Deterministic Compiler System (DCS)

A deterministic compiler that transforms structured requests into verified artifacts.

What is DCS

DCS is a compiler system where:

• Inputs are deterministic: Requests (REQ) specify intent, language, and parameters
• Outputs are reproducible: The same REQ always produces the same artifact
• Verification is built-in: Every artifact passes machine-checkable verification
• External generators are untrusted: Only validated REQ specifications enter the deterministic pipeline

Pipeline

REQ  →  IR  →  Artifact

A request (REQ) specifies intent, language, and parameters. The compiler transforms this into an internal representation (IR) bound to a snapshot-pinned manifest, then produces a verified artifact with a proof bundle.

Usage

Build an artifact from a request specification:

dcs build --req request.json

Example request specifications are available in the examples/ directory.

On success, DCS produces:

• dist/artifact.zip — the generated artifact
• dist/proof_bundle.zip — verification evidence

Verification

DCS artifacts are reproducible. Given the same inputs and snapshot, any machine running the proof produces identical hashes.

To verify:

./scripts/proof/run_clean_proof_v1.sh

Expected hashes are in dcs/expected_proof_hashes_v1.json.

For independent verification instructions, see docs/EXTERNAL_VERIFICATION.md.

Repository layout

This repository is an implementation tree, not a single Python package at the root. The top-level directories (dcs_cli/, dcs_core/, nlc/, orchestrator/, proof/, scripts/, etc.) are the canonical source and spec surfaces for v1. Root files (e.g. pyproject.toml, Dockerfile.tier3, LICENSE) are configuration and entrypoints; the full directory map is in docs/paths.md. There is no demo/ directory in the curated tree—examples live under examples/.

Documentation

• Whitepaper — architecture and design principles
• External Verification — independent proof reproduction
• Reproducibility and schemas — release tag → freeze manifest → proof hashes; schema authority chain (REQ → IR → artifact contract → verifier)
• CLI Reference — command interface
• Paths — directory layout and conventions

Implementation Status

This is the v1 implementation. The system supports:

• Structured intake via REQ.json
• Snapshot-pinned knowledge manifests
• Deterministic artifact generation
• Machine-checkable verification
• Reproducible proof bundles

See governance/ for freeze manifests and version attestations.

License

See LICENSE.