Standalone, vendor-neutral — tied to no model, runtime, or vendor (the property that makes a universal standard worth anything). arl-rs/ is its own workspace; arl-core has zero dependency on any other standard. Models a complete ARL claim and enforces, in code, what the spec says on paper: - The four axes as types: ValidationDepth (1–9), ConvergenceClass (A–E, A most deterministic, E the default), SecurityClass (S0–S4), and EnergyProfile (three joule figures, or an explicit Undisclosed). - Cross-axis gates in Claim::validate(): ARL ≥ 4 ⇒ Class D + S1; ≥ 6 ⇒ C + S2; ≥ 8 ⇒ B + S3; 9 ⇒ S4. Undisclosed energy caps at ARL 3. Security above S0 requires disclosed methodology; S3+ requires independent reproducibility. ARL ≥ 4 requires published error bars + failure modes; ARL ≥ 6 requires methodology published before the claim. validate() returns every violation, not just the first. - Controlled vocabulary (LEXICON): excluded unmeasurable terms (AGI, superintelligence, consciousness, sentience, singularity, human-level) invalidate a claim; partially-hype terms (alignment, hallucination, reasoning, safety, world model, …) are flagged as warnings for operational-sense review. Whole-token matching avoids false positives. Defaults are the uncharacterized floor (ARL 1 / Class E / S0 / energy undisclosed) so a bare claim is honest about how little it has shown. 22 tests; Claim is serde round-trippable. arl-sandbox (ARL-S Supervisor) is next. Co-Authored-By: Claude Opus 4.7 <[email protected]>
14 lines
541 B
TOML
14 lines
541 B
TOML
[package]
|
|
name = "arl-core"
|
|
version.workspace = true
|
|
edition.workspace = true
|
|
rust-version.workspace = true
|
|
license.workspace = true
|
|
repository.workspace = true
|
|
authors.workspace = true
|
|
description = "ARL core — the four-axis AI Readiness Level claim model with the cross-axis gates and controlled-vocabulary enforcement, so an invalid or unmeasurable claim cannot be asserted. Standalone, vendor-neutral, tied to no runtime."
|
|
|
|
[dependencies]
|
|
serde = { workspace = true }
|
|
serde_json = { workspace = true }
|
|
thiserror = { workspace = true }
|