agentfootprint/dist/esm/core/agent/honesty/armed.d.ts

The explainable agent framework — backtrack a wrong answer to the exact context that caused it (evidence, not guesses). Built on footprintjs.

/**
 * honesty/armed — which honesty layers a run has armed: the ONE run constant.
 *
 * Pattern: a run constant, seeded once by `stages/seed.ts` (the
 *          `findingsServe` precedent) and never written again during the run.
 * Role:    core/ layer leaf. A reader of the RECORD — the standing fold, a
 *          lens, a bench — tells "this layer was armed and filed nothing" from
 *          "this layer was never armed" by this key alone, never by guessing
 *          from the rows.
 * Emits:   N/A.
 *
 * Absent on a run with no layer armed, so an agent that armed nothing commits
 * exactly the keys it always did.
 */
/** The armed layers, by name. Grows one key per layer that ships. */
export interface HonestyLayers {
    /** The inputs layer (honesty layer 2) — `sf-inputs` is mounted. */
    readonly inputs?: true;
    /**
     * The inputs layer's declared sources (`.findings({ argumentSources: true
     * })` or `.inputsLayer({ argumentSources: true })`) — ruled tools carry
     * `_findings.from`, and the layer checks it. Which door armed it is on the
     * record too: the findings ledger's run constant `findingsServe` exists
     * exactly when `.findings()` did.
     */
    readonly argumentSources?: true;
    /** The results layer (honesty layer 3, step 7b) — `sf-results` is mounted at the loop head. */
    readonly results?: true;
    /**
     * The answer layer (honesty layer 4) — its stages head the final branch,
     * and the Route decider files the answer's witness rows
     * (`assessment/witness.ts`).
     */
    readonly answer?: true;
}
/**
 * The object form of `AgentOptions.inputsLayer` — `AgentBuilder.inputsLayer(options)`.
 * Written inline on the public option (no new export, no new API page); this
 * is the reader's name for it.
 */
export interface InputsLayerOptions {
    /**
     * DECLARED SOURCES without the findings ledger: each ruled tool's served
     * schema carries the reserved `_findings` argument with `from` alone, and
     * the layer checks every claim before the batch runs. Default off.
     */
    readonly argumentSources?: boolean;
}
/**
 * `AgentOptions.inputsLayer` as the agent reads it — `undefined` when the
 * layer is not asked for (`undefined` or `false`), the options otherwise
 * (`true` reads as `{}`). Anything else is REFUSED, naming the value: a door
 * that silently ignored `{ argumentSources: 'yes' }` would look configured and
 * do nothing.
 */
export declare function readInputsLayerOption(value: unknown): InputsLayerOptions | undefined;
/**
 * The run constant for an agent's armed layers, or `undefined` when none is
 * armed (then seed writes nothing). A layer that is not armed is absent from
 * the object — never `false` — so the bytes of a run that armed only the
 * inputs layer are the bytes that layer has always committed.
 * `argumentSources` only beside `inputs`.
 *
 * @example
 * ```ts
 * honestyLayersOf({ inputs: true, answer: false }); // { inputs: true }
 * honestyLayersOf({ inputs: true, argumentSources: true, answer: false }); // { inputs: true, argumentSources: true }
 * honestyLayersOf({ inputs: false, results: true, answer: false }); // { results: true }
 * honestyLayersOf({ inputs: false, answer: false }); // undefined
 * ```
 */
export declare function honestyLayersOf(armed: {
    readonly inputs: boolean;
    readonly argumentSources?: boolean;
    readonly results?: boolean;
    readonly answer: boolean;
}): HonestyLayers | undefined;