@atlanai/sdk 0.2.1 → 0.2.2

package/dist/raw/eval/apis/ExperimentsApi.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *
@@ -59,6 +59,7 @@ export interface EvalListExperimentResultsRequest {
     experimentId: string;
     limit?: number;
     offset?: number;
+    caseId?: string;
     datasetRecordId?: string;
     sessionId?: string;
     sort?: string;
@@ -103,9 +104,6 @@ export interface EvalSearchExperimentsRequest {
     evalArtifactSearchRequest: EvalArtifactSearchRequest;
     contentEncoding?: EvalSearchExperimentsContentEncodingEnum;
 }
-export interface EvalSummarizeExperimentRequest {
-    experimentId: string;
-}
 /**
  *
  */
@@ -129,12 +127,12 @@ export declare class ExperimentsApi extends runtime.BaseAPI {
      */
     evalCreateExperimentRequestOpts(requestParameters: EvalCreateExperimentRequest): Promise<runtime.RequestOpts>;
     /**
-     * Record the start of an evaluation run: the subject under test (an agent, harness, or skill), the dataset, the configuration, and optionally the baseline it is compared against. The gateway does not execute anything — your runner or CI executes the cases and uploads results here. Created `running` by default; upload results, then finalize with one PATCH to `completed` carrying the score summary.
+     * Record the start of an evaluation run: the subject under test (an agent, harness, or skill), the dataset, the configuration, and optionally the baseline it is compared against. The gateway does not execute anything — your runner or CI executes the cases and uploads results here. Created `running` by default; upload results, then finalize with one PATCH to `completed`. The finalization response contains the server-derived score summary.
      * Create an experiment
      */
     evalCreateExperimentRaw(requestParameters: EvalCreateExperimentRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<runtime.ApiResponse<EvalExperimentArtifactView>>;
     /**
-     * Record the start of an evaluation run: the subject under test (an agent, harness, or skill), the dataset, the configuration, and optionally the baseline it is compared against. The gateway does not execute anything — your runner or CI executes the cases and uploads results here. Created `running` by default; upload results, then finalize with one PATCH to `completed` carrying the score summary.
+     * Record the start of an evaluation run: the subject under test (an agent, harness, or skill), the dataset, the configuration, and optionally the baseline it is compared against. The gateway does not execute anything — your runner or CI executes the cases and uploads results here. Created `running` by default; upload results, then finalize with one PATCH to `completed`. The finalization response contains the server-derived score summary.
      * Create an experiment
      */
     evalCreateExperiment(requestParameters: EvalCreateExperimentRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<EvalExperimentArtifactView>;
@@ -185,12 +183,12 @@ export declare class ExperimentsApi extends runtime.BaseAPI {
      */
     evalGetExperimentResultRequestOpts(requestParameters: EvalGetExperimentResultRequest): Promise<runtime.RequestOpts>;
     /**
-     * Read one evaluated case with its full typed metadata — the input/expected snapshot, the output, the scores, and the session/trace join keys. `experiment_id` addresses the result within its experiment; the result\'s own handle is what resolves it, and the path\'s experiment is **not** currently checked against the result\'s own — see the open item in eval.md.
+     * Read one evaluated case with its full typed metadata — the input/expected snapshot, output, stable case identity, and session/trace join keys. The result must belong to the experiment named in the path; a mismatched parent is returned as 404.
      * Get one result
      */
     evalGetExperimentResultRaw(requestParameters: EvalGetExperimentResultRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<runtime.ApiResponse<EvalExperimentResultArtifactView>>;
     /**
-     * Read one evaluated case with its full typed metadata — the input/expected snapshot, the output, the scores, and the session/trace join keys. `experiment_id` addresses the result within its experiment; the result\'s own handle is what resolves it, and the path\'s experiment is **not** currently checked against the result\'s own — see the open item in eval.md.
+     * Read one evaluated case with its full typed metadata — the input/expected snapshot, output, stable case identity, and session/trace join keys. The result must belong to the experiment named in the path; a mismatched parent is returned as 404.
      * Get one result
      */
     evalGetExperimentResult(requestParameters: EvalGetExperimentResultRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<EvalExperimentResultArtifactView>;
@@ -269,12 +267,12 @@ export declare class ExperimentsApi extends runtime.BaseAPI {
      */
     evalPatchExperimentRequestOpts(requestParameters: EvalPatchExperimentRequest): Promise<runtime.RequestOpts>;
     /**
-     * The finalize write: move `experiment_status` to `completed` or `failed` and record the score `summary`. The run-defining fields — dataset, subject, config, baseline — are not patchable; they describe a run that already happened. A terminal experiment is frozen entirely.
+     * The finalize write: move `experiment_status` to `completed` or `failed`. Runner-observed facts may accompany the transition in `summary`; the server replaces `summary.scores` with aggregates derived from the experiment\'s immutable result rows. The run-defining fields — dataset, subject, config, baseline — are not patchable; they describe a run that already happened. A terminal experiment is frozen entirely.
      * Finalize or annotate an experiment
      */
     evalPatchExperimentRaw(requestParameters: EvalPatchExperimentRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<runtime.ApiResponse<EvalExperimentArtifactView>>;
     /**
-     * The finalize write: move `experiment_status` to `completed` or `failed` and record the score `summary`. The run-defining fields — dataset, subject, config, baseline — are not patchable; they describe a run that already happened. A terminal experiment is frozen entirely.
+     * The finalize write: move `experiment_status` to `completed` or `failed`. Runner-observed facts may accompany the transition in `summary`; the server replaces `summary.scores` with aggregates derived from the experiment\'s immutable result rows. The run-defining fields — dataset, subject, config, baseline — are not patchable; they describe a run that already happened. A terminal experiment is frozen entirely.
      * Finalize or annotate an experiment
      */
     evalPatchExperiment(requestParameters: EvalPatchExperimentRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<EvalExperimentArtifactView>;
@@ -306,20 +304,6 @@ export declare class ExperimentsApi extends runtime.BaseAPI {
      * Search experiments
      */
     evalSearchExperiments(requestParameters: EvalSearchExperimentsRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<EvalPageExperimentArtifactView>;
-    /**
-     * Creates request options for evalSummarizeExperiment without sending the request
-     */
-    evalSummarizeExperimentRequestOpts(requestParameters: EvalSummarizeExperimentRequest): Promise<runtime.RequestOpts>;
-    /**
-     * Use this operation to roll the run\'s score spans up into `summary`.
-     * Roll the run\'s score spans up into `summary`
-     */
-    evalSummarizeExperimentRaw(requestParameters: EvalSummarizeExperimentRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<runtime.ApiResponse<EvalExperimentArtifactView>>;
-    /**
-     * Use this operation to roll the run\'s score spans up into `summary`.
-     * Roll the run\'s score spans up into `summary`
-     */
-    evalSummarizeExperiment(requestParameters: EvalSummarizeExperimentRequest, initOverrides?: RequestInit | runtime.InitOverrideFunction): Promise<EvalExperimentArtifactView>;
 }
 /**
   * @export

package/dist/raw/eval/apis/ExperimentsApi.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *
@@ -139,7 +139,7 @@ class ExperimentsApi extends runtime.BaseAPI {
         };
     }
     /**
-     * Record the start of an evaluation run: the subject under test (an agent, harness, or skill), the dataset, the configuration, and optionally the baseline it is compared against. The gateway does not execute anything — your runner or CI executes the cases and uploads results here. Created `running` by default; upload results, then finalize with one PATCH to `completed` carrying the score summary.
+     * Record the start of an evaluation run: the subject under test (an agent, harness, or skill), the dataset, the configuration, and optionally the baseline it is compared against. The gateway does not execute anything — your runner or CI executes the cases and uploads results here. Created `running` by default; upload results, then finalize with one PATCH to `completed`. The finalization response contains the server-derived score summary.
      * Create an experiment
      */
     async evalCreateExperimentRaw(requestParameters, initOverrides) {
@@ -148,7 +148,7 @@ class ExperimentsApi extends runtime.BaseAPI {
         return new runtime.JSONApiResponse(response, (jsonValue) => (0, EvalExperimentArtifactView_1.EvalExperimentArtifactViewFromJSON)(jsonValue));
     }
     /**
-     * Record the start of an evaluation run: the subject under test (an agent, harness, or skill), the dataset, the configuration, and optionally the baseline it is compared against. The gateway does not execute anything — your runner or CI executes the cases and uploads results here. Created `running` by default; upload results, then finalize with one PATCH to `completed` carrying the score summary.
+     * Record the start of an evaluation run: the subject under test (an agent, harness, or skill), the dataset, the configuration, and optionally the baseline it is compared against. The gateway does not execute anything — your runner or CI executes the cases and uploads results here. Created `running` by default; upload results, then finalize with one PATCH to `completed`. The finalization response contains the server-derived score summary.
      * Create an experiment
      */
     async evalCreateExperiment(requestParameters, initOverrides) {
@@ -327,7 +327,7 @@ class ExperimentsApi extends runtime.BaseAPI {
         };
     }
     /**
-     * Read one evaluated case with its full typed metadata — the input/expected snapshot, the output, the scores, and the session/trace join keys. `experiment_id` addresses the result within its experiment; the result\'s own handle is what resolves it, and the path\'s experiment is **not** currently checked against the result\'s own — see the open item in eval.md.
+     * Read one evaluated case with its full typed metadata — the input/expected snapshot, output, stable case identity, and session/trace join keys. The result must belong to the experiment named in the path; a mismatched parent is returned as 404.
      * Get one result
      */
     async evalGetExperimentResultRaw(requestParameters, initOverrides) {
@@ -336,7 +336,7 @@ class ExperimentsApi extends runtime.BaseAPI {
         return new runtime.JSONApiResponse(response, (jsonValue) => (0, EvalExperimentResultArtifactView_1.EvalExperimentResultArtifactViewFromJSON)(jsonValue));
     }
     /**
-     * Read one evaluated case with its full typed metadata — the input/expected snapshot, the output, the scores, and the session/trace join keys. `experiment_id` addresses the result within its experiment; the result\'s own handle is what resolves it, and the path\'s experiment is **not** currently checked against the result\'s own — see the open item in eval.md.
+     * Read one evaluated case with its full typed metadata — the input/expected snapshot, output, stable case identity, and session/trace join keys. The result must belong to the experiment named in the path; a mismatched parent is returned as 404.
      * Get one result
      */
     async evalGetExperimentResult(requestParameters, initOverrides) {
@@ -403,6 +403,9 @@ class ExperimentsApi extends runtime.BaseAPI {
         if (requestParameters['offset'] != null) {
             queryParameters['offset'] = requestParameters['offset'];
         }
+        if (requestParameters['caseId'] != null) {
+            queryParameters['case_id'] = requestParameters['caseId'];
+        }
         if (requestParameters['datasetRecordId'] != null) {
             queryParameters['dataset_record_id'] = requestParameters['datasetRecordId'];
         }
@@ -657,7 +660,7 @@ class ExperimentsApi extends runtime.BaseAPI {
         };
     }
     /**
-     * The finalize write: move `experiment_status` to `completed` or `failed` and record the score `summary`. The run-defining fields — dataset, subject, config, baseline — are not patchable; they describe a run that already happened. A terminal experiment is frozen entirely.
+     * The finalize write: move `experiment_status` to `completed` or `failed`. Runner-observed facts may accompany the transition in `summary`; the server replaces `summary.scores` with aggregates derived from the experiment\'s immutable result rows. The run-defining fields — dataset, subject, config, baseline — are not patchable; they describe a run that already happened. A terminal experiment is frozen entirely.
      * Finalize or annotate an experiment
      */
     async evalPatchExperimentRaw(requestParameters, initOverrides) {
@@ -666,7 +669,7 @@ class ExperimentsApi extends runtime.BaseAPI {
         return new runtime.JSONApiResponse(response, (jsonValue) => (0, EvalExperimentArtifactView_1.EvalExperimentArtifactViewFromJSON)(jsonValue));
     }
     /**
-     * The finalize write: move `experiment_status` to `completed` or `failed` and record the score `summary`. The run-defining fields — dataset, subject, config, baseline — are not patchable; they describe a run that already happened. A terminal experiment is frozen entirely.
+     * The finalize write: move `experiment_status` to `completed` or `failed`. Runner-observed facts may accompany the transition in `summary`; the server replaces `summary.scores` with aggregates derived from the experiment\'s immutable result rows. The run-defining fields — dataset, subject, config, baseline — are not patchable; they describe a run that already happened. A terminal experiment is frozen entirely.
      * Finalize or annotate an experiment
      */
     async evalPatchExperiment(requestParameters, initOverrides) {
@@ -769,48 +772,6 @@ class ExperimentsApi extends runtime.BaseAPI {
         const response = await this.evalSearchExperimentsRaw(requestParameters, initOverrides);
         return await response.value();
     }
-    /**
-     * Creates request options for evalSummarizeExperiment without sending the request
-     */
-    async evalSummarizeExperimentRequestOpts(requestParameters) {
-        if (requestParameters['experimentId'] == null) {
-            throw new runtime.RequiredError('experimentId', 'Required parameter "experimentId" was null or undefined when calling evalSummarizeExperiment().');
-        }
-        const queryParameters = {};
-        const headerParameters = {};
-        if (this.configuration && this.configuration.accessToken) {
-            const token = this.configuration.accessToken;
-            const tokenString = await token("bearer", []);
-            if (tokenString) {
-                headerParameters["Authorization"] = `Bearer ${tokenString}`;
-            }
-        }
-        let urlPath = `/experiments/{experiment_id}/summarize`;
-        urlPath = urlPath.replace('{experiment_id}', encodeURIComponent(String(requestParameters['experimentId'])));
-        return {
-            path: urlPath,
-            method: 'POST',
-            headers: headerParameters,
-            query: queryParameters,
-        };
-    }
-    /**
-     * Use this operation to roll the run\'s score spans up into `summary`.
-     * Roll the run\'s score spans up into `summary`
-     */
-    async evalSummarizeExperimentRaw(requestParameters, initOverrides) {
-        const requestOptions = await this.evalSummarizeExperimentRequestOpts(requestParameters);
-        const response = await this.request(requestOptions, initOverrides);
-        return new runtime.JSONApiResponse(response, (jsonValue) => (0, EvalExperimentArtifactView_1.EvalExperimentArtifactViewFromJSON)(jsonValue));
-    }
-    /**
-     * Use this operation to roll the run\'s score spans up into `summary`.
-     * Roll the run\'s score spans up into `summary`
-     */
-    async evalSummarizeExperiment(requestParameters, initOverrides) {
-        const response = await this.evalSummarizeExperimentRaw(requestParameters, initOverrides);
-        return await response.value();
-    }
 }
 exports.ExperimentsApi = ExperimentsApi;
 /**

package/dist/raw/eval/apis/ScorersApi.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/apis/ScorersApi.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/AccessView.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/AccessView.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/ArtifactConflictProblem.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/ArtifactConflictProblem.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/CostView.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/CostView.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/CursorPage.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/CursorPage.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalArtifactEnvelope.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalArtifactEnvelope.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalArtifactSearchRequest.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalArtifactSearchRequest.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalBulkDatasetRecordResponse.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalBulkDatasetRecordResponse.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalBulkDatasetRecordResult.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalBulkDatasetRecordResult.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalBulkExperimentResultResponse.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalBulkExperimentResultResponse.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalBulkExperimentResultSlot.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalBulkExperimentResultSlot.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset400Response.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset400Response.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset409Response.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset409Response.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset413Response.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset413Response.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset415Response.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset415Response.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset422Response.d.ts

@@ -1,6 +1,6 @@
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *

package/dist/raw/eval/models/EvalCreateDataset422Response.js

@@ -3,7 +3,7 @@
 /* eslint-disable */
 /**
  * Evaluation catalog
- * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. A verdict is a score span in the run\'s trace; `POST /experiments/{id}/summarize` rolls a run\'s score spans into its durable `summary`. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
+ * The evaluation loop as registry artifacts: curated `dataset`s of examples pinned from production sessions and traces, `experiment` run snapshots with create-only per-case `experiment_result`s, and versioned `scorer` definitions. Finalizing an experiment derives its durable score `summary` from the uploaded result rows. The gateway executes and scores nothing — external harnesses, CI, and judges run the evals, emit score spans, and upload results here.
  *
  * The version of the OpenAPI document: 0.1.0
  *