What ships in the public repository
- The RET requirement algebra and Xarxa dependency-graph substrate, with validated, satisfiable, and monotone requirement refinements.
- Typed scenarios, conditions, and closed observation domains; the factorized evidence model with explicit evidence-use policies and the Present / ObservedAbsent / Missing / Insufficient resolution semantics.
- Pure, I/O-free semantic evaluation with three-valued (True / False / Unknown) results.
- The accepted-run protocol: authority-minted run identities, exact-head acceptance, receipt-first idempotency, and typed outcome families.
- Local operation with in-memory and SQLite-backed accepted-run state.
- MCP, CLI, generated contracts, and SDK projections for integration.
- Runpack export and independent offline verification tooling.
- Documentation, tests, and formal evidence artifacts, including bounded model-checking models with their bounds stated.
Guarantees, with their bounds
- Deterministic evaluation. Identical validated law, accepted progress, and admitted evidence produce identical semantic results. Determinism covers evaluation — not acquisition availability, scheduling, or external effects.
- Fail-closed decisions. Missing, insufficient, or ambiguous proof holds the gate. Uncertainty never advances a workflow by default.
- Replayable records. Accepted runs export to verifiable bundles; an auditor can re-check integrity and re-evaluate gates from recorded evidence without trusting the runtime that produced them.
- Honest claim boundaries. Record integrity, source authenticity, and the truth of original evidence are separate properties, verified separately. No single green light stands in for all three.
Current status
Capability claims below follow three labels: available now in the public repository, designed target architecture, and explicitly deferred.
| Capability | Status |
|---|---|
| Semantic kernel: RET, Xarxa, typed conditions, evidence resolution, pure evaluation | Available now |
| Local accepted-run operation (in-memory and SQLite) | Available now |
| Caller-submitted evidence; named time, environment snapshot, and rooted local document acquisition | Available now |
| MCP, CLI, generated contracts, SDK projections | Available now |
| Runpack export and independent verification | Available now |
| Customer-operated enterprise integration (identity, governance, durable storage backends) | Designed target |
| Registered network evidence providers (HTTP/REST, remote MCP) | Deferred |
| Subprocess and native execution providers; remote evaluators | Deferred |
| Distributed multi-node placement and failover profiles | Deferred |
A host harness can already acquire its own network or command data and submit it as caller evidence; the record then states that channel honestly rather than dressing it up as something stronger.
Commercial licensing
The commercial surface is rights and engineering, not a different theorem. The same semantic kernel that is public under the SSPL can be licensed for:
- Self-hosted commercial deployment in your environment, on your infrastructure.
- Embedded and OEM use inside agent platforms and products.
- Provider-scale strategic deployment, including integration with customer evidence systems, deployment-profile engineering, qualification, and long-term support.
Arxiu does not operate a managed Decision Gate cloud. Deployments are customer-operated by design: your evidence, your infrastructure, your control.
Licenses
- Decision Gate repository: SSPL 1.0.
ret-logicandxarxacrates: Apache 2.0, licensed separately within the repository.
The SSPL is a source-available license, not an OSI-approved open-source license; evaluate its terms against your intended use. Commercial licenses remove SSPL obligations for the deployment surfaces above.
Contact
For commercial licensing, availability, or deployment questions: license@arxiu.io.