix guards

- quantum-portfolio-optimization.js: Added ComplexPool for object reuse to
  reduce GC pressure, plus in-place operations (addInPlace, multiplyInPlace,
  scaleInPlace) to avoid allocations in hot loops

- multi-agent-swarm.js: Added RingBuffer for O(1) bounded memory operations
  and SignalPool for signal object reuse

Validates core WASM bindings work:
- Version check
- Cosine similarity utility
- L2 normalization utility
- Config creation and chaining

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Fixes PreToolUse:Read hook error by supporting --silent flag
for Read/Glob/Task hooks that should not produce output.

Published ruvector@0.1.54

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Adds loader.js with:
- Pre-configured model URLs for 6 popular models
- ModelLoader class with caching and progress reporting
- createEmbedder() helper for quick setup
- embed() and similarity() one-liner helpers

Supported models:
- all-MiniLM-L6-v2 (default)
- all-MiniLM-L12-v2
- bge-small-en-v1.5
- bge-base-en-v1.5
- e5-small-v2
- gte-small

Usage:
```javascript
import { createEmbedder } from './loader.js';
const embedder = await createEmbedder('all-MiniLM-L6-v2');
const embedding = embedder.embedOne("Hello world");
```

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Added npm and crates.io version badges
- WebAssembly and MIT license badges
- Quick start examples for Browser, Node.js, and Cloudflare Workers
- Complete API reference for WasmEmbedder, WasmEmbedderConfig
- Model comparison table with 6 HuggingFace models
- Performance benchmarks and use case examples

Published to npm as ruvector-onnx-embeddings-wasm@0.1.0

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Enable WASM SIMD128 instructions for vectorized operations
- Update simd_available() to properly detect SIMD at compile time
- SIMD build is 180KB smaller than non-SIMD (more compact instructions)
- Published v0.1.1 to both npm and crates.io

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- ParallelEmbedder class using Node.js worker_threads
- Distributes batches across multiple CPU cores
- Benchmark results: 3.6-3.8x speedup on batch processing
- Per-text latency drops from ~390ms to ~103ms with 4 workers
- Published v0.1.2 to npm and crates.io

Usage:
  import { ParallelEmbedder } from 'ruvector-onnx-embeddings-wasm/parallel';
  const embedder = new ParallelEmbedder({ numWorkers: 4 });
  await embedder.init();
  const embeddings = await embedder.embedBatch(texts);

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Added SIMD badge and documentation
- Added ParallelEmbedder API reference and usage examples
- Updated performance benchmarks with parallel vs sequential comparison
- Added browser compatibility table
- Added changelog section
- Added batch processing use case example
- Updated build instructions with SIMD flags

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

New features in this release:
- 13 new CLI hooks commands (AST, Diff, Coverage, Graph, Security, RAG)
- 16 new MCP tools for Claude Code integration
- Enhanced hooks init with v2.0 templates
- 9-phase pretrain with attention/GNN capabilities
- Comprehensive CLAUDE.md documentation generation

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

… Coverage modules

Core Modules Added:
- ast-parser.ts: Full AST parsing with symbol extraction, complexity analysis
- onnx-embedder.ts: ONNX WASM embeddings with SIMD acceleration
- diff-embeddings.ts: Semantic diff analysis with risk scoring
- coverage-router.ts: Test coverage-aware agent routing
- graph-algorithms.ts: MinCut, Spectral clustering, Louvain communities
- graph-wrapper.ts: Code dependency graph builder
- cluster-wrapper.ts: RuVector cluster management
- router-wrapper.ts: Semantic routing wrapper
- parallel-intelligence.ts: Parallel processing orchestration
- parallel-workers.ts: Worker pool with work stealing

CLI Hooks (13 new commands):
- ast-analyze, ast-complexity
- diff-analyze, diff-classify, diff-similar
- coverage-route, coverage-suggest
- graph-mincut, graph-cluster
- security-scan, rag-context, git-churn, route-enhanced

MCP Server (16 new tools):
- hooks_ast_*, hooks_diff_*, hooks_coverage_*
- hooks_graph_*, hooks_security_*, hooks_rag_*
- hooks_attention_info, hooks_gnn_info

Enhanced Features:
- 10 attention mechanisms (Flash, Hyperbolic, MoE, GraphRoPe, etc.)
- 9-phase pretrain with neural capabilities
- ONNX WASM runtime (7.4MB) with all-MiniLM-L6-v2 support
- hooks init generates v2.0 templates with full documentation

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add Claude Code Intelligence v2.0 section with MCP tools
- Add ONNX WASM embeddings documentation
- Add Windows installation with --ignore-scripts
- Add new CLI commands (AST, Diff, Coverage, Graph, Security, RAG)
- Update hooks configuration with new env variables
- Add performance comparison table (40,000x cache speedup)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Added benchmark.js performance suite measuring:
- GNN correlation matrix construction
- Matrix multiplication (original vs optimized)
- Object pooling vs direct allocation
- Ring buffer vs Array.shift()
- Softmax function performance

Additional optimizations:
- attention-regime-detection.js: Optimized softmax avoids spread operator,
  uses loop-based max finding and single-pass exp+sum (2x speedup)
- gnn-correlation-network.js: Pre-computed statistics for Pearson correlation
  via precomputeStats() and calculateCorrelationFast() methods. Avoids
  recomputing mean/std for each pair. Spearman rank also optimized.

Benchmark results:
- Cache-friendly matmul: 1.7-2.9x speedup
- Object pooling: 2.7x speedup
- Ring buffer: 12-14x speedup
- Optimized softmax: 2x speedup

- Add Fractional Kelly engine (1/5th Kelly, 576K ops/s)
- Add Hybrid LSTM-Transformer predictor (1.8K predictions/s)
- Add DRL Portfolio Manager (PPO/SAC/A2C ensemble, 17K ops/s)
- Add Sentiment Alpha pipeline (3.7K signals/s)
- Add comprehensive benchmark suite and documentation

All modules production-ready with sub-millisecond latency.

- LSTM: pre-allocate gate vectors, inline sigmoid/tanh (avoid map/reduce)
- MultiHeadAttention: cache-friendly i-k-j matmul, optimized softmax
- FeedForward: pre-allocate hidden layer, manual loops
- LayerNorm: manual mean/variance computation
- Lexicon: char-based word extraction (avoid regex)

Key improvements:
- Buffer push: 1.1M/s (+67%)
- Buffer sample: 319K/s (+22%)
- Lexicon: 346K/s (+16%)

TensorCompress:
- Adaptive tensor compression with 5 levels (none, half, pq8, pq4, binary)
- Access-frequency based auto-compression
- Up to 10x memory savings for pattern storage
- CLI: compress, compress-stats, compress-store, compress-get
- MCP: hooks_compress, hooks_compress_stats, hooks_compress_store, hooks_compress_get

LearningEngine (9 algorithms):
- Q-Learning: Simple off-policy (default)
- SARSA: On-policy, conservative (error-avoidance)
- Double Q-Learning: Reduces overestimation (precise routing)
- Actor-Critic: Policy gradient + value (confidence scoring)
- PPO: Stable policy updates (preference learning)
- Decision Transformer: Sequence modeling (trajectory patterns)
- Monte Carlo: Full episode learning (unbiased estimates)
- TD(λ): Eligibility traces (credit assignment)
- DQN: Experience replay (high-dim states)

CLI Commands:
- learning-config: Configure algorithms per task
- learning-stats: View algorithm performance
- learning-update: Record experiences
- learn: Combined action + recommendation

MCP Tools:
- hooks_learning_config, hooks_learning_stats
- hooks_learning_update, hooks_learn
- hooks_algorithms_list

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add defensive check for this.data.stats
- Initialize with defaults if missing
- Fixes "Cannot read properties of undefined (reading 'last_session')"

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Check this.data exists before accessing properties
- Initialize trajectories array if missing
- Add try/catch around engine and save operations
- Null-check trajectory objects in filter

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…ess (v0.1.69)

## Multi-Algorithm Learning Engine
- 9 algorithms: q-learning, sarsa, double-q, actor-critic, ppo, decision-transformer, monte-carlo, td-lambda, dqn
- Task-specific configuration:
  - agent-routing: double-q (reduces bias)
  - error-avoidance: sarsa (conservative)
  - confidence-scoring: actor-critic (continuous)
  - trajectory-learning: decision-transformer (sequence)
  - context-ranking: ppo (stable)
  - memory-recall: td-lambda (traces)

## TensorCompress (10x Memory Savings)
- 5 compression levels based on access frequency:
  - none (hot >0.8): 0% savings
  - half (warm >0.4): 50% savings
  - pq8 (cool >0.1): 87.5% savings
  - pq4 (cold >0.01): 93.75% savings
  - binary (archive ≤0.01): 96.9% savings
- Auto-compression on session-end

## Pretrain Phases (11 total)
- Phase 10: Multi-algorithm learning bootstrap
- Phase 11: TensorCompress initialization

## Environment Variables (v2.1)
- RUVECTOR_MULTI_ALGORITHM=true
- RUVECTOR_DEFAULT_ALGORITHM=double-q
- RUVECTOR_TENSOR_COMPRESS=true
- RUVECTOR_AUTO_COMPRESS=true

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add null checks in getQ() method for patterns object
- Add null checks in updateQ() and learn() methods
- Improve load() to merge file data with defaults
- Ensures all required fields exist even with partial data
- Fixes 'Cannot read properties of undefined' errors

Published as v0.1.70

New CLI Commands:
- batch-learn: Process multiple learning experiences at once
- subscribe: Stream real-time learning updates (polling)
- watch: Auto-learn from file changes in real-time

New MCP Tools (49 total):
- hooks_batch_learn: Batch process experiences array
- hooks_subscribe_snapshot: Get state deltas for subscriptions
- hooks_watch_status: Get current watch/learning status

Features:
- Batch learning processes multiple experiences efficiently
- Subscribe streams events: learn, route, memory, compress
- Watch monitors file changes and auto-learns patterns
- Co-edit detection in watch mode (files edited within 1 min)

Published as v0.1.71

Implements the full delta-behavior framework - systems where change is
permitted but collapse is not.

## Core Implementation
- Coherence type with [0,1] bounds and safe constructors
- Three-layer enforcement: energy cost, scheduling, memory gating
- DeltaSystem trait for coherence-preserving systems
- DeltaConfig with strict/relaxed/default presets

## 11 Exotic Applications
1. Self-Limiting Reasoning - AI that does less when uncertain
2. Computational Event Horizon - bounded computation without hard limits
3. Artificial Homeostasis - synthetic life with coherence-based survival
4. Self-Stabilizing World Model - models that refuse to hallucinate
5. Coherence-Bounded Creativity - novelty without chaos
6. Anti-Cascade Financial System - markets that cannot collapse
7. Graceful Aging - systems that simplify over time
8. Swarm Intelligence - collective behavior without pathology
9. Graceful Shutdown - systems that seek safe termination
10. Pre-AGI Containment - bounded intelligence growth
11. Extropic Substrate - goal mutation, agent lifecycles, spike semantics

## Performance Optimizations
- O(n²) → O(n·k) swarm neighbor detection via SpatialGrid
- O(n) → O(1) coherence calculation with incremental cache
- VecDeque for O(1) history removal
- SIMD utilities with 8x loop unrolling
- Bounded history to prevent memory leaks

## Security Fixes
- Replaced unsafe static mut with AtomicU64 for thread-safe RNG
- NaN validation on all coherence inputs
- Overflow protection in calculations

## WASM + TypeScript SDK
- Full wasm-bindgen exports for all 11 applications
- High-level TypeScript SDK with ergonomic APIs
- Browser and Node.js examples

## Test Coverage
- 32 lib tests, 14 WASM tests, 13 doc tests (59 total)

Resolves #140

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Adds modular crates implementing delta-behavior for different domains:

## ruvector-delta-core
- Delta trait with compute/apply/compose/inverse operations
- VectorDelta with sparse/dense encoding
- DeltaStream for event sourcing with checkpoints
- DeltaWindow for time-bounded aggregation
- Compression codecs: LZ4, Zstd, DeltaOfDelta, Quantized

## ruvector-delta-wasm
- JsDelta JavaScript-friendly wrapper
- DeltaEngine for capture/apply operations
- SIMD-accelerated operations
- SharedBuffer and BufferPool for memory management

## ruvector-delta-index
- DeltaHnsw - delta-aware HNSW index
- IncrementalUpdater with multiple strategies
- GraphRepairer with lazy/eager/batched/adaptive modes
- QualityMonitor for recall estimation

## ruvector-delta-graph
- GraphDelta for node/edge operations
- NodeDelta with property and label changes
- EdgeDelta with weight and type changes
- DeltaAwareTraversal (BFS, DFS, shortest path)

## ruvector-delta-consensus
- VectorClock and HybridLogicalClock
- DeltaConsensus coordinator
- DeltaGossip for delta dissemination
- CRDTs: GCounter, PNCounter, LWWRegister, ORSet

Test coverage: 77+ tests across all crates

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add ChatEnhancer for enhanced chat processing with skills, memory,
  and proactive assistance integration
- Add SkillExecutor for skill lifecycle management and execution
- Add builtin skills: CodeSkill, MemorySkill, SummarizeSkill, WebSearchSkill
- Improve server.ts with better error handling and session management
- Update AIDefenceGuard with enhanced security checks
- Update chat UI with improved styling and interactions
- Bump version to 0.1.1 with delta crates integration

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

feat(ruvbot): v0.1.1 with skill system, chat UI, and delta crates

  Built from commit 048b8e5

  Platforms updated:
  - linux-x64-gnu
  - linux-arm64-gnu
  - darwin-x64
  - darwin-arm64
  - win32-x64-msvc

  🤖 Generated by GitHub Actions

This commit fixes a critical P0 bug where HNSW indexes on ruvector
columns would crash PostgreSQL with a segmentation fault when using
parameterized queries (prepared statements, ORMs, application drivers).

Root Cause:
- Query vector extraction failed for parameterized queries
- Code fell back to zero vector without validation
- Zero vector caused segfault during HNSW search

Changes:
- Add multi-method query vector extraction pipeline
  1. Direct RuVector::from_polymorphic_datum()
  2. Text parameter conversion for parameterized queries
  3. Validated varlena fallback with dimension checking
- Add query_valid flag to track extraction success
- Add validation before search execution:
  - Reject empty/invalid query vectors with clear errors
  - Reject all-zero vectors (invalid for similarity search)
  - Validate dimension match between query and index
- Apply same fixes to IVFFlat for consistency

Testing:
- Added regression tests for parameterized queries
- Added tests for zero vector error handling
- Added tests for dimension mismatch errors
- Added 384-dimension production-scale tests

Fixes: #141
See: docs/adr/ADR-0027-hnsw-parameterized-query-fix.md

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Move datum and false arguments to same line in from_polymorphic_datum
- Join split let text_len = ... assignment to single line

These changes fix CI rustfmt check failures.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Run rustfmt on all Rust files to fix CI formatting checks.
This addresses pre-existing formatting inconsistencies across:
- cognitum-gate-kernel
- cognitum-gate-tilezero
- prime-radiant
- ruvector-* crates
- examples/benchmarks
- and other crates

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…ault-141

fix(hnsw): resolve segfault with parameterized queries (#141)

  Built from commit e76f01a

  Platforms updated:
  - linux-x64-gnu
  - linux-arm64-gnu
  - darwin-x64
  - darwin-arm64
  - win32-x64-msvc

  🤖 Generated by GitHub Actions

Resolves the "already exists and is not an empty directory" error by:
- Adding a cleanup step to remove the directory before git clone
- Setting up Node.js for ruvector dependencies
- Installing and verifying ruvector MCP installation

  Built from commit 1eb32f7

  Platforms updated:
  - linux-x64-gnu
  - linux-arm64-gnu
  - darwin-x64
  - darwin-arm64
  - win32-x64-msvc

  🤖 Generated by GitHub Actions

Implements the 6-stage CRV (Coordinate Remote Viewing) signal line
methodology as a ruvector subsystem, mapping each stage to the
appropriate vector database component:

- Stage I: Ideogram gestalts → Poincaré ball hyperbolic embeddings
  (ruvector-attention), encoding hierarchical gestalt taxonomy
- Stage II: Sensory impressions → Multi-head attention vectors
  (ruvector-attention), one head per sensory modality
- Stage III: Dimensional sketches → GNN graph topology
  (ruvector-gnn), spatial elements as nodes, relationships as edges
- Stage IV: Emotional/AOL data → SNN temporal encoding
  (ruvector-mincut SNN), spike rate analysis for AOL detection
- Stage V: Signal line interrogation → Differentiable search
  (ruvector-gnn), soft attention over accumulated session data
- Stage VI: Composite modeling → MinCut partitioning
  (ruvector-mincut), cluster boundary detection for target aspects

Includes session manager with DAG structure, cross-session convergence
analysis for multi-viewer target matching, and 40 passing unit tests.

https://claude.ai/code/session_01CESp4koS81HfLK1HEyCmKJ

…bl4V

  Built from commit 5147a27

  Platforms updated:
  - linux-x64-gnu
  - linux-arm64-gnu
  - darwin-x64
  - darwin-arm64
  - win32-x64-msvc

  🤖 Generated by GitHub Actions

Introduces a complete temporal tensor compression system with:

- ADR-017: SOTA research-backed architecture decision record covering
  groupwise symmetric quantization, temporal segment reuse, access-pattern
  driven tier selection (8/7/5/3 bit), and WASM-compatible design

- ruvector-temporal-tensor crate (zero external dependencies):
  - tier_policy: Score-based hot/warm/cold bit-width selection
  - f16: Software IEEE 754 half-precision conversion
  - bitpack: Arbitrary bit-width stream packing (no alignment waste)
  - quantizer: Groupwise symmetric quantization with f16 scales
  - segment: Binary segment format (TQTC) encode/decode
  - compressor: Temporal segment manager with drift detection
  - ffi: WASM/C FFI with handle-based resource management

- ruvector-temporal-tensor-wasm crate for wasm32 targets

- 33 passing unit tests covering all modules

Compression targets: 4x (hot/8-bit), 4.57x (warm/7-bit),
6.4x (warm/5-bit), 10.67x (cold/3-bit) vs f32 baseline.

https://claude.ai/code/session_01U63xtGd5Q8mUevyY7nUSfJ

Key optimizations:
- Eliminate per-value modulo/division in dequantize by restructuring
  to iterate by frame→group→element (was: flat index with val_idx % tensor_len)
- Cache f32 scales in TemporalTensorCompressor to avoid repeated f16→f32
  conversion on every push_frame (drift check + quantization)
- Add optimized _f32 API variants (frame_fits_scales_f32, quantize_and_pack_f32,
  dequantize_f32) that accept pre-converted scales
- Pre-reserve Vec capacity in quantize_and_pack (avoids reallocations)
- Add #[inline] on qmax_from_bits, f32_to_f16_bits, f16_bits_to_f32
- Use chunks() iterator instead of manual index tracking in compute_scales

New tests (41 total, up from 33):
- Roundtrip tests for 5-bit and 7-bit quantization
- Non-finite value handling (NaN, Inf, -Inf)
- Single-element group edge case
- Compression ratio validation for all tiers
- Cold-tier (3-bit) full roundtrip
- Large tensor multi-group (512-dim, 50 frames)
- Accessor method coverage

https://claude.ai/code/session_01U63xtGd5Q8mUevyY7nUSfJ

…ties

- Fix all clippy warnings: module-level //! docs, .div_ceil(), is_empty()
- Optimize segment::decode to call dequantize_f32 directly (skip legacy wrapper)
- Add decode_single_frame() for random-access frame decoding
- Add compression_ratio() utility for segment inspection
- Add comprehensive doc-examples with 3 tested examples in lib.rs
- Fix HEADER_SIZE offset bug in decode_single_frame (22 vs 26)
- All 41 unit tests + 3 doc-tests pass, 0 clippy warnings

https://claude.ai/code/session_01U63xtGd5Q8mUevyY7nUSfJ

Plain-language introduction explaining what temporal tensor compression
does and why it matters, feature tables, Quick Start with 4 code
examples (basic, streaming, random-access, custom policy), full API
reference, segment binary format spec, FFI/WASM guide, and build
instructions.

https://claude.ai/code/session_01U63xtGd5Q8mUevyY7nUSfJ

…n-1r58N

  Built from commit ddb0408

  Platforms updated:
  - linux-x64-gnu
  - linux-arm64-gnu
  - darwin-x64
  - darwin-arm64
  - win32-x64-msvc

  🤖 Generated by GitHub Actions

Implements ADR-018: Visual World Model as a Bounded Nervous System.

Core crate (ruvector-vwm):
- 4D Gaussian primitives with temporal deformation and screen projection
- Spacetime tile system with quantization tiers (Hot8/Warm7/Warm5/Cold3)
- Packed draw list protocol for deterministic GPU rendering
- Coherence gate for update acceptance/rejection with rollback support
- Append-only lineage log with full provenance tracking
- Entity graph for objects, tracks, regions with typed edges
- Streaming protocol with keyframe/delta/semantic packets and bandwidth budget

WASM bindings (ruvector-vwm-wasm):
- Browser-ready wasm-bindgen wrappers for all core types
- WasmGaussian4D, WasmDrawList, WasmCoherenceGate, WasmEntityGraph
- WasmLineageLog, WasmActiveMask, WasmBandwidthBudget

WebGPU viewer (examples/vwm-viewer):
- WGSL shaders for Gaussian splatting with alpha blending
- CPU-side projection, depth sorting, and active mask filtering
- Orbit camera controls
- Synthetic demo data generator
- Time scrubber UI with FPS counter and entity search

Zero external dependencies in core crate for full WASM compatibility.
Both crates compile cleanly against the workspace.

https://claude.ai/code/session_012MQauGiqSnQbszfmFKpsNT

…r VWM

Documentation:
- README for ruvector-vwm (712 lines) with collapsible groups covering
  all core concepts, 13 use cases across product/research/frontier tiers,
  architecture diagrams, and quick start examples
- README for ruvector-vwm-wasm with full API reference, JS examples,
  and type mapping tables
- README for vwm-viewer with quick start, controls, and WebGPU pipeline docs

Architecture Decision Records:
- ADR-019: Three-Cadence Loop Architecture (fast/medium/slow rate separation)
- ADR-020: GNN-to-Coherence-Gate Feedback Pipeline (identity verdicts,
  mincut signal, confidence calibration)
- ADR-021: Four-Level Attention Architecture (view/temporal/semantic/write)
- ADR-022: Query-First Rendering Pattern (retrieve → select → render)

Integration Tests:
- 28 end-to-end tests covering full pipeline, dynamic scenes, coherence
  gate scenarios, entity graph warehouse scene, lineage audit trail,
  streaming protocol, multi-tile scenes, privacy tags, roundtrip fidelity,
  and edge cases

All 78 tests pass (49 unit + 28 integration + 1 doc-test).

https://claude.ai/code/session_012MQauGiqSnQbszfmFKpsNT

…ks, and embedding search

- Add four-level attention pipeline (view/temporal/semantic/write) per ADR-021
- Add query-first rendering engine with SceneQuery/QueryResult per ADR-022
- Add three-cadence loop scheduler (fast 60Hz, medium 5Hz, slow 0.5Hz) per ADR-019
- Add static/dynamic layer separation with automatic Gaussian classification
- Add cosine-similarity embedding search (search_by_embedding, top_k_by_embedding) to EntityGraph
- Add Criterion benchmark suite (20 benchmarks across 8 groups: gaussian, tile, draw_list, coherence, entity, mask, streaming, sort)
- Add performance acceptance tests
- Implement WASM integration path in viewer (coherence gate, entity graph, active mask, draw list)
- 177 tests passing, clippy clean, zero dependencies in core crate

https://claude.ai/code/session_012MQauGiqSnQbszfmFKpsNT

Integration tests now use tolerance-based comparison for float fields
since PrimitiveBlock::encode uses real 8-bit quantization (lossy).
IDs remain exact. All 28 integration tests pass.

https://claude.ai/code/session_012MQauGiqSnQbszfmFKpsNT