# DARKOZA Arrival Graph — research preview

Version: 0.1.0-alpha.1
Status: experimental local developer preview. Not connected to Darkoza production chat or a hosted execution API.

## Requirements

- Python 3.10+
- Python standard library only

## Try it

From this directory:

    python3 -m unittest discover -s tests -v
    bin/dkzg check examples/optimized-graph.dgz --optimize
    bin/dkzg run examples/optimized-graph.dgz --inputs '{"x":123456789}' --optimize

## What is included

- compiler/arrival_graph.py — parser, typed integer dependency graph and DAG evaluator
- compiler/equality_saturation.py — bounded conservative E-Graph / equality-saturation optimizer
- bin/dkzg — command-line launcher
- examples — base and optimized sample programs
- tests — Arrival Graph and E-Graph regression tests
- benchmarks — synthetic warm-run benchmark and measured results
- docs — language model, semantics, safety policy and limits

## Important limitations

The current language supports pure signed 64-bit integer graph operations only. It has no filesystem, network, shell, import, callback or native-code opcodes. It is not formally verified, independently security-audited, self-hosted or production-approved. Hosted code execution for end users is not available.

The equality-saturation optimizer has explicit node/class/round limits and falls back to the original graph on budget exhaustion. It intentionally avoids algebraic rewrites that can suppress checked-overflow errors.

The benchmark is a synthetic five-operation identity chain. Its measured warm-run speedup is not representative of general programs and does not include optimization/setup overhead in each execution. Optimization needs repeated reuse to amortize.

## License and redistribution

No open-source or commercial reuse license has been selected for this research preview. Public download is not a grant to rebrand, republish, sublicense or incorporate the package into unrelated products. Contact Darkoza for any broader licensing request.
