GEMLA SDK

GEMLA is a Γ–EML–α transport architecture research SDK for observable path measures, lifted-phase diagnostics, adversarial-control gates, and finite transport-topology evaluation.

The SDK evaluates whether a supplied trajectory produces a stable lifted transport signature that passes diagnostic gates and rejects adversarial controls.

---

Current Status

Research preview: v0.5.0-practical-demos

Current capabilities:

• package import and editable install
• synthetic transport generation
• lifted phase construction
• RevA v2-SF gate evaluation
• wrong-sign, phase-shuffle, and residue-scramble controls
• anchor irregularity diagnostics
• integer winding diagnostics
• Markdown report export
• benchmark suite
• latent embedding evaluation
• V-JEPA-style embedding adapter
• industrial telemetry demo
• market microstructure demo
• cyber event transport demo
• pytest validation
• GitHub Actions CI

---

Core Pipeline

source trajectory
→ complex / latent surrogate
→ lifted phase Θ(t)
→ RevA v2-SF gate
→ adversarial controls
→ anchor diagnostics
→ winding diagnostics
→ PASS / FAIL verdict

---

Install

From the repository root:

pip install -e ".[dev]"

Check installation:

gemla --help

---

Quickstart

Run the minimal vertical-slice pipeline:

gemla evaluate

Or run the original Python example:

python examples/quickstart/01_minimal_pipeline.py

Expected output includes:

GEMLA Transport Evaluation
--------------------------
RevA v2-SF main pass: True
Wrong-sign rejected: True
Phase-shuffle rejected: True
Residue-scramble rejected: True
Final verdict: PASS

The command writes a Markdown report to:

reports/gemla_transport_report.md

---

CLI Usage

Default transport evaluation

gemla evaluate

Custom output:

gemla evaluate --output reports/custom_report.md

Custom synthetic trajectory:

gemla evaluate --n 1500 --t-max 100 --noise 0.02 --seed 9

Benchmark suite

gemla benchmark

Custom benchmark output directory:

gemla benchmark --output-dir benchmarks/results

Latent embedding evaluation

Run synthetic latent evaluation:

gemla evaluate-latent

Evaluate saved latent embeddings:

gemla evaluate-latent --input path/to/latents.npy

Expected input shape:

(n_steps, latent_dim)

V-JEPA-style embedding evaluation

Run synthetic V-JEPA-style demo:

gemla evaluate-vjepa --synthetic

Evaluate external embeddings:

gemla evaluate-vjepa --input path/to/embeddings.npy

GEMLA does not download or redistribute third-party model weights. The V-JEPA-style adapter treats external embeddings as surrogate transport trajectories.

Practical deployment demos

Industrial telemetry:

gemla demo-industrial

Market microstructure:

gemla demo-market

Cyber event transport:

gemla demo-cyber

Each command generates synthetic domain-proxy data, runs the GEMLA lifted-phase gate stack, rejects adversarial controls, and writes a Markdown report.

---

Python Examples

Minimal transport example:

python examples/quickstart/01_minimal_pipeline.py

Industrial telemetry:

python examples/industrial_telemetry/run_industrial_telemetry_demo.py

Market microstructure:

python examples/market_microstructure/run_market_microstructure_demo.py

Cyber event transport:

python examples/cybersecurity/run_cyber_transport_demo.py

Latent embeddings:

python examples/latent_embeddings/run_latent_transport_demo.py

V-JEPA-style embeddings:

python examples/vjepa_latent_transport/run_vjepa_latent_transport_demo.py

---

Run Tests

pytest

---

Repository Structure

src/gemla/
  data/          synthetic and domain-proxy data generators
  lifted/        lifted phase, anchors, spectral flatness, winding
  gates/         RevA v2-SF gate evaluation
  controls/      adversarial control generation
  pipelines/     end-to-end GEMLA pipelines
  integrations/  latent and V-JEPA-style adapters
  benchmarks/    benchmark runner and result writers
  reports/       Markdown report exporter
  cli/           command-line interface

examples/
  quickstart/
  industrial_telemetry/
  market_microstructure/
  cybersecurity/
  latent_embeddings/
  vjepa_latent_transport/

tests/
  unit and integration tests

---

Documentation

• Architecture Card
• Quickstart
• CLI Reference
• Examples
• Limitations
• Release Checklist
• Changelog

---

Interpretation

The current version is a finite diagnostic SDK. It does not make universal claims. A PASS means the supplied trajectory satisfied the current GEMLA gate stack and rejected the included adversarial controls. A PASS does not mean the system is safe, optimal, causal in a physical sense, or ready for high-stakes deployment.

---

Roadmap

Planned next steps:

• documentation site
• sample output reports
• richer benchmark summaries
• configurable gate thresholds
• real-data adapter examples
• hosted demo or release bundle
• public release preparation

---

License

See LICENSE.

---

Citation

See CITATION.cff.