@workglow/eval 0.3.38 → 0.3.40

package/dist/commands/report.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../src/commands/report.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAGzC,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAG7C,sEAAsE;AACtE,wBAAsB,WAAW,CAC/B,MAAM,EAAE,UAAU,EAClB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,GACb,OAAO,CAAC,IAAI,CAAC,CAyDf;AAED,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,OAAO,EAChB,UAAU,EAAE,MAAM,OAAO,CAAC,UAAU,CAAC,GACpC,IAAI,CA2CN"}
+{"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../src/commands/report.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAGzC,OAAO,KAAK,EAAoB,UAAU,EAAE,MAAM,YAAY,CAAC;AAG/D,sEAAsE;AACtE,wBAAsB,WAAW,CAC/B,MAAM,EAAE,UAAU,EAClB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,GACb,OAAO,CAAC,IAAI,CAAC,CA0Ff;AAED,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,OAAO,EAChB,UAAU,EAAE,MAAM,OAAO,CAAC,UAAU,CAAC,GACpC,IAAI,CA2CN"}

package/dist/commands/run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../src/commands/run.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAMzC,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AA0C7C,wBAAgB,kBAAkB,CAChC,OAAO,EAAE,OAAO,EAChB,UAAU,EAAE,MAAM,OAAO,CAAC,UAAU,CAAC,EACrC,eAAe,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,GACnC,IAAI,CA0CN"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../src/commands/run.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAOzC,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AA2C7C,wBAAgB,kBAAkB,CAChC,OAAO,EAAE,OAAO,EAChB,UAAU,EAAE,MAAM,OAAO,CAAC,UAAU,CAAC,EACrC,eAAe,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,GACnC,IAAI,CA2CN"}

package/dist/evals/classify.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"classify.d.ts","sourceRoot":"","sources":["../../src/evals/classify.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAEhD,OAAO,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAE1D,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAE1E;;;;;GAKG;AACH,wBAAgB,aAAa,CAC3B,GAAG,EAAE,UAAU,EACf,WAAW,EAAE,MAAM,EACnB,UAAU,EAAE,UAAU,EACtB,eAAe,GAAE,SAAS,MAAM,EAAO,GACtC,MAAM,CAOR;AAED,mFAAmF;AACnF,wBAAgB,sBAAsB,CACpC,OAAO,EAAE,aAAa,EACtB,OAAO,EAAE,cAAc,GACtB,SAAS,MAAM,EAAE,CAQnB;AAED,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,CAOnF;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,WAAW,EAClB,OAAO,EAAE,aAAa,EACtB,OAAO,EAAE,cAAc,GACtB,WAAW,CAkCb"}
+{"version":3,"file":"classify.d.ts","sourceRoot":"","sources":["../../src/evals/classify.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAGhD,OAAO,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAG1D,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAE1E;;;;;GAKG;AACH,wBAAgB,aAAa,CAC3B,GAAG,EAAE,UAAU,EACf,WAAW,EAAE,MAAM,EACnB,UAAU,EAAE,UAAU,EACtB,eAAe,GAAE,SAAS,MAAM,EAAO,GACtC,MAAM,CAOR;AAED,mFAAmF;AACnF,wBAAgB,sBAAsB,CACpC,OAAO,EAAE,aAAa,EACtB,OAAO,EAAE,cAAc,GACtB,SAAS,MAAM,EAAE,CAQnB;AAED,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,CAOnF;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,WAAW,EAClB,OAAO,EAAE,aAAa,EACtB,OAAO,EAAE,cAAc,GACtB,WAAW,CAsCb"}

package/dist/evals/extract.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"extract.d.ts","sourceRoot":"","sources":["../../src/evals/extract.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAEhD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AAExD,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAW1E;;;;GAIG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,GAAG,YAAY,EAAE,CAShF;AAED,uFAAuF;AACvF,wBAAgB,uBAAuB,CACrC,YAAY,EAAE,SAAS,YAAY,EAAE,EACrC,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,GACpC,MAAM,EAAE,CAaV;AAED,wBAAgB,kBAAkB,CAChC,IAAI,EAAE,MAAM,EACZ,WAAW,EAAE,MAAM,EACnB,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,SAAS,MAAM,EAAE,GACxB,MAAM,CASR;AA+BD;;;;;;GAMG;AACH,wBAAgB,mBAAmB,CACjC,KAAK,EAAE,WAAW,EAClB,OAAO,EAAE,aAAa,EACtB,OAAO,EAAE,cAAc,GACtB,WAAW,CA4Cb"}
+{"version":3,"file":"extract.d.ts","sourceRoot":"","sources":["../../src/evals/extract.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAGhD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AAGxD,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAW1E;;;;GAIG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,GAAG,YAAY,EAAE,CAShF;AAED,uFAAuF;AACvF,wBAAgB,uBAAuB,CACrC,YAAY,EAAE,SAAS,YAAY,EAAE,EACrC,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,GACpC,MAAM,EAAE,CAaV;AAED,wBAAgB,kBAAkB,CAChC,IAAI,EAAE,MAAM,EACZ,WAAW,EAAE,MAAM,EACnB,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,SAAS,MAAM,EAAE,GACxB,MAAM,CASR;AA+BD;;;;;;GAMG;AACH,wBAAgB,mBAAmB,CACjC,KAAK,EAAE,WAAW,EAClB,OAAO,EAAE,aAAa,EACtB,OAAO,EAAE,cAAc,GACtB,WAAW,CAgDb"}

package/dist/evals/runner.d.ts

@@ -3,6 +3,8 @@
  * Copyright 2026 Steven Roussey <sroussey@gmail.com>
  * SPDX-License-Identifier: Apache-2.0
  */
+import type { ModelPricing } from "@workglow/ai";
+import type { StreamEvent, Usage } from "@workglow/task-graph";
 import type { EvalKind } from "../models";
 import type { DatasetRowRecord, EvalStores } from "../storage";
 import type { ColumnOptions, DatasetContext } from "./types";
@@ -13,8 +15,30 @@ export interface SweepOptions {
     readonly models: readonly string[];
     readonly columns: ColumnOptions;
     readonly context: DatasetContext;
-    readonly onProgress?: ((done: number, total: number, model: string, ok: boolean) => void) | undefined;
+    readonly onProgress?: ((done: number, total: number, model: string, ok: boolean, usage: Usage | undefined) => void) | undefined;
+    /**
+     * Forwarded to each row's executor as its stream-chunk listener. Left
+     * undefined for an ordinary sweep, so nothing subscribes and a 500-row run
+     * pays nothing for it.
+     */
+    readonly onStreamChunk?: ((event: StreamEvent) => void) | undefined;
 }
+/**
+ * Map one run's token accounting onto the result row's columns.
+ *
+ * An unreported counter is stored as `null`, never `0` — the database has to
+ * keep the distinction the in-memory type protects, or a model that never
+ * mentions caching becomes indistinguishable from one that cached nothing.
+ */
+export declare function usageColumns(usage: Usage | undefined, pricing: ModelPricing | undefined): {
+    input_tokens: number | null;
+    output_tokens: number | null;
+    cached_tokens: number | null;
+    cache_write_tokens: number | null;
+    total_tokens: number | null;
+    cost: number | null;
+    currency: string | null;
+};
 /**
  * Run every stored dataset row through the eval workflow for each model,
  * persisting one result row per (run, model, dataset row). Failures are

package/dist/evals/runner.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAIH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,WAAW,CAAC;AAE1C,OAAO,KAAK,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAI/D,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAe,MAAM,SAAS,CAAC;AAE1E,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC;IACxB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,CAAC;IACnC,QAAQ,CAAC,OAAO,EAAE,aAAa,CAAC;IAChC,QAAQ,CAAC,OAAO,EAAE,cAAc,CAAC;IACjC,QAAQ,CAAC,UAAU,CAAC,EAClB,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,OAAO,KAAK,IAAI,CAAC,GAAG,SAAS,CAAC;CACnF;AAoBD;;;;;GAKG;AACH,wBAAsB,QAAQ,CAC5B,MAAM,EAAE,UAAU,EAClB,IAAI,EAAE,SAAS,gBAAgB,EAAE,EACjC,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAwEjB"}
+{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAe,YAAY,EAAE,MAAM,cAAc,CAAC;AAE9D,OAAO,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,sBAAsB,CAAC;AAE/D,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,WAAW,CAAC;AAE1C,OAAO,KAAK,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAI/D,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAe,MAAM,SAAS,CAAC;AAE1E,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC;IACxB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,CAAC;IACnC,QAAQ,CAAC,OAAO,EAAE,aAAa,CAAC;IAChC,QAAQ,CAAC,OAAO,EAAE,cAAc,CAAC;IACjC,QAAQ,CAAC,UAAU,CAAC,EAChB,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,GAAG,SAAS,KAAK,IAAI,CAAC,GAC7F,SAAS,CAAC;IACd;;;;OAIG;IACH,QAAQ,CAAC,aAAa,CAAC,EAAE,CAAC,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,CAAC,GAAG,SAAS,CAAC;CACrE;AAED;;;;;;GAMG;AACH,wBAAgB,YAAY,CAC1B,KAAK,EAAE,KAAK,GAAG,SAAS,EACxB,OAAO,EAAE,YAAY,GAAG,SAAS,GAChC;IACD,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,kBAAkB,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IACpB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;CACzB,CAWA;AAoBD;;;;;GAKG;AACH,wBAAsB,QAAQ,CAC5B,MAAM,EAAE,UAAU,EAClB,IAAI,EAAE,SAAS,gBAAgB,EAAE,EACjC,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CA6EjB"}

package/dist/evals/similarity.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"similarity.d.ts","sourceRoot":"","sources":["../../src/evals/similarity.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAIhD,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAE1E;;;;;GAKG;AACH,wBAAgB,sBAAsB,CACpC,KAAK,EAAE,WAAW,EAClB,OAAO,EAAE,aAAa,EACtB,OAAO,EAAE,cAAc,GACtB,WAAW,CAiCb"}
+{"version":3,"file":"similarity.d.ts","sourceRoot":"","sources":["../../src/evals/similarity.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAMhD,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAE1E;;;;;GAKG;AACH,wBAAgB,sBAAsB,CACpC,KAAK,EAAE,WAAW,EAClB,OAAO,EAAE,aAAa,EACtB,OAAO,EAAE,cAAc,GACtB,WAAW,CAqCb"}

package/dist/evals/streamSubscribe.d.ts

@@ -0,0 +1,15 @@
+/**
+ * @license
+ * Copyright 2026 Steven Roussey <sroussey@gmail.com>
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import type { StreamEvent, Workflow } from "@workglow/task-graph";
+/**
+ * Run a workflow, forwarding its `stream_chunk` events to `onStreamChunk` for
+ * the duration of the run only.
+ *
+ * When no listener is given, nothing is subscribed at all — a 500-row sweep
+ * without `--stream` pays nothing for it.
+ */
+export declare function runWithStreamChunks<Output>(workflow: Workflow, onStreamChunk: ((event: StreamEvent) => void) | undefined): Promise<Output>;
+//# sourceMappingURL=streamSubscribe.d.ts.map

package/dist/evals/streamSubscribe.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"streamSubscribe.d.ts","sourceRoot":"","sources":["../../src/evals/streamSubscribe.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAc,QAAQ,EAAE,MAAM,sBAAsB,CAAC;AAE9E;;;;;;GAMG;AACH,wBAAsB,mBAAmB,CAAC,MAAM,EAC9C,QAAQ,EAAE,QAAQ,EAClB,aAAa,EAAE,CAAC,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,CAAC,GAAG,SAAS,GACxD,OAAO,CAAC,MAAM,CAAC,CAUjB"}

package/dist/evals/types.d.ts

@@ -3,6 +3,7 @@
  * Copyright 2026 Steven Roussey <sroussey@gmail.com>
  * SPDX-License-Identifier: Apache-2.0
  */
+import type { StreamEvent, Usage } from "@workglow/task-graph";
 import type { DatasetRow, LabelNames } from "../hf/types";
 /** What one workflow execution produced for one dataset row. */
 export interface RowPrediction {
@@ -10,9 +11,16 @@ export interface RowPrediction {
     readonly predicted?: string | undefined;
     readonly expectedValue?: number | undefined;
     readonly predictedValue?: number | undefined;
+    readonly usage?: Usage | undefined;
 }
-/** Runs the eval workflow for a single dataset row against one model. */
-export type RowExecutor = (row: DatasetRow) => Promise<RowPrediction>;
+/**
+ * Runs the eval workflow for a single dataset row against one model.
+ *
+ * `onStreamChunk`, when passed, is subscribed to the row's underlying task for
+ * the life of the call only — `--stream` is the only caller that passes it, so
+ * an ordinary sweep never pays for the subscription.
+ */
+export type RowExecutor = (row: DatasetRow, onStreamChunk?: (event: StreamEvent) => void) => Promise<RowPrediction>;
 export interface ColumnOptions {
     readonly textColumn: string;
     /** classify: gold label column. */

package/dist/evals/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/evals/types.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAE1D,gEAAgE;AAChE,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACvC,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACxC,QAAQ,CAAC,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5C,QAAQ,CAAC,cAAc,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC9C;AAED,yEAAyE;AACzE,MAAM,MAAM,WAAW,GAAG,CAAC,GAAG,EAAE,UAAU,KAAK,OAAO,CAAC,aAAa,CAAC,CAAC;AAEtE,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,mCAAmC;IACnC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,8EAA8E;IAC9E,QAAQ,CAAC,MAAM,CAAC,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,CAAC;IAChD,0CAA0C;IAC1C,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,gDAAgD;IAChD,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,6EAA6E;IAC7E,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;IAChC,mEAAmE;IACnE,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,oFAAoF;IACpF,QAAQ,CAAC,MAAM,CAAC,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,CAAC;IAChD,uDAAuD;IACvD,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC3C;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,OAAO,EAAE,SAAS,MAAM,EAAE,CAAC;IACpC,QAAQ,CAAC,UAAU,EAAE,UAAU,CAAC;CACjC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/evals/types.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,sBAAsB,CAAC;AAC/D,OAAO,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAE1D,gEAAgE;AAChE,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACvC,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACxC,QAAQ,CAAC,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5C,QAAQ,CAAC,cAAc,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7C,QAAQ,CAAC,KAAK,CAAC,EAAE,KAAK,GAAG,SAAS,CAAC;CACpC;AAED;;;;;;GAMG;AACH,MAAM,MAAM,WAAW,GAAG,CACxB,GAAG,EAAE,UAAU,EACf,aAAa,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,KACzC,OAAO,CAAC,aAAa,CAAC,CAAC;AAE5B,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,mCAAmC;IACnC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,8EAA8E;IAC9E,QAAQ,CAAC,MAAM,CAAC,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,CAAC;IAChD,0CAA0C;IAC1C,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,gDAAgD;IAChD,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,6EAA6E;IAC7E,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;IAChC,mEAAmE;IACnE,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,oFAAoF;IACpF,QAAQ,CAAC,MAAM,CAAC,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,CAAC;IAChD,uDAAuD;IACvD,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC3C;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,OAAO,EAAE,SAAS,MAAM,EAAE,CAAC;IACpC,QAAQ,CAAC,UAAU,EAAE,UAAU,CAAC;CACjC"}

package/dist/models.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAMhD,MAAM,MAAM,QAAQ,GAAG,UAAU,GAAG,YAAY,GAAG,SAAS,CAAC;AAI7D;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,wBAAgB,kBAAkB,CAAC,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,GAAG,WAAW,CA8C1E;AAED;;;;;GAKG;AACH,wBAAsB,yBAAyB,CAAC,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC,CAkBzF;AAED;;;;;GAKG;AACH,wBAAsB,qBAAqB,CAAC,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CAI9E;AAQD,yEAAyE;AACzE,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,CASvD"}
+{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAgB,MAAM,cAAc,CAAC;AAM9D,MAAM,MAAM,QAAQ,GAAG,UAAU,GAAG,YAAY,GAAG,SAAS,CAAC;AA8J7D;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,wBAAgB,kBAAkB,CAAC,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,GAAG,WAAW,CA8D1E;AAED;;;;;GAKG;AACH,wBAAsB,yBAAyB,CAAC,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC,CAkBzF;AAED;;;;;GAKG;AACH,wBAAsB,qBAAqB,CAAC,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CAI9E;AAQD,yEAAyE;AACzE,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,CASvD"}

package/dist/report/aggregate.d.ts

@@ -32,4 +32,17 @@ export interface AggregateOptions {
  * tie-breaker.
  */
 export declare function aggregateResults(kind: EvalKind, results: readonly EvalResultRecord[], options?: AggregateOptions): ModelReport[];
+/**
+ * Sum a model's token spend across a run's result rows.
+ *
+ * A null column means the provider reported nothing for that row, so it is
+ * skipped rather than counted as 0 — a model that never reports caching must not
+ * end up looking like one that cached nothing.
+ */
+export declare function sumUsageColumns(rows: readonly EvalResultRecord[]): {
+    inputTokens: number | undefined;
+    outputTokens: number | undefined;
+    cachedTokens: number | undefined;
+    cost: number | undefined;
+};
 //# sourceMappingURL=aggregate.d.ts.map

package/dist/report/aggregate.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"aggregate.d.ts","sourceRoot":"","sources":["../../src/report/aggregate.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,WAAW,CAAC;AAK1C,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAEnD,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,uDAAuD;IACvD,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,wFAAwF;IACxF,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,6FAA6F;IAC7F,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC/B;AAED,wEAAwE;AACxE,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACvC,QAAQ,CAAC,MAAM,CAAC,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,CAAC;CACjD;AAWD;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAC9B,IAAI,EAAE,QAAQ,EACd,OAAO,EAAE,SAAS,gBAAgB,EAAE,EACpC,OAAO,GAAE,gBAAqB,GAC7B,WAAW,EAAE,CAmDf"}
+{"version":3,"file":"aggregate.d.ts","sourceRoot":"","sources":["../../src/report/aggregate.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,WAAW,CAAC;AAK1C,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAEnD,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,uDAAuD;IACvD,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,wFAAwF;IACxF,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,6FAA6F;IAC7F,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC/B;AAED,wEAAwE;AACxE,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACvC,QAAQ,CAAC,MAAM,CAAC,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,CAAC;CACjD;AAWD;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAC9B,IAAI,EAAE,QAAQ,EACd,OAAO,EAAE,SAAS,gBAAgB,EAAE,EACpC,OAAO,GAAE,gBAAqB,GAC7B,WAAW,EAAE,CAmDf;AA0BD;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,SAAS,gBAAgB,EAAE,GAAG;IAClE,WAAW,EAAE,MAAM,GAAG,SAAS,CAAC;IAChC,YAAY,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,YAAY,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,IAAI,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1B,CAgBA"}

package/dist/report/liveTally.d.ts

@@ -0,0 +1,22 @@
+/**
+ * @license
+ * Copyright 2026 Steven Roussey <sroussey@gmail.com>
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import type { ModelPricing } from "@workglow/ai";
+import type { Usage } from "@workglow/task-graph";
+/**
+ * Running per-model totals for a sweep, rendered as a small table on stderr.
+ *
+ * Sums usage across rows because each row is a separate request — unlike a task's
+ * cumulative in-run total, these genuinely add up.
+ */
+export declare class LiveTally {
+    private readonly pricing;
+    private readonly byModel;
+    private readonly startedAt;
+    constructor(pricing: (model: string) => ModelPricing | undefined, now: number);
+    record(model: string, ok: boolean, usage: Usage | undefined): void;
+    render(done: number, total: number, now: number): string;
+}
+//# sourceMappingURL=liveTally.d.ts.map

package/dist/report/liveTally.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"liveTally.d.ts","sourceRoot":"","sources":["../../src/report/liveTally.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAEjD,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,sBAAsB,CAAC;AASlD;;;;;GAKG;AACH,qBAAa,SAAS;IAKlB,OAAO,CAAC,QAAQ,CAAC,OAAO;IAJ1B,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAiC;IACzD,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAS;IAEnC,YACmB,OAAO,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,YAAY,GAAG,SAAS,EACrE,GAAG,EAAE,MAAM,EAGZ;IAED,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,GAAG,SAAS,GAAG,IAAI,CAMjE;IAED,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,GAAG,MAAM,CAevD;CACF"}

package/dist/storage.d.ts

@@ -133,6 +133,27 @@ export declare const EvalResultSchema: {
         readonly latency_ms: {
             readonly type: "number";
         };
+        readonly input_tokens: {
+            readonly type: readonly ["integer", "null"];
+        };
+        readonly output_tokens: {
+            readonly type: readonly ["integer", "null"];
+        };
+        readonly cached_tokens: {
+            readonly type: readonly ["integer", "null"];
+        };
+        readonly cache_write_tokens: {
+            readonly type: readonly ["integer", "null"];
+        };
+        readonly total_tokens: {
+            readonly type: readonly ["integer", "null"];
+        };
+        readonly cost: {
+            readonly type: readonly ["number", "null"];
+        };
+        readonly currency: {
+            readonly type: readonly ["string", "null"];
+        };
     };
     readonly required: readonly ["run_id", "model", "row_index", "ok", "latency_ms"];
     readonly additionalProperties: false;

package/dist/storage.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"storage.d.ts","sourceRoot":"","sources":["../src/storage.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAEzD,OAAO,KAAK,EAAwB,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAC9E,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,UAAU,CAAC;AAE3C;;;GAGG;AACH,eAAO,MAAM,iBAAiB;mBACtB,QAAQ;;iBAEZ,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,KAAK;qBAAI,IAAI,EAAE,QAAQ;;iBACvB,MAAM;qBAAI,IAAI,EAAE,QAAQ;;iBACxB,QAAQ;qBAAI,IAAI,EAAE,QAAQ;;iBAC1B,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,WAAW;qBAAI,IAAI;;iBACnB,MAAM;qBAAI,IAAI,EAAE,QAAQ;;iBACxB,UAAU;qBAAI,IAAI,EAAE,QAAQ;;;;;CAIS,CAAC;AAE1C,eAAO,MAAM,0BAA0B,YAAI,SAAS,EAAE,OAAO,CAAU,CAAC;AAExE,2EAA2E;AAC3E,eAAO,MAAM,gBAAgB;mBACrB,QAAQ;;iBAEZ,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,KAAK;qBAAI,IAAI,EAAE,QAAQ;;iBACvB,SAAS;qBAAI,IAAI,EAAE,QAAQ;;iBAC3B,IAAI;qBAAI,IAAI,EAAE,QAAQ;;;;;CAIe,CAAC;AAE1C,eAAO,MAAM,yBAAyB,YAAI,SAAS,EAAE,OAAO,EAAE,WAAW,CAAU,CAAC;AACpF,MAAM,MAAM,gBAAgB,GAAG,UAAU,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAEnE,gFAAgF;AAChF,eAAO,MAAM,aAAa;mBAClB,QAAQ;;iBAEZ,MAAM;qBAAI,IAAI,EAAE,QAAQ;qBAAE,kBAAkB;;iBAC5C,IAAI;qBAAI,IAAI,EAAE,QAAQ;;iBACtB,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,KAAK;qBAAI,IAAI,EAAE,QAAQ;;iBACvB,MAAM;qBAAI,IAAI,EAAE,QAAQ;;iBACxB,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,UAAU;qBAAI,IAAI,EAAE,QAAQ;;;;;CAIS,CAAC;AAE1C,eAAO,MAAM,sBAAsB,YAAI,QAAQ,CAAU,CAAC;AAE1D;;;;GAIG;AACH,eAAO,MAAM,gBAAgB;mBACrB,QAAQ;;iBAEZ,MAAM;qBAAI,IAAI,EAAE,QAAQ;;iBACxB,KAAK;qBAAI,IAAI,EAAE,QAAQ;;iBACvB,SAAS;qBAAI,IAAI,EAAE,QAAQ;;iBAC3B,EAAE;qBAAI,IAAI,EAAE,QAAQ;;iBACpB,KAAK;qBAAI,IAAI;;iBACb,QAAQ;qBAAI,IAAI;;iBAChB,SAAS;qBAAI,IAAI;;iBACjB,cAAc;qBAAI,IAAI;;iBACtB,eAAe;qBAAI,IAAI;;iBACvB,UAAU;qBAAI,IAAI,EAAE,QAAQ;;;;;CAIS,CAAC;AAE1C,eAAO,MAAM,yBAAyB,YAAI,QAAQ,EAAE,OAAO,EAAE,WAAW,CAAU,CAAC;AACnF,MAAM,MAAM,gBAAgB,GAAG,UAAU,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAEnE,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,QAAQ,EAAE,eAAe,CAAC,OAAO,iBAAiB,EAAE,OAAO,0BAA0B,CAAC,CAAC;IAChG,QAAQ,CAAC,IAAI,EAAE,eAAe,CAAC,OAAO,gBAAgB,EAAE,OAAO,yBAAyB,CAAC,CAAC;IAC1F,QAAQ,CAAC,IAAI,EAAE,eAAe,CAAC,OAAO,aAAa,EAAE,OAAO,sBAAsB,CAAC,CAAC;IACpF,QAAQ,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,gBAAgB,EAAE,OAAO,yBAAyB,CAAC,CAAC;CAC9F;AAED,mEAAmE;AACnE,wBAAsB,kBAAkB,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC,CA0BhF;AAED,oEAAoE;AACpE,wBAAsB,oBAAoB,IAAI,OAAO,CAAC,UAAU,CAAC,CAShE"}
+{"version":3,"file":"storage.d.ts","sourceRoot":"","sources":["../src/storage.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAEzD,OAAO,KAAK,EAAwB,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAC9E,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,UAAU,CAAC;AAE3C;;;GAGG;AACH,eAAO,MAAM,iBAAiB;mBACtB,QAAQ;;iBAEZ,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,KAAK;qBAAI,IAAI,EAAE,QAAQ;;iBACvB,MAAM;qBAAI,IAAI,EAAE,QAAQ;;iBACxB,QAAQ;qBAAI,IAAI,EAAE,QAAQ;;iBAC1B,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,WAAW;qBAAI,IAAI;;iBACnB,MAAM;qBAAI,IAAI,EAAE,QAAQ;;iBACxB,UAAU;qBAAI,IAAI,EAAE,QAAQ;;;;;CAIS,CAAC;AAE1C,eAAO,MAAM,0BAA0B,YAAI,SAAS,EAAE,OAAO,CAAU,CAAC;AAExE,2EAA2E;AAC3E,eAAO,MAAM,gBAAgB;mBACrB,QAAQ;;iBAEZ,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,KAAK;qBAAI,IAAI,EAAE,QAAQ;;iBACvB,SAAS;qBAAI,IAAI,EAAE,QAAQ;;iBAC3B,IAAI;qBAAI,IAAI,EAAE,QAAQ;;;;;CAIe,CAAC;AAE1C,eAAO,MAAM,yBAAyB,YAAI,SAAS,EAAE,OAAO,EAAE,WAAW,CAAU,CAAC;AACpF,MAAM,MAAM,gBAAgB,GAAG,UAAU,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAEnE,gFAAgF;AAChF,eAAO,MAAM,aAAa;mBAClB,QAAQ;;iBAEZ,MAAM;qBAAI,IAAI,EAAE,QAAQ;qBAAE,kBAAkB;;iBAC5C,IAAI;qBAAI,IAAI,EAAE,QAAQ;;iBACtB,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,KAAK;qBAAI,IAAI,EAAE,QAAQ;;iBACvB,MAAM;qBAAI,IAAI,EAAE,QAAQ;;iBACxB,OAAO;qBAAI,IAAI,EAAE,QAAQ;;iBACzB,UAAU;qBAAI,IAAI,EAAE,QAAQ;;;;;CAIS,CAAC;AAE1C,eAAO,MAAM,sBAAsB,YAAI,QAAQ,CAAU,CAAC;AAE1D;;;;GAIG;AACH,eAAO,MAAM,gBAAgB;mBACrB,QAAQ;;iBAEZ,MAAM;qBAAI,IAAI,EAAE,QAAQ;;iBACxB,KAAK;qBAAI,IAAI,EAAE,QAAQ;;iBACvB,SAAS;qBAAI,IAAI,EAAE,QAAQ;;iBAC3B,EAAE;qBAAI,IAAI,EAAE,QAAQ;;iBACpB,KAAK;qBAAI,IAAI;;iBACb,QAAQ;qBAAI,IAAI;;iBAChB,SAAS;qBAAI,IAAI;;iBACjB,cAAc;qBAAI,IAAI;;iBACtB,eAAe;qBAAI,IAAI;;iBACvB,UAAU;qBAAI,IAAI,EAAE,QAAQ;;iBAC5B,YAAY;qBAAI,IAAI;;iBACpB,aAAa;qBAAI,IAAI;;iBACrB,aAAa;qBAAI,IAAI;;iBACrB,kBAAkB;qBAAI,IAAI;;iBAC1B,YAAY;qBAAI,IAAI;;iBACpB,IAAI;qBAAI,IAAI;;iBACZ,QAAQ;qBAAI,IAAI;;;;;CAIqB,CAAC;AAE1C,eAAO,MAAM,yBAAyB,YAAI,QAAQ,EAAE,OAAO,EAAE,WAAW,CAAU,CAAC;AACnF,MAAM,MAAM,gBAAgB,GAAG,UAAU,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAEnE,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,QAAQ,EAAE,eAAe,CAAC,OAAO,iBAAiB,EAAE,OAAO,0BAA0B,CAAC,CAAC;IAChG,QAAQ,CAAC,IAAI,EAAE,eAAe,CAAC,OAAO,gBAAgB,EAAE,OAAO,yBAAyB,CAAC,CAAC;IAC1F,QAAQ,CAAC,IAAI,EAAE,eAAe,CAAC,OAAO,aAAa,EAAE,OAAO,sBAAsB,CAAC,CAAC;IACpF,QAAQ,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,gBAAgB,EAAE,OAAO,yBAAyB,CAAC,CAAC;CAC9F;AAED,mEAAmE;AACnE,wBAAsB,kBAAkB,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC,CA0BhF;AAED,oEAAoE;AACpE,wBAAsB,oBAAoB,IAAI,OAAO,CAAC,UAAU,CAAC,CAShE"}

package/dist/test/liveTally.test.d.ts

@@ -0,0 +1,7 @@
+/**
+ * @license
+ * Copyright 2026 Steven Roussey <sroussey@gmail.com>
+ * SPDX-License-Identifier: Apache-2.0
+ */
+export {};
+//# sourceMappingURL=liveTally.test.d.ts.map

package/dist/test/liveTally.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"liveTally.test.d.ts","sourceRoot":"","sources":["../../src/test/liveTally.test.ts"],"names":[],"mappings":"AAAA;;;;GAIG"}

package/dist/test/runSweep.test.d.ts

@@ -0,0 +1,7 @@
+/**
+ * @license
+ * Copyright 2026 Steven Roussey <sroussey@gmail.com>
+ * SPDX-License-Identifier: Apache-2.0
+ */
+export {};
+//# sourceMappingURL=runSweep.test.d.ts.map

package/dist/test/runSweep.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runSweep.test.d.ts","sourceRoot":"","sources":["../../src/test/runSweep.test.ts"],"names":[],"mappings":"AAAA;;;;GAIG"}

package/dist/test/streamSubscribe.test.d.ts

@@ -0,0 +1,7 @@
+/**
+ * @license
+ * Copyright 2026 Steven Roussey <sroussey@gmail.com>
+ * SPDX-License-Identifier: Apache-2.0
+ */
+export {};
+//# sourceMappingURL=streamSubscribe.test.d.ts.map

package/dist/test/streamSubscribe.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"streamSubscribe.test.d.ts","sourceRoot":"","sources":["../../src/test/streamSubscribe.test.ts"],"names":[],"mappings":"AAAA;;;;GAIG"}

package/dist/test/usageColumns.test.d.ts

@@ -0,0 +1,7 @@
+/**
+ * @license
+ * Copyright 2026 Steven Roussey <sroussey@gmail.com>
+ * SPDX-License-Identifier: Apache-2.0
+ */
+export {};
+//# sourceMappingURL=usageColumns.test.d.ts.map

package/dist/test/usageColumns.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"usageColumns.test.d.ts","sourceRoot":"","sources":["../../src/test/usageColumns.test.ts"],"names":[],"mappings":"AAAA;;;;GAIG"}

package/dist/workglow-eval.js

@@ -562,6 +562,24 @@ function extractionCountsFor(pairs, options) {
   }
   return counts;
 }
+function sumUsageColumns(rows) {
+  const sum = (pick) => {
+    let total;
+    for (const row of rows) {
+      const value = pick(row);
+      if (value === null || value === undefined)
+        continue;
+      total = (total ?? 0) + value;
+    }
+    return total;
+  };
+  return {
+    inputTokens: sum((r) => r.input_tokens),
+    outputTokens: sum((r) => r.output_tokens),
+    cachedTokens: sum((r) => r.cached_tokens),
+    cost: sum((r) => r.cost)
+  };
+}
 
 // src/commands/report.ts
 async function printReport(stores, runId, format) {
@@ -589,22 +607,55 @@ async function printReport(stores, runId, format) {
   }
   console.log(`run ${run.run_id} \u2014 ${run.kind} on ${run.dataset} [${run.split}]`);
   const REPORT_COLUMNS = {
-    classify: ["model", "rows", "ok", "accuracy", "avg_ms"],
-    similarity: ["model", "rows", "ok", "pearson", "spearman", "avg_ms"],
-    extract: ["model", "rows", "ok", "score", "found", "prec", "avg_ms"]
+    classify: ["model", "rows", "ok", "accuracy", "avg_ms", "tokens", "cached", "cost"],
+    similarity: [
+      "model",
+      "rows",
+      "ok",
+      "pearson",
+      "spearman",
+      "avg_ms",
+      "tokens",
+      "cached",
+      "cost"
+    ],
+    extract: [
+      "model",
+      "rows",
+      "ok",
+      "score",
+      "found",
+      "prec",
+      "avg_ms",
+      "tokens",
+      "cached",
+      "cost"
+    ]
   };
-  const tableRows = reports.map((r) => ({
-    model: r.model,
-    rows: String(r.rows),
-    ok: String(r.okRows),
-    accuracy: formatMetric(r.accuracy),
-    pearson: formatMetric(r.pearson),
-    spearman: formatMetric(r.spearman),
-    score: formatMetric(r.score),
-    found: formatMetric(r.found),
-    prec: formatMetric(r.prec),
-    avg_ms: formatMetric(r.avgLatencyMs, 0)
-  }));
+  const resultsByModel = new Map;
+  for (const result of results) {
+    const list = resultsByModel.get(result.model) ?? [];
+    list.push(result);
+    resultsByModel.set(result.model, list);
+  }
+  const tableRows = reports.map((r) => {
+    const totals = sumUsageColumns(resultsByModel.get(r.model) ?? []);
+    return {
+      model: r.model,
+      rows: String(r.rows),
+      ok: String(r.okRows),
+      accuracy: formatMetric(r.accuracy),
+      pearson: formatMetric(r.pearson),
+      spearman: formatMetric(r.spearman),
+      score: formatMetric(r.score),
+      found: formatMetric(r.found),
+      prec: formatMetric(r.prec),
+      avg_ms: formatMetric(r.avgLatencyMs, 0),
+      tokens: `${totals.inputTokens ?? "-"}/${totals.outputTokens ?? "-"}`,
+      cached: String(totals.cachedTokens ?? "-"),
+      cost: totals.cost === undefined ? "-" : `$${totals.cost.toFixed(4)}`
+    };
+  });
   console.log(formatTable(tableRows, REPORT_COLUMNS[kind]));
   const failures = results.filter((r) => r.ok !== 1);
   if (failures.length > 0) {
@@ -647,6 +698,9 @@ function registerReportCommand(program, openStores) {
   });
 }
 
+// src/evals/runner.ts
+import { estimateCost } from "@workglow/ai";
+
 // src/models.ts
 import { downloadModel } from "@workglow/ai";
 
@@ -673,6 +727,136 @@ function loadConfig() {
 
 // src/models.ts
 var GGUF_PREFIX = "gguf:";
+var ANTHROPIC_PRICING = {
+  "claude-haiku-4-5": {
+    currency: "USD",
+    input: 1,
+    output: 5,
+    cached: 0.1,
+    cacheWrite: 1.25,
+    cacheStoragePerHour: undefined
+  },
+  "claude-sonnet-5": {
+    currency: "USD",
+    input: 2,
+    output: 10,
+    cached: 0.2,
+    cacheWrite: 2.5,
+    cacheStoragePerHour: undefined
+  },
+  "claude-opus-5": {
+    currency: "USD",
+    input: 15,
+    output: 75,
+    cached: 1.5,
+    cacheWrite: 18.75,
+    cacheStoragePerHour: undefined
+  }
+};
+var OPENAI_PRICING = {
+  "gpt-5.6-sol": {
+    currency: "USD",
+    input: 5,
+    output: 30,
+    cached: 0.5,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  },
+  "gpt-5.6-terra": {
+    currency: "USD",
+    input: 2,
+    output: 12,
+    cached: 0.2,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  },
+  "gpt-5.6-luna": {
+    currency: "USD",
+    input: 0.2,
+    output: 1.2,
+    cached: 0.02,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  },
+  "gpt-5.5": {
+    currency: "USD",
+    input: 5,
+    output: 15,
+    cached: 2.5,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  },
+  "gpt-5.4-mini": {
+    currency: "USD",
+    input: 0.4,
+    output: 1.6,
+    cached: 0.2,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  },
+  "text-embedding-3-small": {
+    currency: "USD",
+    input: 0.02,
+    output: undefined,
+    cached: undefined,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  }
+};
+var GEMINI_PRICING = {
+  "gemini-3.1-pro-preview": {
+    currency: "USD",
+    input: 2.5,
+    output: 10,
+    cached: 0.625,
+    cacheWrite: undefined,
+    cacheStoragePerHour: 4.5
+  },
+  "gemini-3-flash-preview": {
+    currency: "USD",
+    input: 0.3,
+    output: 2.5,
+    cached: 0.075,
+    cacheWrite: undefined,
+    cacheStoragePerHour: 1
+  }
+};
+var XAI_PRICING = {
+  "grok-4.6": {
+    currency: "USD",
+    input: 2,
+    output: 6,
+    cached: 0.5,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  }
+};
+var DEEPSEEK_PRICING = {
+  "deepseek-v4-flash": {
+    currency: "USD",
+    input: 0.14,
+    output: 0.28,
+    cached: 0.014,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  },
+  "deepseek-v4-pro": {
+    currency: "USD",
+    input: 0.56,
+    output: 1.68,
+    cached: 0.07,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  },
+  "deepseek-v4-pro-0813": {
+    currency: "USD",
+    input: 0.435,
+    output: 0.87,
+    cached: 0.0435,
+    cacheWrite: undefined,
+    cacheStoragePerHour: undefined
+  }
+};
 function resolveModelConfig(id, kind) {
   if (id.startsWith(GGUF_PREFIX)) {
     const ref = id.slice(GGUF_PREFIX.length);
@@ -702,19 +886,35 @@ function resolveModelConfig(id, kind) {
     };
   }
   if (/^claude-/.test(id)) {
-    return { provider: "ANTHROPIC", provider_config: { model_name: id, max_tokens: 1024 } };
+    return {
+      provider: "ANTHROPIC",
+      provider_config: { model_name: id },
+      pricing: ANTHROPIC_PRICING[id]
+    };
   }
   if (/^(?:gpt-|o\d|chatgpt-|text-embedding-)/.test(id)) {
-    return { provider: "OPENAI", provider_config: { model_name: id } };
+    return {
+      provider: "OPENAI",
+      provider_config: { model_name: id },
+      pricing: OPENAI_PRICING[id]
+    };
   }
   if (/^gemini-/.test(id)) {
-    return { provider: "GOOGLE_GEMINI", provider_config: { model_name: id } };
+    return {
+      provider: "GOOGLE_GEMINI",
+      provider_config: { model_name: id },
+      pricing: GEMINI_PRICING[id]
+    };
   }
   if (/^grok-/.test(id)) {
-    return { provider: "XAI", provider_config: { model_name: id } };
+    return { provider: "XAI", provider_config: { model_name: id }, pricing: XAI_PRICING[id] };
   }
   if (/^deepseek-/.test(id)) {
-    return { provider: "DEEPSEEK", provider_config: { model_name: id } };
+    return {
+      provider: "DEEPSEEK",
+      provider_config: { model_name: id },
+      pricing: DEEPSEEK_PRICING[id]
+    };
   }
   throw new Error(`cannot infer a provider for model "${id}" \u2014 use a claude-*/gpt-*/gemini-*/grok-*/deepseek-* ` + `cloud id or an org/name HuggingFace ONNX model path`);
 }
@@ -759,7 +959,7 @@ function parseModelList(models) {
 }
 
 // src/evals/classify.ts
-import { Workflow } from "@workglow/task-graph";
+import { USAGE_OUTPUT_KEY, Workflow } from "@workglow/task-graph";
 
 // src/evals/prompt.ts
 function fenceText(text) {
@@ -771,6 +971,19 @@ ${text}
 ${fence}`;
 }
 
+// src/evals/streamSubscribe.ts
+async function runWithStreamChunks(workflow, onStreamChunk) {
+  const listener = onStreamChunk ? (_taskId, event) => onStreamChunk(event) : undefined;
+  if (listener)
+    workflow.on("stream_chunk", listener);
+  try {
+    return await workflow.run();
+  } finally {
+    if (listener)
+      workflow.off("stream_chunk", listener);
+  }
+}
+
 // src/evals/classify.ts
 function expectedLabel(row, labelColumn, labelNames, candidateLabels = []) {
   const raw = row[labelColumn];
@@ -810,7 +1023,7 @@ function makeClassifyExecutor(model, options, context) {
     required: ["label"],
     additionalProperties: false
   };
-  return async (row) => {
+  return async (row, onStreamChunk) => {
     const text = String(row[options.textColumn] ?? "");
     const workflow = new Workflow;
     workflow.structuredGeneration({
@@ -820,16 +1033,17 @@ function makeClassifyExecutor(model, options, context) {
       temperature: 0,
       maxTokens: 256
     });
-    const result = await workflow.run();
+    const output = await runWithStreamChunks(workflow, onStreamChunk);
     return {
       expected: expectedLabel(row, options.labelColumn, context.labelNames, labels),
-      predicted: String(result.object?.label ?? "")
+      predicted: String(output.object?.label ?? ""),
+      usage: output[USAGE_OUTPUT_KEY]
     };
   };
 }
 
 // src/evals/extract.ts
-import { Workflow as Workflow2 } from "@workglow/task-graph";
+import { USAGE_OUTPUT_KEY as USAGE_OUTPUT_KEY2, Workflow as Workflow2 } from "@workglow/task-graph";
 var DEFAULT_INSTRUCTION = "Extract every entity mentioned in the text.";
 var UNSAFE_FIELDS = new Set(["__proto__", "constructor", "prototype"]);
 function parseExpectedRows(value, column) {
@@ -896,7 +1110,7 @@ function makeExtractExecutor(model, options, context) {
   const instruction = options.instruction ?? DEFAULT_INSTRUCTION;
   const explicitFields = options.fields && options.fields.length > 0 ? resolveExtractionFields([], options.keyField, options.fields) : undefined;
   const explicitSchema = explicitFields ? buildItemsSchema(explicitFields, options.keyField) : undefined;
-  return async (row) => {
+  return async (row, onStreamChunk) => {
     const text = String(row[options.textColumn] ?? "");
     const expectedRows = parseExpectedRows(row[options.expectedColumn], options.expectedColumn);
     const fields = explicitFields ?? resolveExtractionFields(expectedRows, options.keyField, undefined);
@@ -909,17 +1123,18 @@ function makeExtractExecutor(model, options, context) {
       temperature: 0,
       maxTokens: 2048
     });
-    const result = await workflow.run();
-    const items = Array.isArray(result.object?.items) ? result.object.items : [];
+    const output = await runWithStreamChunks(workflow, onStreamChunk);
+    const items = Array.isArray(output.object?.items) ? output.object.items : [];
     return {
       expected: JSON.stringify(expectedRows),
-      predicted: JSON.stringify(items)
+      predicted: JSON.stringify(items),
+      usage: output[USAGE_OUTPUT_KEY2]
     };
   };
 }
 
 // src/evals/similarity.ts
-import { Workflow as Workflow3 } from "@workglow/task-graph";
+import { USAGE_OUTPUT_KEY as USAGE_OUTPUT_KEY3, Workflow as Workflow3 } from "@workglow/task-graph";
 import { cosineSimilarity } from "@workglow/util/schema";
 function makeSimilarityExecutor(model, options, context) {
   for (const column of [options.textColumn, options.pairColumn, options.scoreColumn]) {
@@ -927,13 +1142,13 @@ function makeSimilarityExecutor(model, options, context) {
       throw new Error(`dataset has no column "${column}" (columns: ${context.columns.join(", ")}) \u2014 ` + `set --text-column/--pair-column/--score-column`);
     }
   }
-  return async (row) => {
+  return async (row, onStreamChunk) => {
     const a = String(row[options.textColumn] ?? "");
     const b = String(row[options.pairColumn] ?? "");
     const workflow = new Workflow3;
     workflow.textEmbedding({ model, text: [a, b] });
-    const result = await workflow.run();
-    const vectors = Array.isArray(result.vector) ? result.vector : [result.vector];
+    const output = await runWithStreamChunks(workflow, onStreamChunk);
+    const vectors = Array.isArray(output.vector) ? output.vector : [output.vector];
     if (vectors.length < 2) {
       throw new Error(`expected 2 embeddings, got ${vectors.length}`);
     }
@@ -943,12 +1158,25 @@ function makeSimilarityExecutor(model, options, context) {
     }
     return {
       expectedValue: gold,
-      predictedValue: cosineSimilarity(vectors[0], vectors[1])
+      predictedValue: cosineSimilarity(vectors[0], vectors[1]),
+      usage: output[USAGE_OUTPUT_KEY3]
     };
   };
 }
 
 // src/evals/runner.ts
+function usageColumns(usage, pricing) {
+  const estimate = usage ? estimateCost(usage, pricing) : undefined;
+  return {
+    input_tokens: usage?.input ?? null,
+    output_tokens: usage?.output ?? null,
+    cached_tokens: usage?.cached ?? null,
+    cache_write_tokens: usage?.cacheWrite ?? null,
+    total_tokens: usage?.total ?? null,
+    cost: estimate?.amount ?? null,
+    currency: estimate?.currency ?? null
+  };
+}
 var EXECUTORS = {
   classify: makeClassifyExecutor,
   similarity: makeSimilarityExecutor,
@@ -979,11 +1207,13 @@ async function runSweep(stores, rows, options) {
   for (const modelId of options.models) {
     let executor;
     let setupError;
+    let modelPricing;
     try {
       let config = resolveModelConfig(modelId, options.kind);
       if (options.kind === "similarity")
         config = await ensureEmbeddingDimensions(config);
       await ensureModelDownloaded(config);
+      modelPricing = config.pricing;
       executor = makeExecutor(options.kind, config, options.columns, options.context);
     } catch (err) {
       setupError = err instanceof Error ? err.message : String(err);
@@ -991,13 +1221,15 @@ async function runSweep(stores, rows, options) {
     for (const { record, row, parseError } of parsedRows) {
       const t0 = performance.now();
       let outcome;
+      let outcomeUsage;
       if (!executor) {
         outcome = failedOutcome(setupError ?? "executor unavailable");
       } else if (!row) {
         outcome = failedOutcome(parseError ?? "invalid stored row");
       } else {
         try {
-          const prediction = await executor(row);
+          const prediction = await executor(row, options.onStreamChunk);
+          outcomeUsage = prediction.usage;
           outcome = {
             ok: 1,
             error: null,
@@ -1016,10 +1248,11 @@ async function runSweep(stores, rows, options) {
         model: modelId,
         row_index: record.row_index,
         latency_ms: Math.round(latency * 100) / 100,
+        ...usageColumns(outcomeUsage, modelPricing),
         ...outcome
       });
       done++;
-      options.onProgress?.(done, total, modelId, outcome.ok === 1);
+      options.onProgress?.(done, total, modelId, outcome.ok === 1, outcomeUsage);
     }
   }
   return run.run_id;
@@ -1035,6 +1268,39 @@ function failedOutcome(error) {
   };
 }
 
+// src/report/liveTally.ts
+import { estimateCost as estimateCost2, formatCost, formatUsage } from "@workglow/ai";
+import { mergeUsage } from "@workglow/task-graph";
+
+class LiveTally {
+  pricing;
+  byModel = new Map;
+  startedAt;
+  constructor(pricing, now) {
+    this.pricing = pricing;
+    this.startedAt = now;
+  }
+  record(model, ok, usage) {
+    const tally = this.byModel.get(model) ?? { rows: 0, ok: 0, usage: undefined };
+    tally.rows += 1;
+    if (ok)
+      tally.ok += 1;
+    tally.usage = mergeUsage(tally.usage, usage);
+    this.byModel.set(model, tally);
+  }
+  render(done, total, now) {
+    const elapsedSec = Math.max((now - this.startedAt) / 1000, 0.001);
+    const lines = [`[${done}/${total}]`];
+    for (const [model, tally] of this.byModel) {
+      const tps = tally.usage?.output === undefined ? "" : ` ${Math.round(tally.usage.output / elapsedSec)} tok/s`;
+      const cost = tally.usage ? formatCost(estimateCost2(tally.usage, this.pricing(model))) : "";
+      lines.push(`  ${model}  ${tally.ok}/${tally.rows} ok  ` + `${formatUsage(tally.usage, "detailed")}${tps}${cost ? `  ${cost}` : ""}`);
+    }
+    return lines.join(`
+`);
+  }
+}
+
 // src/commands/run.ts
 var DEFAULT_MODELS = {
   classify: "onnx-community/LFM2.5-350M-ONNX",
@@ -1053,7 +1319,7 @@ var DESCRIPTIONS = {
 };
 function registerRunCommand(program, openStores, ensureProviders) {
   for (const kind of ["classify", "similarity", "extract"]) {
-    const command = program.command(`run-${kind}`).description(DESCRIPTIONS[kind]).requiredOption("--dataset <id>", "a dataset previously stored via `dataset pull`").option("--split <split>", "dataset split", "test").option("--models <list>", `comma-separated model ids (default: ${DEFAULT_MODELS[kind]})`).option("--limit <n>", "cap the number of stored rows to run").option("--text-column <name>", `input text column (default: ${DEFAULT_TEXT_COLUMN[kind]})`).option("--format <fmt>", "report output: table or json", "table");
+    const command = program.command(`run-${kind}`).description(DESCRIPTIONS[kind]).requiredOption("--dataset <id>", "a dataset previously stored via `dataset pull`").option("--split <split>", "dataset split", "test").option("--models <list>", `comma-separated model ids (default: ${DEFAULT_MODELS[kind]})`).option("--limit <n>", "cap the number of stored rows to run").option("--text-column <name>", `input text column (default: ${DEFAULT_TEXT_COLUMN[kind]})`).option("--format <fmt>", "report output: table or json", "table").option("--stream", "render in-flight generation text for the current row", false);
     if (kind === "classify") {
       command.option("--label-column <name>", "gold label column (default: label)").option("--labels <list>", "comma-separated candidate labels override");
     } else if (kind === "similarity") {
@@ -1113,6 +1379,11 @@ async function runEval(kind, flags, stores) {
     labelNames: meta.label_names ? JSON.parse(meta.label_names) : {}
   };
   console.error(`running ${kind} over ${limited.length} rows of ${flags.dataset} [${flags.split}] ` + `with ${models.length} model(s)`);
+  const tally = new LiveTally((id) => resolveModelConfig(id, kind).pricing, Date.now());
+  const onStreamChunk = flags.stream ? (event) => {
+    if (event.type === "text-delta")
+      process.stderr.write(event.textDelta);
+  } : undefined;
   const runId = await runSweep(stores, limited, {
     kind,
     dataset: flags.dataset,
@@ -1120,9 +1391,12 @@ async function runEval(kind, flags, stores) {
     models,
     columns,
     context,
-    onProgress: (done, total, model, ok) => {
-      console.error(`[${done}/${total}] ${model} ${ok ? "ok" : "FAIL"}`);
-    }
+    onProgress: (done, total, model, ok, usage) => {
+      tally.record(model, ok, usage);
+      process.stderr.write(`${tally.render(done, total, Date.now())}
+`);
+    },
+    onStreamChunk
   });
   console.error(`run ${runId} complete
 `);
@@ -1232,7 +1506,14 @@ var EvalResultSchema = {
     predicted: { type: ["string", "null"] },
     expected_value: { type: ["number", "null"] },
     predicted_value: { type: ["number", "null"] },
-    latency_ms: { type: "number" }
+    latency_ms: { type: "number" },
+    input_tokens: { type: ["integer", "null"] },
+    output_tokens: { type: ["integer", "null"] },
+    cached_tokens: { type: ["integer", "null"] },
+    cache_write_tokens: { type: ["integer", "null"] },
+    total_tokens: { type: ["integer", "null"] },
+    cost: { type: ["number", "null"] },
+    currency: { type: ["string", "null"] }
   },
   required: ["run_id", "model", "row_index", "ok", "latency_ms"],
   additionalProperties: false

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@workglow/eval",
   "type": "module",
-  "version": "0.3.38",
+  "version": "0.3.40",
   "repository": {
     "type": "git",
     "url": "https://github.com/workglow-dev/libs.git",
@@ -21,7 +21,7 @@
     "build-js": "bun build --target=bun --packages=external --outdir ./dist ./src/workglow-eval.ts",
     "build-js-worker-hft": "bun build --target=node --packages=external --outdir ./dist ./src/worker_hft.ts",
     "build-types": "rm -f tsconfig.tsbuildinfo && tsgo",
-    "test": "vitest run",
+    "test": "vitest run --config ../../vitest.config.ts --project eval",
     "test:watch": "vitest",
     "build-js-worker-llamacpp": "bun build --target=node --packages=external --outdir ./dist ./src/worker_llamacpp.ts"
   },
@@ -33,21 +33,21 @@
     "src/**/*.md"
   ],
   "dependencies": {
-    "@workglow/ai": "0.3.38",
-    "@workglow/anthropic": "0.3.38",
-    "@workglow/deepseek": "0.3.38",
-    "@workglow/google-gemini": "0.3.38",
-    "@workglow/huggingface-transformers": "0.3.38",
-    "@workglow/node-llama-cpp": "0.3.38",
-    "@workglow/openai": "0.3.38",
-    "@workglow/sqlite": "0.3.38",
-    "@workglow/storage": "0.3.38",
-    "@workglow/task-graph": "0.3.38",
-    "@workglow/util": "0.3.38",
-    "@workglow/xai": "0.3.38",
+    "@workglow/ai": "0.3.40",
+    "@workglow/anthropic": "0.3.40",
+    "@workglow/deepseek": "0.3.40",
+    "@workglow/google-gemini": "0.3.40",
+    "@workglow/huggingface-transformers": "0.3.40",
+    "@workglow/node-llama-cpp": "0.3.40",
+    "@workglow/openai": "0.3.40",
+    "@workglow/sqlite": "0.3.40",
+    "@workglow/storage": "0.3.40",
+    "@workglow/task-graph": "0.3.40",
+    "@workglow/util": "0.3.40",
+    "@workglow/xai": "0.3.40",
     "@huggingface/transformers": "^4.2.0",
     "commander": "^15.0.0",
-    "hyparquet": "^1.27.1",
+    "hyparquet": "^1.28.1",
     "hyparquet-compressors": "^1.1.1"
   },
   "publishConfig": {