vsm-pattern-engine

Elixir library for temporal pattern detection, correlation analysis, and anomaly detection. Designed for use within the Viable System Model (VSM) ecosystem but usable independently.

Status

Version 0.1.0. Not published to Hex. The main module (VsmPatternEngine) still contains the default hello/0 stub, but the subsystem modules under lib/vsm_pattern_engine/ are implemented.

What it does

• Temporal detection -- identifies periodic, trend, burst, decay, and cyclic patterns in time-series data using sliding windows.
• Correlation analysis -- computes Pearson, Spearman, Kendall, and mutual information correlations between pattern sets. Includes a basic causal analysis pass.
• Anomaly detection -- statistical (z-score/IQR), isolation forest, and a VSM-specific method that factors in variety ratios and recursion levels.
• Vector store integration -- optional client for persisting and querying patterns against vsm-vector-store.

Repository structure

Path	Contents
lib/vsm_pattern_engine/engine.ex	Core dispatch: analyze_pattern, detect_anomaly, correlate_patterns
lib/vsm_pattern_engine/temporal/	Detector, analyzer, window, stream processor, pattern struct, cache (7 files)
lib/vsm_pattern_engine/correlation/	Analyzer, matrix, matrix processor, statistics, relationship, cache (7 files)
lib/vsm_pattern_engine/anomaly/	Detector, alert manager, baseline manager (4 files)
lib/vsm_pattern_engine/vector_store/	Client, connection, encoder (3 files)
lib/vsm_pattern_engine/telemetry.ex	Telemetry event definitions
test/	1 test file (basic)

Module count

24 .ex files under lib/. 1 test file.

Dependencies

Category	Packages
HTTP	tesla, mint, finch
Math/stats	statistics
Time	timex
Concurrency	flow, gen_stage
Telemetry	telemetry, telemetry_metrics, telemetry_poller
Serialization	jason
Dev	ex_doc, credo, dialyxir

Requires Elixir >= 1.14.

Quick start

git clone https://github.com/viable-systems/vsm-pattern-engine.git
cd vsm-pattern-engine
mix deps.get
mix test

{:ok, _} = VsmPatternEngine.Application.start(:normal, [])

data = [1.0, 2.1, 3.0, 3.9, 5.1, 6.0, 7.1, 8.0]
{:ok, result} = VsmPatternEngine.Engine.analyze_pattern(data)

baseline = [1.0, 1.1, 0.9, 1.0, 1.2, 0.8, 1.1]
new_data = [1.0, 1.1, 5.0, 1.0]
{:ok, anomaly} = VsmPatternEngine.Engine.detect_anomaly(new_data, baseline)

Configuration

config :vsm_pattern_engine,
  vector_store_url: "http://localhost:4000/api",
  detection_interval: 5000,
  anomaly_threshold: 0.8,
  correlation_threshold: 0.7

config :vsm_pattern_engine, :vsm,
  recursion_levels: 5,
  variety_management: :requisite,
  algedonic_signals: true

Known limitations

• The top-level VsmPatternEngine module is a stub (hello/0). Use VsmPatternEngine.Engine for actual functionality.
• Only 1 test file; coverage is minimal.
• Vector store client assumes vsm-vector-store is running locally; no fallback if unavailable.
• No CI badge despite having a .github/workflows/ci.yml.
• The statistics hex package handles basic stats; more advanced methods (isolation forest, LOF) are implemented from scratch with limited validation.

Related packages

• vsm-vector-store -- vector persistence backend
• vsm-event-bus -- event coordination
• vsm-mcp -- top-level VSM orchestrator
• vsm-external-interfaces -- HTTP/WS/gRPC adapters

License

MIT