@recursiv/sdk 0.7.152 → 0.7.195

package/dist/resources/evidence.d.ts

@@ -27,12 +27,14 @@ export interface CreateEvalInput {
 }
 export interface SimulateInput {
     target: EvidenceTarget;
-    /** The batch of cases to run through the target. */
-    cases: Array<{
+    /** The batch of cases to run through the target. Pass this or `eval_set_id`. */
+    cases?: Array<{
         input: string;
         expected?: Record<string, unknown>;
         label?: string;
     }>;
+    /** Run a saved set from {@link EvidenceResource.createEvalSet} instead of inline cases. */
+    eval_set_id?: string;
     /** Optionally attach evals for this run (also saved to the target). */
     evals?: Array<Omit<CreateEvalInput, 'target'>>;
     run_type?: 'test' | 'real';
@@ -41,6 +43,40 @@ export interface SimulateInput {
     /** Attach the run to a project so it appears there, not only on the agent. */
     project_id?: string;
 }
+/** A reusable set of cases with known answers, run with `simulate({ eval_set_id })`. */
+export interface CreateEvalSetInput {
+    name: string;
+    /** 1 to 50 cases. `weight` 1-100 (default 1); `trap` marks a case built to catch a wrong answer. */
+    cases: Array<{
+        input: string;
+        expected?: Record<string, unknown>;
+        weight?: number;
+        trap?: boolean;
+        label?: string;
+    }>;
+    description?: string;
+    scoring_mode?: 'exact' | 'judge';
+    /** For judge scoring: what a good answer looks like. */
+    rubric?: string;
+    /** Defaults to the key's organization; omit both for a personal set. */
+    organization_id?: string;
+    project_id?: string;
+}
+export interface EvalSetRow {
+    id: string;
+    name: string;
+    description: string | null;
+    scoringMode: string;
+    organizationId: string | null;
+    projectId: string | null;
+    createdAt: string;
+}
+/** A run's result matrix as a file body. */
+export interface EvalRunExport {
+    filename: string;
+    mimeType: string;
+    content: string;
+}
 export interface EvalRunRow {
     id: string;
     targetType: string;
@@ -167,6 +203,25 @@ export declare class EvidenceResource {
             reason: string;
         }>;
     }>>;
+    /**
+     * Save a reusable set of cases with known answers. Run it as often as you
+     * like with `simulate({ target, eval_set_id })`, so runs compare like with like.
+     */
+    createEvalSet(input: CreateEvalSetInput): Promise<SingleResponse<{
+        eval_set_id: string;
+        case_count: number;
+    }>>;
+    /**
+     * The saved sets you can run: an organization's (the key's own by default),
+     * or, for a personal key with no organization, the ones you made.
+     */
+    listEvalSets(opts?: {
+        organizationId?: string;
+    }): Promise<{
+        data: EvalSetRow[];
+    }>;
+    /** A run's result matrix as CSV (default) or JSON, for offline analysis. */
+    exportRun(runId: string, format?: 'csv' | 'json'): Promise<SingleResponse<EvalRunExport>>;
     /** Recent runs for a target. */
     listRuns(target: EvidenceTarget, limit?: number): Promise<{
         data: EvalRunRow[];

package/dist/resources/evidence.js

@@ -26,6 +26,25 @@ export class EvidenceResource {
     runTaskChecks(taskId) {
         return this.client.post('/evidence/task-checks/run', { task_id: taskId });
     }
+    /**
+     * Save a reusable set of cases with known answers. Run it as often as you
+     * like with `simulate({ target, eval_set_id })`, so runs compare like with like.
+     */
+    createEvalSet(input) {
+        return this.client.post('/evidence/eval-sets', input);
+    }
+    /**
+     * The saved sets you can run: an organization's (the key's own by default),
+     * or, for a personal key with no organization, the ones you made.
+     */
+    listEvalSets(opts = {}) {
+        const query = opts.organizationId ? `?organization_id=${encodeURIComponent(opts.organizationId)}` : '';
+        return this.client.get(`/evidence/eval-sets${query}`);
+    }
+    /** A run's result matrix as CSV (default) or JSON, for offline analysis. */
+    exportRun(runId, format = 'csv') {
+        return this.client.get(`/evidence/runs/${runId}/export?format=${format}`);
+    }
     /** Recent runs for a target. */
     listRuns(target, limit = 20) {
         return this.client.get(`/evidence/runs?target_type=${target.type}&target_id=${encodeURIComponent(target.id)}&limit=${limit}`);

package/dist/resources/feeds.d.ts

@@ -160,6 +160,57 @@ export interface FeedTopResponse {
     lead: FeedStory | null;
     meta: FeedTopMeta;
 }
+/** The networks a firehose item comes from: Minds, and the open networks the top's lanes carry. */
+export type FeedFirehoseNetwork = 'minds' | 'bluesky' | 'activitypub' | 'nostr';
+/**
+ * One firehose item (0.7.195): an item of the top of a feed, with its network, its state, when it was published and when
+ * the server chose it. An open-network post carries its `card` to draw (no URL on its network: open it by `target.path`).
+ */
+export interface FeedFirehoseItem extends FeedTopItem {
+    network: FeedFirehoseNetwork;
+    /** 'breaking' (its radar story broke in the last 6 hours) or 'viral'; at most one, else null. */
+    state: 'breaking' | 'viral' | null;
+    /** When the post was published (ISO). */
+    publishedAt: string;
+    /** When the server chose it for this answer (ISO, the server's clock). */
+    chosenAt: string;
+    /** An open-network post's card. */
+    card?: FeedTopNetworkCard;
+}
+export interface FeedFirehoseMeta {
+    /** The soonest to read again, in ms: longer when the answer was short or the server is busy. */
+    refillAfterMs: number;
+    /** The oldest shared snapshot the stream's pool was composed from (ISO). */
+    poolBuiltAt: string;
+    poolSize: number;
+    /** 'safe' for a member on safe defaults, 'open' for one who turned them off. */
+    safety: 'safe' | 'open';
+    /** Fewer items than asked for: supply is thin, so read again after `refillAfterMs`. */
+    short: boolean;
+    networks: Record<'bluesky' | 'activitypub' | 'nostr', 'ok' | 'unavailable'>;
+    /**
+     * Whether the server kept this session from repeating: 'on'; 'full' when the session's record is at its cap; and
+     * 'unavailable' when it could not (guard against repeats in the app, over more of what it showed).
+     */
+    sent: 'on' | 'full' | 'unavailable';
+    compass_filter?: CompassFilterMeta;
+}
+/** GET /feeds/firehose. `retract` names refs the session was sent that were taken down since: remove them wherever they are held. */
+export interface FeedFirehoseResponse {
+    items: FeedFirehoseItem[];
+    retract: string[];
+    meta: FeedFirehoseMeta;
+}
+export interface FeedFirehoseParams {
+    /** Opaque per app session ([A-Za-z0-9_-], up to 64), required: nothing is sent twice within one session. */
+    session: string;
+    /** 1..25, default 20. */
+    limit?: number;
+    /** A Politics choice, as `posts.list` takes it. */
+    compass?: string;
+    /** Only one network's items: `all` (default), `minds`, `bluesky`, `activitypub` or `nostr`. */
+    network?: 'all' | FeedFirehoseNetwork;
+}
 export interface FeedStoriesResponse {
     data: FeedStory[];
     meta: {
@@ -230,6 +281,18 @@ export declare class FeedsResource {
     top(params: FeedTopParams, options?: {
         retry?: boolean;
     }): Promise<FeedTopResponse>;
+    /**
+     * The firehose for the signed-in member, with their own key of the app (0.7.195): the next items of their live stream of
+     * the most engaged, viral and breaking posts of Minds and the open networks, every one re-checked for them on the
+     * server, none sent twice in one `session`. Reveal them one at a time; read again no sooner than `meta.refillAfterMs`.
+     * Off (404) until the app's operator turns it on; 429 `rate_limited` and 503 `firehose_unavailable` carry Retry-After.
+     * A Mix or Choose sides `compass` spends one of the member's hourly sided Politics reads at most once per `session` and
+     * choice in ten minutes; 429 `compass_rate_limited` when they are spent (choose Everything or Hide politics, or wait).
+     * Always private: never cached for anyone else.
+     */
+    firehose(params: FeedFirehoseParams, options?: {
+        retry?: boolean;
+    }): Promise<FeedFirehoseResponse>;
     /** The stories the member should meet next (the story tail), by the same ranker as the top's lead. */
     stories(params?: FeedStoriesParams, options?: {
         retry?: boolean;

package/dist/resources/feeds.js

@@ -14,6 +14,18 @@ export class FeedsResource {
     top(params, options) {
         return this.http.get('/feeds/top', { ref: params.ref, limit: params.limit, session: params.session, compass: params.compass }, options);
     }
+    /**
+     * The firehose for the signed-in member, with their own key of the app (0.7.195): the next items of their live stream of
+     * the most engaged, viral and breaking posts of Minds and the open networks, every one re-checked for them on the
+     * server, none sent twice in one `session`. Reveal them one at a time; read again no sooner than `meta.refillAfterMs`.
+     * Off (404) until the app's operator turns it on; 429 `rate_limited` and 503 `firehose_unavailable` carry Retry-After.
+     * A Mix or Choose sides `compass` spends one of the member's hourly sided Politics reads at most once per `session` and
+     * choice in ten minutes; 429 `compass_rate_limited` when they are spent (choose Everything or Hide politics, or wait).
+     * Always private: never cached for anyone else.
+     */
+    firehose(params, options) {
+        return this.http.get('/feeds/firehose', { session: params.session, limit: params.limit, compass: params.compass, network: params.network }, options);
+    }
     /** The stories the member should meet next (the story tail), by the same ranker as the top's lead. */
     stories(params = {}, options) {
         return this.http.get('/feeds/stories', { after: params.after, limit: params.limit, compass: params.compass }, options);