Module policy 

Policy - status, rank, and primary-agent eligibility for known models.

This module owns the bundled policy record (feedback 01/02/05/06): the status, rank, default flag, and primary-agent eligibility of every model the project has an explicit opinion about. Unknown models default to Available; the policy never claims a model is ineligible unless a policy entry says so. The JSON contract serializes as snake_case so the CLI and the Hermes plugin can consume it without transformation.

Structs

ModelPolicy

Versioned policy document - the JSON contract envelope for policy get.

PolicyEntry

One policy opinion about one model.

RankingBreakdown

Evidence breakdown for one model’s ranking_score - the picker renders one line per component from this struct, so every ranking decision is explainable (feedback 02 evidence lines).

Enums

ModelStatus

Picker status for a model (feedback 02 status vocabulary).

Constants

POLICY_VERSION

Version of the bundled default policy document (feedback 02 YAML version: 1).

REFERENCE_CONTEXT_TOKENS

Context window (tokens) that normalizes to 1.0 - DeepSeek V4 Flash’s advertised window. Anything at or beyond this caps at 1.0.

REFERENCE_LATENCY_MS

Median latency (ms) at or beyond which the latency score is 0.0 (60s).

REFERENCE_PRICE_PER_MILLION

Average per-million-token price (USD) at or beyond which the price score is 0.0.

WEIGHT_CONTEXT

Weight of the context-window suitability component.

WEIGHT_DELIVERY

Weight of the recent-delivery component in the provider ranking score (feedback 02: score = 0.40R + 0.25T + 0.15C + 0.10L + 0.10P).

WEIGHT_LATENCY

Weight of the latency component.

WEIGHT_PRICE

Weight of the price/value component.

WEIGHT_TOOL_LOOP

Weight of the multi-turn tool-loop conformance component.

Functions

ranking_score

Provider ranking score (feedback 02: score = 0.40R + 0.25T + 0.15C + 0.10L + 0.10P).