Module ontology 

A complete, self-describing ontology of agentic-eval itself.

Agentic-first design means a consumer should never have to read prose docs to use the library — it should be able to discover every capability through a compact, machine-readable manifest and expand any entry on demand. That is the same progressive-disclosure pattern this crate measures (a cheap root index + describe(...) for detail), applied reflexively to the crate’s own surface: its four axes, the Effect taxonomy with the policy [Decision] each effect gets under each Mode, the supported tokenizer Models, and the built-in CLI command classifier.

The ontology is deterministic (built from static data in fixed order, no map iteration), compact (manifest is a few hundred tokens), and machine-readable (every type derives serde::Serialize under the serde feature). Start at manifest, expand with describe, or take the whole structured catalog with ontology.

// Discover the surface without reading docs:
let root = agentic_eval::ontology::manifest();
assert!(root.contains("axes:"));
// Expand one entry on demand:
let safety = agentic_eval::ontology::describe("safety").unwrap();
assert!(safety.contains("assess_safety"));
let rm = agentic_eval::ontology::describe("destructive").unwrap();
assert!(rm.contains("agent=approve")); // the policy decision, machine-checkable

Structs

AxisDoc

One of the four evaluation axes — what it measures and how to invoke it.

EffectDoc

An effect class with the policy decision it receives under each mode.

ModelDoc

A supported tokenizer/model.

Ontology

The complete, structured ontology of the crate.

SubjectDoc

A profiled evaluation subject (language, framework, or VM system) — compact index entry; expand with describe for full per-axis scores and evidence.

Constants

VERSION

The crate’s own version (so the ontology never drifts from the build).

Functions

axes

The four evaluation axes, in canonical order.

describe

Expand one ontology entry by name (case-insensitive): an axis (tokens…), an effect (destructive…), a model (gpt4/claude…), or a section keyword (axes / effects / models / modes / commands). Returns None for an unknown query — the manifest lists every valid name.

effects

The effect taxonomy, each annotated with the policy decision it gets under each mode and a few example commands the built-in classifier maps to it.

frameworks

Compact index of the profiled AI frameworks.

languages

Compact index of the profiled programming languages.

manifest

A compact, deterministic manifest — the cheap discovery root. Lists the axes, effect classes, modes, models, and command count, plus how to expand any entry with describe. A few hundred tokens; the agentic-first entry point.

models

The tokenizer models token efficiency can be counted under (exact or estimated).

ontology

The complete structured ontology of the crate.

vms

Compact index of the profiled VM/sandbox systems.

web_stacks

Compact index of the profiled web stacks / wire protocols.