@dalmia/calibrate-mcp 0.0.89 → 0.0.91

package/src/models/taskagreementannotationtaskstaskuuidagreementgetop.ts

@@ -3,6 +3,7 @@
  */
 
 import * as z from "zod";
+import { ClosedEnum } from "../types/enums.js";
 import {
   HTTPValidationError,
   HTTPValidationError$zodSchema,
@@ -12,14 +13,65 @@ import {
   TaskAgreementResponse$zodSchema,
 } from "./taskagreementresponse.js";
 
+/**
+ * Which date `start` and `end` apply to:
+ *
+ * @remarks
+ *
+ * - `row_added`: when the row was added to the task
+ * - `annotator_answered`: when the annotator first gave the answer in their latest labelling job. Editing does not change this date
+ *
+ * Omit to use `row_added`
+ */
+export const TaskAgreementAnnotationTasksTaskUuidAgreementGetDateField = {
+  RowAdded: "row_added",
+  AnnotatorAnswered: "annotator_answered",
+} as const;
+/**
+ * Which date `start` and `end` apply to:
+ *
+ * @remarks
+ *
+ * - `row_added`: when the row was added to the task
+ * - `annotator_answered`: when the annotator first gave the answer in their latest labelling job. Editing does not change this date
+ *
+ * Omit to use `row_added`
+ */
+export type TaskAgreementAnnotationTasksTaskUuidAgreementGetDateField =
+  ClosedEnum<typeof TaskAgreementAnnotationTasksTaskUuidAgreementGetDateField>;
+
+export const TaskAgreementAnnotationTasksTaskUuidAgreementGetDateField$zodSchema =
+  z.enum([
+    "row_added",
+    "annotator_answered",
+  ]).describe(
+    "Which date `start` and `end` apply to:\n\n- `row_added`: when the row was added to the task\n- `annotator_answered`: when the annotator first gave the answer in their latest labelling job. Editing does not change this date\n\nOmit to use `row_added`",
+  );
+
 export type TaskAgreementAnnotationTasksTaskUuidAgreementGetRequest = {
   task_uuid: string;
+  date_field?:
+    | TaskAgreementAnnotationTasksTaskUuidAgreementGetDateField
+    | undefined;
+  start?: string | null | undefined;
+  end?: string | null | undefined;
   xAPIKey?: string | null | undefined;
 };
 
 export const TaskAgreementAnnotationTasksTaskUuidAgreementGetRequest$zodSchema:
   z.ZodType<TaskAgreementAnnotationTasksTaskUuidAgreementGetRequest> = z.object(
     {
+      date_field:
+        TaskAgreementAnnotationTasksTaskUuidAgreementGetDateField$zodSchema
+          .default("row_added").describe(
+            "Which date `start` and `end` apply to:\n\n- `row_added`: when the row was added to the task\n- `annotator_answered`: when the annotator first gave the answer in their latest labelling job. Editing does not change this date\n\nOmit to use `row_added`",
+          ),
+      end: z.iso.datetime({ offset: true }).describe(
+        "Return only rows or answers dated before this time (ISO 8601). A time with no timezone is read as UTC. Omit for no latest date",
+      ).nullable().optional(),
+      start: z.iso.datetime({ offset: true }).describe(
+        "Return only rows or answers dated at or after this time (ISO 8601). A time with no timezone is read as UTC. Omit for no earliest date",
+      ).nullable().optional(),
       task_uuid: z.string().describe("Annotation task to act on"),
       xAPIKey: z.string().nullable().optional(),
     },

package/src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts

@@ -13,6 +13,42 @@ import {
   TaskSummaryResponse$zodSchema,
 } from "./tasksummaryresponse.js";
 
+/**
+ * Which date `start` and `end` apply to:
+ *
+ * @remarks
+ *
+ * - `row_added`: when the row was added to the task
+ * - `annotator_answered`: when the annotator first gave the answer in their latest labelling job. Editing does not change this date
+ *
+ * Omit to use `row_added`
+ */
+export const TaskSummaryAnnotationTasksTaskUuidSummaryGetDateField = {
+  RowAdded: "row_added",
+  AnnotatorAnswered: "annotator_answered",
+} as const;
+/**
+ * Which date `start` and `end` apply to:
+ *
+ * @remarks
+ *
+ * - `row_added`: when the row was added to the task
+ * - `annotator_answered`: when the annotator first gave the answer in their latest labelling job. Editing does not change this date
+ *
+ * Omit to use `row_added`
+ */
+export type TaskSummaryAnnotationTasksTaskUuidSummaryGetDateField = ClosedEnum<
+  typeof TaskSummaryAnnotationTasksTaskUuidSummaryGetDateField
+>;
+
+export const TaskSummaryAnnotationTasksTaskUuidSummaryGetDateField$zodSchema = z
+  .enum([
+    "row_added",
+    "annotator_answered",
+  ]).describe(
+    "Which date `start` and `end` apply to:\n\n- `row_added`: when the row was added to the task\n- `annotator_answered`: when the annotator first gave the answer in their latest labelling job. Editing does not change this date\n\nOmit to use `row_added`",
+  );
+
 /**
  * Sort direction
  */
@@ -35,6 +71,11 @@ export type TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest = {
   item_id?: string | null | undefined;
   live_only?: boolean | undefined;
   disagreement_only?: boolean | undefined;
+  date_field?:
+    | TaskSummaryAnnotationTasksTaskUuidSummaryGetDateField
+    | undefined;
+  start?: string | null | undefined;
+  end?: string | null | undefined;
   q?: string | null | undefined;
   sort_by?: string | undefined;
   order?: Order | undefined;
@@ -49,9 +90,16 @@ export const TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest$zodSchema:
     compact: z.boolean().default(false).describe(
       "Return a compact response that omits heavy detail fields (`rows.payload`, `rows.evaluator_reasoning`, `rows.annotations.reasoning`, `evaluators.versions.system_prompt`, `evaluators.versions.output_config`, `evaluators.versions.variables`, `item_comments`), keeping only the lightweight decision fields. Omit for full detail",
     ),
+    date_field: TaskSummaryAnnotationTasksTaskUuidSummaryGetDateField$zodSchema
+      .default("row_added").describe(
+        "Which date `start` and `end` apply to:\n\n- `row_added`: when the row was added to the task\n- `annotator_answered`: when the annotator first gave the answer in their latest labelling job. Editing does not change this date\n\nOmit to use `row_added`",
+      ),
     disagreement_only: z.boolean().default(false).describe(
       "When true, keep only rows where the evaluator disagreed with at least one annotator",
     ),
+    end: z.iso.datetime({ offset: true }).describe(
+      "Return only rows or answers dated before this time (ISO 8601). A time with no timezone is read as UTC. Omit for no latest date",
+    ).nullable().optional(),
     item_id: z.string().describe(
       "Filter rows to a single item. The full task-wide annotator union is still returned in `annotators`",
     ).nullable().optional(),
@@ -69,6 +117,9 @@ export const TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest$zodSchema:
     sort_by: z.string().default("created_at").describe(
       "Sort key for the results",
     ),
+    start: z.iso.datetime({ offset: true }).describe(
+      "Return only rows or answers dated at or after this time (ISO 8601). A time with no timezone is read as UTC. Omit for no earliest date",
+    ).nullable().optional(),
     task_uuid: z.string().describe("Annotation task to act on"),
     xAPIKey: z.string().nullable().optional(),
   });

package/src/models/tasksummaryresponse.ts

@@ -54,7 +54,7 @@ export const TaskSummaryResponse$zodSchema: z.ZodType<TaskSummaryResponse> =
         .describe("Annotator comments, keyed by item ID"),
       pagination: PaginationMeta$zodSchema,
       rows: z.array(z.record(z.string(), z.any())).describe(
-        "One row per item, evaluator, and version on this page, with the evaluator's value and each annotator's label",
+        "One row per item, evaluator, and version on this page, with the evaluator's value and each annotator's label. `item_added_at` is when the row was added (ISO 8601 UTC). Each label's `answered_at` is when the annotator first gave it in their latest labelling job, and editing does not change it",
       ),
       task_id: z.string().describe("ID of the task"),
       task_type: TaskType$zodSchema.describe("Type of the task"),

package/src/models/tracelistresponse.ts

@@ -29,7 +29,7 @@ export const TraceListResponse$zodSchema: z.ZodType<TraceListResponse> = z
     offset: z.int().default(0).describe("The applied `offset`"),
     score_averages: z.array(TraceScoreAverage$zodSchema).nullable().optional()
       .describe(
-        "One entry per evaluator that scored at least one matching trace, averaged over every matching trace rather than this page. Ask for it with `include_score_averages`",
+        "One average for each evaluator linked to the agent that scored at least one trace matching the filters, covering all the matching traces it scored rather than this page. Ask for it with `include_score_averages`",
       ),
     total: z.int().describe(
       "Total number of items matching the query, before pagination",

package/src/models/tracesummary.ts

@@ -94,7 +94,7 @@ export const TraceSummary$zodSchema: z.ZodType<TraceSummary> = z.object({
     "The agent reply, truncated for display",
   ),
   results: z.array(TraceScoreResult$zodSchema).optional().describe(
-    "Results for each evaluator on the latest scoring run",
+    "Results on the latest scoring run from the evaluators linked to the agent",
   ),
   tool_call_count: z.int().describe(
     "Number of tool calls the agent issued for this turn",