@bitfab/sdk 0.38.10 → 0.40.0

package/dist/{chunk-UE4GGPM6.js → chunk-5NT4YDCQ.js}

@@ -7,19 +7,21 @@ import {
   runWithAutoTraceRootContext
 } from "./chunk-ZUD7OFYB.js";
 import {
-  BitfabError,
-  DEFAULT_SERVICE_URL,
-  HttpClient,
   NO_MOCK_OVERRIDE,
   deserializeValue,
-  getReplayContext,
   randomUuid,
   resolveMockValue,
   serializeValue,
   toJsonSafe,
-  toJsonSafeReport,
+  toJsonSafeReport
+} from "./chunk-EXT5FK54.js";
+import {
+  BitfabError,
+  DEFAULT_SERVICE_URL,
+  HttpClient,
+  getReplayContext,
   warnOnce
-} from "./chunk-BNOVHUQB.js";
+} from "./chunk-A22EYRSY.js";
 import {
   __privateAdd,
   __privateGet,
@@ -947,6 +949,131 @@ async function runFunctionWithBaml(bamlSource, inputs, providers, envVars) {
   };
 }
 
+// src/datasets.ts
+var DEFAULT_RERUN_TIMEOUT_MS = 9e4;
+var DEFAULT_RERUN_POLL_INTERVAL_MS = 1e3;
+var TERMINAL_RERUN_STATUSES = /* @__PURE__ */ new Set([
+  "completed",
+  "errored"
+]);
+function sleep(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+function datasetPath(datasetId, suffix = "") {
+  return `/api/sdk/datasets/${encodeURIComponent(datasetId)}${suffix}`;
+}
+var DatasetsClient = class {
+  constructor(httpClient) {
+    this.httpClient = httpClient;
+  }
+  /**
+   * Create a dataset, or update the one already named this way under the same
+   * trace function. `created` reports which happened. An omitted description
+   * leaves an existing one untouched.
+   */
+  async save(params) {
+    return this.httpClient.request("/api/sdk/datasets", {
+      traceFunctionKey: params.traceFunctionKey,
+      name: params.name,
+      ...params.description === void 0 ? {} : { description: params.description }
+    });
+  }
+  /**
+   * List datasets, scoped to one trace function when `traceFunctionKey` is
+   * given and organization-wide otherwise.
+   */
+  async list(params = {}) {
+    const query = params.traceFunctionKey === void 0 ? "" : `?traceFunctionKey=${encodeURIComponent(params.traceFunctionKey)}`;
+    const response = await this.httpClient.get(
+      `/api/sdk/datasets${query}`
+    );
+    return response.datasets;
+  }
+  /** Fetch one dataset by id. Rejects with a 404 `BitfabError` when it is not in this organization. */
+  async get(datasetId) {
+    const response = await this.httpClient.get(
+      datasetPath(datasetId)
+    );
+    return response.dataset;
+  }
+  /** The ids of every trace in the dataset, the same membership a replay with `datasetId` selects. */
+  async listTraces(datasetId) {
+    return this.httpClient.get(
+      datasetPath(datasetId, "/traces")
+    );
+  }
+  /**
+   * Add traces to the dataset (1 to 100 ids per call). Traces outside the
+   * organization or under another trace function are reported in
+   * `skippedTraceIds` rather than failing the call.
+   */
+  async addTraces(datasetId, traceIds) {
+    return this.httpClient.request(
+      datasetPath(datasetId, "/traces"),
+      { traceIds }
+    );
+  }
+  /** Remove traces from the dataset. The traces themselves are never deleted. */
+  async removeTraces(datasetId, traceIds) {
+    return this.httpClient.request(
+      datasetPath(datasetId, "/removeTraces"),
+      { traceIds }
+    );
+  }
+  /**
+   * Assign graders to the dataset (1 to 100 ids per call). Graders outside the
+   * organization or under another trace function are reported in
+   * `skippedGraderIds` rather than failing the call.
+   */
+  async addGraders(datasetId, graderIds) {
+    return this.httpClient.request(
+      datasetPath(datasetId, "/graders"),
+      { graderIds }
+    );
+  }
+  /** Unassign graders from the dataset. */
+  async removeGraders(datasetId, graderIds) {
+    return this.httpClient.request(
+      datasetPath(datasetId, "/removeGraders"),
+      { graderIds }
+    );
+  }
+  /**
+   * Re-run graders over every trace in the dataset. Defaults to every assigned
+   * grader; an unassigned id is rejected. Waits for the run to finish (up to
+   * `timeoutMs`, default 90s) unless `wait` is `false`, and returns the last
+   * run state seen either way. A request matching an in-flight run joins it.
+   */
+  async rerunGraders(datasetId, options = {}) {
+    const started = await this.httpClient.request(
+      datasetPath(datasetId, "/rerunGraders"),
+      options.graderIds === void 0 ? {} : { graderIds: options.graderIds }
+    );
+    if (options.wait === false) {
+      return started;
+    }
+    const deadline = Date.now() + (options.timeoutMs ?? DEFAULT_RERUN_TIMEOUT_MS);
+    const interval = options.pollIntervalMs ?? DEFAULT_RERUN_POLL_INTERVAL_MS;
+    let run = started.run;
+    while (!TERMINAL_RERUN_STATUSES.has(run.status) && Date.now() < deadline) {
+      await sleep(interval);
+      run = await this.getGraderRerun(datasetId, run.id) ?? run;
+    }
+    return { run, joinedExisting: started.joinedExisting };
+  }
+  /**
+   * The dataset's active grader re-run, or the run named by `runId`. Returns
+   * `null` when nothing is active or the run does not belong to this dataset.
+   */
+  async getGraderRerun(datasetId, runId) {
+    const query = runId === void 0 ? "" : `?runId=${encodeURIComponent(runId)}`;
+    const response = await this.httpClient.get(
+      datasetPath(datasetId, `/rerunGraders${query}`)
+    );
+    return response.run;
+  }
+};
+
 // src/dbSnapshot.ts
 var SUPPORTED_PROVIDERS = ["neon"];
 function validateDbSnapshotConfig(config) {
@@ -2658,6 +2785,7 @@ var Bitfab = class {
       serviceUrl: this.serviceUrl,
       timeout: this.timeout
     });
+    this.datasets = new DatasetsClient(this.httpClient);
   }
   /**
    * Decorate a class method as an automatically expanded trace root.
@@ -3568,6 +3696,7 @@ var Bitfab = class {
                 inputSourceTraceId: traceState?.inputSourceTraceId,
                 dbSnapshotRef: traceState?.dbSnapshotRef,
                 dropped: traceState?.dropped,
+                ingestionType: traceState?.ingestionType,
                 // Built AFTER the wrapped fn finished, so `accessed` reflects
                 // whether customer code obtained the branch URL during this
                 // item. Omitted entirely when no lease was attached, so the
@@ -3791,6 +3920,7 @@ var Bitfab = class {
     Object.defineProperty(wrappedFn, "_bitfabTraceFunctionKey", {
       value: traceFunctionKey
     });
+    Object.defineProperty(wrappedFn, "_bitfabWrappedFn", { value: fn });
     return wrappedFn;
   }
   /**
@@ -3953,6 +4083,9 @@ var Bitfab = class {
     if (params.dbSnapshotRef) {
       rawTrace.db_snapshot_ref = params.dbSnapshotRef;
     }
+    if (params.ingestionType) {
+      rawTrace.ingestion_type = params.ingestionType;
+    }
     if (params.dbSnapshotUsage) {
       rawTrace.db_snapshot_usage = {
         neon_branch_id: params.dbSnapshotUsage.neonBranchId,
@@ -4099,6 +4232,68 @@ var Bitfab = class {
   clearMockOverrides() {
     this.mockOverrides.length = 0;
   }
+  /**
+   * Write a replayable trace from a case, without running anything.
+   *
+   * Use this to turn a corpus you already hold (a Braintrust dataset, a
+   * spreadsheet, hand-written cases) into traces that {@link replay} can
+   * select. The recorded root span carries `input` as its input and `expected`
+   * as its output, so replay reports each item against the value you expected
+   * rather than against a previous run.
+   *
+   * A seeded trace has no child spans and no database pin, so replay mocking
+   * has nothing recorded to substitute and `dbBranch` refuses it. Pass
+   * `mockOverride` at replay time for calls that must not run.
+   *
+   * @returns The trace ID, usable with `replay({ traceIds: [...] })`.
+   */
+  seedTrace(traceFunctionKey, options) {
+    const { input } = options;
+    const fn = options.fn?._bitfabWrappedFn ?? options.fn;
+    if (fn && input.length < fn.length) {
+      throw new BitfabError(
+        `Seeded case supplies ${input.length} argument(s) but ${fn.name === "" ? "the function" : fn.name} requires ${fn.length}. Fix the case, or omit fn to seed it anyway.`
+      );
+    }
+    const traceId = randomUuid();
+    const startedAt = nowIsoTimestamp();
+    activeTraceStates.set(traceId, {
+      traceId,
+      startedAt,
+      contexts: [],
+      ingestionType: "seeded",
+      ...options.sessionId !== void 0 && { sessionId: options.sessionId },
+      ...options.metadata !== void 0 && { metadata: options.metadata }
+    });
+    try {
+      this.sendWrapperSpan({
+        traceFunctionKey,
+        spanName: options.spanName ?? traceFunctionKey,
+        traceId,
+        spanId: randomUuid(),
+        parentSpanId: null,
+        inputs: input,
+        result: options.expected,
+        startedAt,
+        endedAt: startedAt,
+        spanType: options.spanType ?? "agent",
+        captureContent: true
+      });
+      this.sendTraceCompletion({
+        traceFunctionKey,
+        traceId,
+        startedAt,
+        endedAt: startedAt,
+        sessionId: options.sessionId,
+        metadata: options.metadata,
+        contexts: [],
+        ingestionType: "seeded"
+      });
+    } finally {
+      activeTraceStates.delete(traceId);
+    }
+    return traceId;
+  }
   async replay(traceFunctionKey, fn, options) {
     const wrappedKey = fn._bitfabTraceFunctionKey;
     let replayFn = fn;
@@ -4113,7 +4308,7 @@ var Bitfab = class {
         `Function is wrapped with trace function key '${wrappedKey}' but replay was called with '${traceFunctionKey}'. Pass matching keys, or pass the unwrapped function to replay it under the explicit key.`
       );
     }
-    const { replay: doReplay } = await import("./replay-SMUBBMNK.js");
+    const { replay: doReplay } = await import("./replay-352POXG3.js");
     return doReplay(
       this.httpClient,
       this.serviceUrl,
@@ -4349,9 +4544,41 @@ var finalizers = {
 function defineReplayRegistry(registry) {
   return registry;
 }
+async function seedFromRegistry(registry, pipeline, cases) {
+  const registration = registry[pipeline];
+  if (registration === void 0) {
+    throw new BitfabError(
+      `Unknown pipeline '${pipeline}'. Registered: ${Object.keys(registry).join(", ")}`
+    );
+  }
+  const traceFunctionKey = resolveTraceFunctionKey(registration);
+  const traceIds = cases.map(
+    (seedCase) => registration.client.seedTrace(traceFunctionKey, {
+      input: seedCase.input,
+      expected: seedCase.expected,
+      fn: registration.fn,
+      metadata: seedCase.metadata,
+      sessionId: seedCase.sessionId
+    })
+  );
+  const { flushTraces: flushTraces2 } = await import("./http-2OFIGQOK.js");
+  await flushTraces2(3e4);
+  return { pipeline, traceFunctionKey, traceIds };
+}
+function resolveTraceFunctionKey(registration) {
+  const wrappedKey = registration.fn._bitfabTraceFunctionKey;
+  const key = registration.traceFunctionKey ?? wrappedKey;
+  if (key === void 0) {
+    throw new BitfabError(
+      "Replay registry entry uses a plain function. Set traceFunctionKey to the key its production handler records."
+    );
+  }
+  return key;
+}
 
 export {
   BitfabClaudeAgentHandler,
+  DatasetsClient,
   SUPPORTED_PROVIDERS,
   BitfabLangGraphCallbackHandler,
   BitfabLangGraphIntegration,
@@ -4364,6 +4591,7 @@ export {
   Bitfab,
   BitfabFunction,
   finalizers,
-  defineReplayRegistry
+  defineReplayRegistry,
+  seedFromRegistry
 };
-//# sourceMappingURL=chunk-UE4GGPM6.js.map
+//# sourceMappingURL=chunk-5NT4YDCQ.js.map