Browse documentation
Guarantees and boundaries
What each platform claim means, where it applies and what it does not establish.
The first-release model combines the supported Rust profile, durable execution, explicit effects, scoped authority and development tools. A guarantee applies to the relevant runtime version, execution route and stated conditions.
Release perspective
This documentation describes the integrated first-public-release target, with the active runtime, authority, development and ecosystem programmes completed and qualified together. Concept pages use that target to explain how the pieces fit.
Runnable tutorials preserve the currently documented effectful command surface and state their prerequisites. Feature and distribution availability depend on the installed runtime version. A first-release design statement is not evidence that an earlier build already implements or qualifies it.
Language and development
| Claim | Meaning | Boundary |
|---|---|---|
| Build with Rust | Code uses Rust within a supported profile | Arbitrary Rust programs and Cargo dependency graphs are not automatically admitted |
| Rust safety foundations | Types, ownership checks and defined semantics reduce specific bug classes | Memory safety does not establish business correctness or eliminate all security failures |
| Zero-build loop | Supported application and library edits use an installed runtime | Checking and lowering remain; runtime and native components have separate builds |
| Exact code identities | Defined artifacts and dependencies have exact content identities | Identity does not prove benign behaviour or publisher authenticity |
Execution and authority
| Claim | Meaning | Boundary |
|---|---|---|
| Deterministic core | Computation has defined semantics under its execution contract | Environmental inputs and external delivery need explicit rules |
| Durable execution | State and progress persist within the storage and recovery contract | No universal exactly-once external delivery or total-disk-loss recovery is implied |
| Actor boundaries | State ownership and message contracts give the system explicit structure | No automatic OS sandbox, independent deployment or freedom from resource exhaustion |
| Scoped capabilities | Grants and authenticated context constrain permitted operations | A declaration is not a grant; native host authority remains part of the trust boundary |
| Inspectable communication | Declared, permitted and observed views answer distinct questions | A diagram is not proof of complete field-level information-flow control |
Evidence
| Evidence | What it can establish | Conditions to retain |
|---|---|---|
| Type or profile check | Acceptance under the checked rules | Exact inputs, route and checker identity |
| Test | Observed behaviour for selected cases | Inputs, expected result, execution origin and environment |
| Replay | Reconstruction under recorded inputs and execution rules | Retained history, code and applicable replay contract |
| Benchmark | Performance for a measured workload | Workload, state and log sizes, hardware, configuration and samples |
| Formal proof | A stated property under formal assumptions | The model, proved property and scope of connection to implementation |
| Live observation | Behaviour in a particular operating context | Time, deployment identity, coverage and collection limits |
These forms of evidence are complementary. One does not silently stand in for another.
Security boundary
The trusted runtime, native hosts and drivers, secret handling, storage and deployment environment belong in the system’s threat model. Narrow capabilities can limit code’s reach within the enforced runtime boundary. They do not remove the need to review trusted components.
Authority for completing admitted work and authority to disclose protected data can differ. Inspection and retained history must respect that distinction.
Scope beyond the first release
Cross-node execution, a managed cloud, recovery after losing all local storage, arbitrary Rust compatibility and a complete graph query-and-rules language are separate milestones. Building a web framework on the foundation is different from shipping a bundled Rails-like framework.
Comparative security or performance claims require a defined threat model or workload and reproducible evidence. The architecture alone does not establish overall superiority over another system.
Build software you can reason about.