@atlanai/sdk 0.2.1 → 0.2.3

package/dist/raw/eval/models/EvalExperimentArtifactView.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
  *
@@ -22,6 +22,7 @@ const GatewayArtifactIdentityRevealView_1 = require("./GatewayArtifactIdentityRe
 const EvalSubjectKind_1 = require("./EvalSubjectKind");
 const EvalExperimentStatus_1 = require("./EvalExperimentStatus");
 const RegistryRelationshipsView_1 = require("./RegistryRelationshipsView");
+const EvalDatasetSnapshot_1 = require("./EvalDatasetSnapshot");
 /**
  * Check if a given object implements the EvalExperimentArtifactView interface.
  */
@@ -62,6 +63,7 @@ function EvalExperimentArtifactViewFromJSONTyped(json, ignoreDiscriminator) {
         return json;
     }
     return {
+        'app': json['app'] == null ? undefined : json['app'],
         'artifactStatus': json['artifact_status'],
         'commitTs': json['commit_ts'],
         'contentHashHex': json['content_hash_hex'] == null ? undefined : json['content_hash_hex'],
@@ -69,7 +71,6 @@ function EvalExperimentArtifactViewFromJSONTyped(json, ignoreDiscriminator) {
         'createdBy': json['created_by'],
         'description': json['description'] == null ? undefined : json['description'],
         'displayName': json['display_name'] == null ? undefined : json['display_name'],
-        'extension': json['extension'] == null ? undefined : json['extension'],
         'id': json['id'],
         'identity': json['identity'] == null ? undefined : (0, GatewayArtifactIdentityRevealView_1.GatewayArtifactIdentityRevealViewFromJSON)(json['identity']),
         'isLatest': json['is_latest'],
@@ -98,6 +99,7 @@ function EvalExperimentArtifactViewFromJSONTyped(json, ignoreDiscriminator) {
         'baselineExperimentId': json['baseline_experiment_id'] === undefined ? undefined : json['baseline_experiment_id'] === null ? null : json['baseline_experiment_id'],
         'config': json['config'] === undefined ? undefined : json['config'] === null ? null : json['config'],
         'datasetId': json['dataset_id'] === undefined ? undefined : json['dataset_id'] === null ? null : json['dataset_id'],
+        'datasetSnapshot': json['dataset_snapshot'] === undefined ? undefined : json['dataset_snapshot'] === null ? null : (0, EvalDatasetSnapshot_1.EvalDatasetSnapshotFromJSON)(json['dataset_snapshot']),
         'experimentStatus': json['experiment_status'] == null ? undefined : (0, EvalExperimentStatus_1.EvalExperimentStatusFromJSON)(json['experiment_status']),
         'subjectId': json['subject_id'] === undefined ? undefined : json['subject_id'] === null ? null : json['subject_id'],
         'subjectKind': json['subject_kind'] === undefined ? undefined : json['subject_kind'] === null ? null : (0, EvalSubjectKind_1.EvalSubjectKindFromJSON)(json['subject_kind']),
@@ -114,6 +116,7 @@ function EvalExperimentArtifactViewToJSONTyped(value, ignoreDiscriminator = fals
         return value;
     }
     return {
+        'app': value['app'],
         'artifact_status': value['artifactStatus'],
         'commit_ts': value['commitTs'],
         'content_hash_hex': value['contentHashHex'],
@@ -121,7 +124,6 @@ function EvalExperimentArtifactViewToJSONTyped(value, ignoreDiscriminator = fals
         'created_by': value['createdBy'],
         'description': value['description'],
         'display_name': value['displayName'],
-        'extension': value['extension'],
         'id': value['id'],
         'identity': (0, GatewayArtifactIdentityRevealView_1.GatewayArtifactIdentityRevealViewToJSON)(value['identity']),
         'is_latest': value['isLatest'],
@@ -150,6 +152,7 @@ function EvalExperimentArtifactViewToJSONTyped(value, ignoreDiscriminator = fals
         'baseline_experiment_id': value['baselineExperimentId'],
         'config': value['config'],
         'dataset_id': value['datasetId'],
+        'dataset_snapshot': (0, EvalDatasetSnapshot_1.EvalDatasetSnapshotToJSON)(value['datasetSnapshot']),
         'experiment_status': (0, EvalExperimentStatus_1.EvalExperimentStatusToJSON)(value['experimentStatus']),
         'subject_id': value['subjectId'],
         'subject_kind': (0, EvalSubjectKind_1.EvalSubjectKindToJSON)(value['subjectKind']),

package/dist/raw/eval/models/EvalExperimentCreate.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/EvalExperimentCreate.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/EvalExperimentMeta.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
  *
@@ -11,6 +11,7 @@
  */
 import type { EvalSubjectKind } from './EvalSubjectKind';
 import type { EvalExperimentStatus } from './EvalExperimentStatus';
+import type { EvalDatasetSnapshot } from './EvalDatasetSnapshot';
 /**
  * The `experiment` artifact kind: the recorded snapshot of one eval run.
  *
@@ -20,9 +21,8 @@ import type { EvalExperimentStatus } from './EvalExperimentStatus';
  * baseline — is pinned after create, and a terminal experiment is frozen
  * entirely: that is what makes a completed experiment comparable over time.
  *
- * `dataset_id` is a line pointer, deliberately **not** version-pinned: results
- * snapshot their own `input`/`expected`, so the experiment stays an honest
- * record even as the dataset evolves — and the dataset stays archivable.
+ * When `dataset_id` is present, `dataset_snapshot` is stamped by the server at
+ * create time and freezes the exact dataset and record versions used.
  * @export
  * @interface EvalExperimentMeta
  */
@@ -52,6 +52,12 @@ export interface EvalExperimentMeta {
      * @memberof EvalExperimentMeta
      */
     datasetId?: string | null;
+    /**
+     * Value of the `dataset_snapshot` field.
+     * @type {EvalDatasetSnapshot}
+     * @memberof EvalExperimentMeta
+     */
+    datasetSnapshot?: EvalDatasetSnapshot | null;
     /**
      *
      * @type {EvalExperimentStatus}

package/dist/raw/eval/models/EvalExperimentMeta.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
  *
@@ -20,6 +20,7 @@ exports.EvalExperimentMetaToJSON = EvalExperimentMetaToJSON;
 exports.EvalExperimentMetaToJSONTyped = EvalExperimentMetaToJSONTyped;
 const EvalSubjectKind_1 = require("./EvalSubjectKind");
 const EvalExperimentStatus_1 = require("./EvalExperimentStatus");
+const EvalDatasetSnapshot_1 = require("./EvalDatasetSnapshot");
 /**
  * Check if a given object implements the EvalExperimentMeta interface.
  */
@@ -37,6 +38,7 @@ function EvalExperimentMetaFromJSONTyped(json, ignoreDiscriminator) {
         'baselineExperimentId': json['baseline_experiment_id'] === undefined ? undefined : json['baseline_experiment_id'] === null ? null : json['baseline_experiment_id'],
         'config': json['config'] === undefined ? undefined : json['config'] === null ? null : json['config'],
         'datasetId': json['dataset_id'] === undefined ? undefined : json['dataset_id'] === null ? null : json['dataset_id'],
+        'datasetSnapshot': json['dataset_snapshot'] === undefined ? undefined : json['dataset_snapshot'] === null ? null : (0, EvalDatasetSnapshot_1.EvalDatasetSnapshotFromJSON)(json['dataset_snapshot']),
         'experimentStatus': json['experiment_status'] == null ? undefined : (0, EvalExperimentStatus_1.EvalExperimentStatusFromJSON)(json['experiment_status']),
         'subjectId': json['subject_id'] === undefined ? undefined : json['subject_id'] === null ? null : json['subject_id'],
         'subjectKind': json['subject_kind'] === undefined ? undefined : json['subject_kind'] === null ? null : (0, EvalSubjectKind_1.EvalSubjectKindFromJSON)(json['subject_kind']),
@@ -54,6 +56,7 @@ function EvalExperimentMetaToJSONTyped(value, ignoreDiscriminator = false) {
         'baseline_experiment_id': value['baselineExperimentId'],
         'config': value['config'],
         'dataset_id': value['datasetId'],
+        'dataset_snapshot': (0, EvalDatasetSnapshot_1.EvalDatasetSnapshotToJSON)(value['datasetSnapshot']),
         'experiment_status': (0, EvalExperimentStatus_1.EvalExperimentStatusToJSON)(value['experimentStatus']),
         'subject_id': value['subjectId'],
         'subject_kind': (0, EvalSubjectKind_1.EvalSubjectKindToJSON)(value['subjectKind']),

package/dist/raw/eval/models/EvalExperimentPatch.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
  *
@@ -44,8 +44,9 @@ export interface EvalExperimentPatch {
      */
     name?: string | null;
     /**
-     * Set or clear the score summary. Writable while the run is live; sealed
-     * once terminal.
+     * Set or clear runner-observed summary facts. On transition to a terminal
+     * status, the server replaces `summary.scores` with aggregates derived
+     * from the experiment's immutable result rows.
      * @type {{ [key: string]: any; }}
      * @memberof EvalExperimentPatch
      */

package/dist/raw/eval/models/EvalExperimentPatch.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/EvalExperimentResultArtifactView.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
  *
@@ -17,6 +17,12 @@ import type { RegistryRelationshipsView } from './RegistryRelationshipsView';
  * @interface EvalExperimentResultArtifactView
  */
 export interface EvalExperimentResultArtifactView {
+    /**
+     * Value of the `app` field.
+     * @type {string}
+     * @memberof EvalExperimentResultArtifactView
+     */
+    app?: string;
     /**
      * Value of the `artifact_status` field.
      * @type {string}
@@ -59,12 +65,6 @@ export interface EvalExperimentResultArtifactView {
      * @memberof EvalExperimentResultArtifactView
      */
     displayName?: string;
-    /**
-     * Value of the `extension` field.
-     * @type {string}
-     * @memberof EvalExperimentResultArtifactView
-     */
-    extension?: string;
     /**
      * Kind-prefixed Crockford-base32 handle (`<kind>_<base32>`).
      * @type {string}
@@ -103,19 +103,28 @@ export interface EvalExperimentResultArtifactView {
      * The Control-Plane OIDC client id this artifact authenticates as
      * (`x-gateway-identity` kinds); absent for every other kind.
      *
-     * **Derived, not stored.** The registry resolves it from the artifact's
-     * `service_principals` row rather than keeping a second copy on the artifact,
-     * so it appears on the create response and on the single-artifact read — the
-     * two paths that make that lookup — and **not** on list, search, or a `PATCH`
-     * response, none of which do a per-row identity read. It is the public half of
-     * the credential pair and is not secret.
+     * **Derived, not stored, and present on the create response only.** The
+     * registry would resolve it from the artifact's `service_principals` row
+     * rather than keep a second copy on the artifact — but only the create path
+     * does that lookup, because it has just provisioned the identity. Every read
+     * shape (single read, list, search, `PATCH`) is built by one constructor that
+     * sets this `None`, so it is absent from all four.
+     *
+     * This block used to name the single-artifact read as a second populated path.
+     * It was never one. Adding it means deciding whether list and search pay a
+     * per-row identity read too; the lookup a future implementation would call is
+     * `registry_server::identity::artifact_principal`. It is the public half of the
+     * credential pair and is not secret.
      * @type {string}
      * @memberof EvalExperimentResultArtifactView
      */
     oidcClientId?: string;
     /**
      * Canonical owner: the `users.user_id` this artifact belongs to, FK-backed.
-     * Absent when unset (legacy rows / system-plane creates).
+     *
+     * Defaulted to the creating principal, so this is normally set. Absent for a row created
+     * before that default landed, for a system-plane write, and for a principal with no directory
+     * row — a service account or a seed has nothing to own anything with.
      * @type {string}
      * @memberof EvalExperimentResultArtifactView
      */
@@ -223,13 +232,16 @@ export interface EvalExperimentResultArtifactView {
      */
     updatedBy: string;
     /**
-     * Optional human version selector (semver); absent when unset.
+     * The free-form label the write set, verbatim. Absent when none was sent, and *not* the name
+     * the version shipped under — that is the release's own `version_label`.
      * @type {string}
      * @memberof EvalExperimentResultArtifactView
      */
     versionLabel?: string;
     /**
-     * Monotonic version coordinate within the artifact line.
+     * This version's position in its line — the write index, 1-based and dense over versions
+     * *written*. Distinct from a release's `version_ordinal`, which counts versions **shipped**;
+     * the two agree under the default `auto` release policy and diverge under `manual`.
      * @type {number}
      * @memberof EvalExperimentResultArtifactView
      */
@@ -240,6 +252,13 @@ export interface EvalExperimentResultArtifactView {
      * @memberof EvalExperimentResultArtifactView
      */
     workspaceId: string;
+    /**
+     * Stable runner-owned identity for this case within the experiment.
+     * Unique per experiment, enabling safe retry and server-backed resume.
+     * @type {string}
+     * @memberof EvalExperimentResultArtifactView
+     */
+    caseId?: string | null;
     /**
      * The dataset row this case came from; absent for an ad-hoc case. The
      * per-case join key when comparing two experiments over one dataset.
@@ -290,6 +309,16 @@ export interface EvalExperimentResultArtifactView {
     output?: {
         [key: string]: any;
     } | null;
+    /**
+     * Immutable per-case score snapshot as reported by the runner. Score
+     * spans remain the queryable evidence; this copy survives trace retention
+     * and makes result retrieval self-contained.
+     * @type {{ [key: string]: number | null; }}
+     * @memberof EvalExperimentResultArtifactView
+     */
+    scores?: {
+        [key: string]: number | null;
+    };
     /**
      * The gateway `session` this case ran as, when the run was cataloged
      * (resolved at create). Links a result to its transcript.

package/dist/raw/eval/models/EvalExperimentResultArtifactView.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
  *
@@ -60,6 +60,7 @@ function EvalExperimentResultArtifactViewFromJSONTyped(json, ignoreDiscriminator
         return json;
     }
     return {
+        'app': json['app'] == null ? undefined : json['app'],
         'artifactStatus': json['artifact_status'],
         'commitTs': json['commit_ts'],
         'contentHashHex': json['content_hash_hex'] == null ? undefined : json['content_hash_hex'],
@@ -67,7 +68,6 @@ function EvalExperimentResultArtifactViewFromJSONTyped(json, ignoreDiscriminator
         'createdBy': json['created_by'],
         'description': json['description'] == null ? undefined : json['description'],
         'displayName': json['display_name'] == null ? undefined : json['display_name'],
-        'extension': json['extension'] == null ? undefined : json['extension'],
         'id': json['id'],
         'identity': json['identity'] == null ? undefined : (0, GatewayArtifactIdentityRevealView_1.GatewayArtifactIdentityRevealViewFromJSON)(json['identity']),
         'isLatest': json['is_latest'],
@@ -93,6 +93,7 @@ function EvalExperimentResultArtifactViewFromJSONTyped(json, ignoreDiscriminator
         'versionLabel': json['version_label'] == null ? undefined : json['version_label'],
         'versionOrdinal': json['version_ordinal'],
         'workspaceId': json['workspace_id'],
+        'caseId': json['case_id'] === undefined ? undefined : json['case_id'] === null ? null : json['case_id'],
         'datasetRecordId': json['dataset_record_id'] === undefined ? undefined : json['dataset_record_id'] === null ? null : json['dataset_record_id'],
         'durationMs': json['duration_ms'] === undefined ? undefined : json['duration_ms'] === null ? null : json['duration_ms'],
         'error': json['error'] === undefined ? undefined : json['error'] === null ? null : json['error'],
@@ -100,6 +101,7 @@ function EvalExperimentResultArtifactViewFromJSONTyped(json, ignoreDiscriminator
         'experimentId': json['experiment_id'] == null ? undefined : json['experiment_id'],
         'input': json['input'] === undefined ? undefined : json['input'] === null ? null : json['input'],
         'output': json['output'] === undefined ? undefined : json['output'] === null ? null : json['output'],
+        'scores': json['scores'] == null ? undefined : json['scores'],
         'sessionId': json['session_id'] === undefined ? undefined : json['session_id'] === null ? null : json['session_id'],
         'traceId': json['trace_id'] === undefined ? undefined : json['trace_id'] === null ? null : json['trace_id'],
         'permissions': json['permissions'] === undefined ? undefined : json['permissions'] === null ? null : json['permissions'],
@@ -114,6 +116,7 @@ function EvalExperimentResultArtifactViewToJSONTyped(value, ignoreDiscriminator
         return value;
     }
     return {
+        'app': value['app'],
         'artifact_status': value['artifactStatus'],
         'commit_ts': value['commitTs'],
         'content_hash_hex': value['contentHashHex'],
@@ -121,7 +124,6 @@ function EvalExperimentResultArtifactViewToJSONTyped(value, ignoreDiscriminator
         'created_by': value['createdBy'],
         'description': value['description'],
         'display_name': value['displayName'],
-        'extension': value['extension'],
         'id': value['id'],
         'identity': (0, GatewayArtifactIdentityRevealView_1.GatewayArtifactIdentityRevealViewToJSON)(value['identity']),
         'is_latest': value['isLatest'],
@@ -147,6 +149,7 @@ function EvalExperimentResultArtifactViewToJSONTyped(value, ignoreDiscriminator
         'version_label': value['versionLabel'],
         'version_ordinal': value['versionOrdinal'],
         'workspace_id': value['workspaceId'],
+        'case_id': value['caseId'],
         'dataset_record_id': value['datasetRecordId'],
         'duration_ms': value['durationMs'],
         'error': value['error'],
@@ -154,6 +157,7 @@ function EvalExperimentResultArtifactViewToJSONTyped(value, ignoreDiscriminator
         'experiment_id': value['experimentId'],
         'input': value['input'],
         'output': value['output'],
+        'scores': value['scores'],
         'session_id': value['sessionId'],
         'trace_id': value['traceId'],
         'permissions': value['permissions'],

package/dist/raw/eval/models/EvalExperimentResultBulkCreate.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/EvalExperimentResultBulkCreate.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/EvalExperimentResultCreate.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
  *
@@ -132,6 +132,13 @@ export interface EvalExperimentResultCreate {
      * @memberof EvalExperimentResultCreate
      */
     workspaceId: string;
+    /**
+     * Stable identity for this case within the experiment. Reusing it is
+     * rejected, so interrupted uploads can list and skip completed cases.
+     * @type {string}
+     * @memberof EvalExperimentResultCreate
+     */
+    caseId?: string | null;
     /**
      * The dataset row this case came from; omit for an ad-hoc case.
      * @type {string}
@@ -174,6 +181,14 @@ export interface EvalExperimentResultCreate {
     output?: {
         [key: string]: any;
     } | null;
+    /**
+     * Per-case scorer values. Numeric null means the scorer abstained.
+     * @type {{ [key: string]: number | null; }}
+     * @memberof EvalExperimentResultCreate
+     */
+    scores?: {
+        [key: string]: number | null;
+    };
     /**
      * The gateway session this case ran as, when the run was cataloged.
      * @type {string}

package/dist/raw/eval/models/EvalExperimentResultCreate.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
  *
@@ -53,12 +53,14 @@ function EvalExperimentResultCreateFromJSONTyped(json, ignoreDiscriminator) {
         'projectId': json['project_id'] == null ? undefined : json['project_id'],
         'repoId': json['repo_id'] == null ? undefined : json['repo_id'],
         'workspaceId': json['workspace_id'],
+        'caseId': json['case_id'] === undefined ? undefined : json['case_id'] === null ? null : json['case_id'],
         'datasetRecordId': json['dataset_record_id'] === undefined ? undefined : json['dataset_record_id'] === null ? null : json['dataset_record_id'],
         'durationMs': json['duration_ms'] === undefined ? undefined : json['duration_ms'] === null ? null : json['duration_ms'],
         'error': json['error'] === undefined ? undefined : json['error'] === null ? null : json['error'],
         'expected': json['expected'] === undefined ? undefined : json['expected'] === null ? null : json['expected'],
         'input': json['input'] === undefined ? undefined : json['input'] === null ? null : json['input'],
         'output': json['output'] === undefined ? undefined : json['output'] === null ? null : json['output'],
+        'scores': json['scores'] == null ? undefined : json['scores'],
         'sessionId': json['session_id'] === undefined ? undefined : json['session_id'] === null ? null : json['session_id'],
         'traceId': json['trace_id'] === undefined ? undefined : json['trace_id'] === null ? null : json['trace_id'],
     };
@@ -88,12 +90,14 @@ function EvalExperimentResultCreateToJSONTyped(value, ignoreDiscriminator = fals
         'project_id': value['projectId'],
         'repo_id': value['repoId'],
         'workspace_id': value['workspaceId'],
+        'case_id': value['caseId'],
         'dataset_record_id': value['datasetRecordId'],
         'duration_ms': value['durationMs'],
         'error': value['error'],
         'expected': value['expected'],
         'input': value['input'],
         'output': value['output'],
+        'scores': value['scores'],
         'session_id': value['sessionId'],
         'trace_id': value['traceId'],
     };

package/dist/raw/eval/models/EvalExperimentResultMeta.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
  *
@@ -26,6 +26,13 @@
  * @interface EvalExperimentResultMeta
  */
 export interface EvalExperimentResultMeta {
+    /**
+     * Stable runner-owned identity for this case within the experiment.
+     * Unique per experiment, enabling safe retry and server-backed resume.
+     * @type {string}
+     * @memberof EvalExperimentResultMeta
+     */
+    caseId?: string | null;
     /**
      * The dataset row this case came from; absent for an ad-hoc case. The
      * per-case join key when comparing two experiments over one dataset.
@@ -76,6 +83,16 @@ export interface EvalExperimentResultMeta {
     output?: {
         [key: string]: any;
     } | null;
+    /**
+     * Immutable per-case score snapshot as reported by the runner. Score
+     * spans remain the queryable evidence; this copy survives trace retention
+     * and makes result retrieval self-contained.
+     * @type {{ [key: string]: number | null; }}
+     * @memberof EvalExperimentResultMeta
+     */
+    scores?: {
+        [key: string]: number | null;
+    };
     /**
      * The gateway `session` this case ran as, when the run was cataloged
      * (resolved at create). Links a result to its transcript.

package/dist/raw/eval/models/EvalExperimentResultMeta.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
  *
@@ -34,6 +34,7 @@ function EvalExperimentResultMetaFromJSONTyped(json, ignoreDiscriminator) {
         return json;
     }
     return {
+        'caseId': json['case_id'] === undefined ? undefined : json['case_id'] === null ? null : json['case_id'],
         'datasetRecordId': json['dataset_record_id'] === undefined ? undefined : json['dataset_record_id'] === null ? null : json['dataset_record_id'],
         'durationMs': json['duration_ms'] === undefined ? undefined : json['duration_ms'] === null ? null : json['duration_ms'],
         'error': json['error'] === undefined ? undefined : json['error'] === null ? null : json['error'],
@@ -41,6 +42,7 @@ function EvalExperimentResultMetaFromJSONTyped(json, ignoreDiscriminator) {
         'experimentId': json['experiment_id'],
         'input': json['input'] === undefined ? undefined : json['input'] === null ? null : json['input'],
         'output': json['output'] === undefined ? undefined : json['output'] === null ? null : json['output'],
+        'scores': json['scores'] == null ? undefined : json['scores'],
         'sessionId': json['session_id'] === undefined ? undefined : json['session_id'] === null ? null : json['session_id'],
         'traceId': json['trace_id'] === undefined ? undefined : json['trace_id'] === null ? null : json['trace_id'],
     };
@@ -53,6 +55,7 @@ function EvalExperimentResultMetaToJSONTyped(value, ignoreDiscriminator = false)
         return value;
     }
     return {
+        'case_id': value['caseId'],
         'dataset_record_id': value['datasetRecordId'],
         'duration_ms': value['durationMs'],
         'error': value['error'],
@@ -60,6 +63,7 @@ function EvalExperimentResultMetaToJSONTyped(value, ignoreDiscriminator = false)
         'experiment_id': value['experimentId'],
         'input': value['input'],
         'output': value['output'],
+        'scores': value['scores'],
         'session_id': value['sessionId'],
         'trace_id': value['traceId'],
     };

package/dist/raw/eval/models/EvalExperimentStatus.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/EvalExperimentStatus.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/EvalGetDataset404Response.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/EvalGetDataset404Response.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
  *