@atlanai/sdk 0.2.1 → 0.2.2

package/dist/raw/eval/models/EvalDatasetArtifactView.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 EvalDatasetArtifactView
  */
 export interface EvalDatasetArtifactView {
+    /**
+     * Value of the `app` field.
+     * @type {string}
+     * @memberof EvalDatasetArtifactView
+     */
+    app?: string;
     /**
      * Value of the `artifact_status` field.
      * @type {string}
@@ -59,12 +65,6 @@ export interface EvalDatasetArtifactView {
      * @memberof EvalDatasetArtifactView
      */
     displayName?: string;
-    /**
-     * Value of the `extension` field.
-     * @type {string}
-     * @memberof EvalDatasetArtifactView
-     */
-    extension?: string;
     /**
      * Kind-prefixed Crockford-base32 handle (`<kind>_<base32>`).
      * @type {string}
@@ -103,19 +103,28 @@ export interface EvalDatasetArtifactView {
      * 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 EvalDatasetArtifactView
      */
     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 EvalDatasetArtifactView
      */
@@ -223,13 +232,16 @@ export interface EvalDatasetArtifactView {
      */
     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 EvalDatasetArtifactView
      */
     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 EvalDatasetArtifactView
      */
@@ -252,6 +264,13 @@ export interface EvalDatasetArtifactView {
     extra?: {
         [key: string]: any;
     } | null;
+    /**
+     * The key in each record's `input` object that should be rendered as the
+     * human-readable case prompt. Existing datasets default to `question`.
+     * @type {string}
+     * @memberof EvalDatasetArtifactView
+     */
+    inputKey?: string;
     /**
      * Value of the `permissions` field.
      * @type {{ [key: string]: boolean; }}

package/dist/raw/eval/models/EvalDatasetArtifactView.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 EvalDatasetArtifactViewFromJSONTyped(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 EvalDatasetArtifactViewFromJSONTyped(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'],
@@ -94,6 +94,7 @@ function EvalDatasetArtifactViewFromJSONTyped(json, ignoreDiscriminator) {
         'versionOrdinal': json['version_ordinal'],
         'workspaceId': json['workspace_id'],
         'extra': json['extra'] === undefined ? undefined : json['extra'] === null ? null : json['extra'],
+        'inputKey': json['input_key'] == null ? undefined : json['input_key'],
         'permissions': json['permissions'] === undefined ? undefined : json['permissions'] === null ? null : json['permissions'],
         'relationships': json['relationships'] === undefined ? undefined : json['relationships'] === null ? null : (0, RegistryRelationshipsView_1.RegistryRelationshipsViewFromJSON)(json['relationships']),
     };
@@ -106,6 +107,7 @@ function EvalDatasetArtifactViewToJSONTyped(value, ignoreDiscriminator = false)
         return value;
     }
     return {
+        'app': value['app'],
         'artifact_status': value['artifactStatus'],
         'commit_ts': value['commitTs'],
         'content_hash_hex': value['contentHashHex'],
@@ -113,7 +115,6 @@ function EvalDatasetArtifactViewToJSONTyped(value, ignoreDiscriminator = false)
         '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'],
@@ -140,6 +141,7 @@ function EvalDatasetArtifactViewToJSONTyped(value, ignoreDiscriminator = false)
         'version_ordinal': value['versionOrdinal'],
         'workspace_id': value['workspaceId'],
         'extra': value['extra'],
+        'input_key': value['inputKey'],
         'permissions': value['permissions'],
         'relationships': (0, RegistryRelationshipsView_1.RegistryRelationshipsViewToJSON)(value['relationships']),
     };

package/dist/raw/eval/models/EvalDatasetCreate.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
  *
@@ -139,6 +139,12 @@ export interface EvalDatasetCreate {
     extra?: {
         [key: string]: any;
     } | null;
+    /**
+     * Input-object key rendered as the case question. Defaults to `question`.
+     * @type {string}
+     * @memberof EvalDatasetCreate
+     */
+    inputKey?: string;
 }
 /**
  * Check if a given object implements the EvalDatasetCreate interface.

package/dist/raw/eval/models/EvalDatasetCreate.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
  *
@@ -54,6 +54,7 @@ function EvalDatasetCreateFromJSONTyped(json, ignoreDiscriminator) {
         'repoId': json['repo_id'] == null ? undefined : json['repo_id'],
         'workspaceId': json['workspace_id'],
         'extra': json['extra'] === undefined ? undefined : json['extra'] === null ? null : json['extra'],
+        'inputKey': json['input_key'] == null ? undefined : json['input_key'],
     };
 }
 function EvalDatasetCreateToJSON(json) {
@@ -82,5 +83,6 @@ function EvalDatasetCreateToJSONTyped(value, ignoreDiscriminator = false) {
         'repo_id': value['repoId'],
         'workspace_id': value['workspaceId'],
         'extra': value['extra'],
+        'input_key': value['inputKey'],
     };
 }

package/dist/raw/eval/models/EvalDatasetMeta.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
  *
@@ -35,6 +35,13 @@ export interface EvalDatasetMeta {
     extra?: {
         [key: string]: any;
     } | null;
+    /**
+     * The key in each record's `input` object that should be rendered as the
+     * human-readable case prompt. Existing datasets default to `question`.
+     * @type {string}
+     * @memberof EvalDatasetMeta
+     */
+    inputKey?: string;
 }
 /**
  * Check if a given object implements the EvalDatasetMeta interface.

package/dist/raw/eval/models/EvalDatasetMeta.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
  *
@@ -33,6 +33,7 @@ function EvalDatasetMetaFromJSONTyped(json, ignoreDiscriminator) {
     }
     return {
         'extra': json['extra'] === undefined ? undefined : json['extra'] === null ? null : json['extra'],
+        'inputKey': json['input_key'] == null ? undefined : json['input_key'],
     };
 }
 function EvalDatasetMetaToJSON(json) {
@@ -44,5 +45,6 @@ function EvalDatasetMetaToJSONTyped(value, ignoreDiscriminator = false) {
     }
     return {
         'extra': value['extra'],
+        'input_key': value['inputKey'],
     };
 }

package/dist/raw/eval/models/EvalDatasetPatch.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
  *
@@ -35,6 +35,12 @@ export interface EvalDatasetPatch {
     extra?: {
         [key: string]: any;
     } | null;
+    /**
+     * Change the display key used for future reads and experiment snapshots.
+     * @type {string}
+     * @memberof EvalDatasetPatch
+     */
+    inputKey?: string | null;
     /**
      * Stable machine-readable name.
      * @type {string}

package/dist/raw/eval/models/EvalDatasetPatch.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
  *
@@ -35,6 +35,7 @@ function EvalDatasetPatchFromJSONTyped(json, ignoreDiscriminator) {
         'description': json['description'] === undefined ? undefined : json['description'] === null ? null : json['description'],
         'displayName': json['display_name'] === undefined ? undefined : json['display_name'] === null ? null : json['display_name'],
         'extra': json['extra'] === undefined ? undefined : json['extra'] === null ? null : json['extra'],
+        'inputKey': json['input_key'] === undefined ? undefined : json['input_key'] === null ? null : json['input_key'],
         'name': json['name'] === undefined ? undefined : json['name'] === null ? null : json['name'],
     };
 }
@@ -49,6 +50,7 @@ function EvalDatasetPatchToJSONTyped(value, ignoreDiscriminator = false) {
         'description': value['description'],
         'display_name': value['displayName'],
         'extra': value['extra'],
+        'input_key': value['inputKey'],
         'name': value['name'],
     };
 }

package/dist/raw/eval/models/EvalDatasetRecordArtifactView.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
  *
@@ -18,6 +18,12 @@ import type { RegistryRelationshipsView } from './RegistryRelationshipsView';
  * @interface EvalDatasetRecordArtifactView
  */
 export interface EvalDatasetRecordArtifactView {
+    /**
+     * Value of the `app` field.
+     * @type {string}
+     * @memberof EvalDatasetRecordArtifactView
+     */
+    app?: string;
     /**
      * Value of the `artifact_status` field.
      * @type {string}
@@ -60,12 +66,6 @@ export interface EvalDatasetRecordArtifactView {
      * @memberof EvalDatasetRecordArtifactView
      */
     displayName?: string;
-    /**
-     * Value of the `extension` field.
-     * @type {string}
-     * @memberof EvalDatasetRecordArtifactView
-     */
-    extension?: string;
     /**
      * Kind-prefixed Crockford-base32 handle (`<kind>_<base32>`).
      * @type {string}
@@ -104,19 +104,28 @@ export interface EvalDatasetRecordArtifactView {
      * 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 EvalDatasetRecordArtifactView
      */
     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 EvalDatasetRecordArtifactView
      */
@@ -224,13 +233,16 @@ export interface EvalDatasetRecordArtifactView {
      */
     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 EvalDatasetRecordArtifactView
      */
     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 EvalDatasetRecordArtifactView
      */
@@ -257,6 +269,12 @@ export interface EvalDatasetRecordArtifactView {
      * @memberof EvalDatasetRecordArtifactView
      */
     datasetId?: string;
+    /**
+     *
+     * @type {any}
+     * @memberof EvalDatasetRecordArtifactView
+     */
+    displayInput?: any | null;
     /**
      * The expected outcome, when the case has one. A labeled example carries
      * it; a raw traffic sample often does not.

package/dist/raw/eval/models/EvalDatasetRecordArtifactView.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
  *
@@ -61,6 +61,7 @@ function EvalDatasetRecordArtifactViewFromJSONTyped(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'],
@@ -68,7 +69,6 @@ function EvalDatasetRecordArtifactViewFromJSONTyped(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'],
@@ -96,6 +96,7 @@ function EvalDatasetRecordArtifactViewFromJSONTyped(json, ignoreDiscriminator) {
         'workspaceId': json['workspace_id'],
         'categories': json['categories'] == null ? undefined : json['categories'],
         'datasetId': json['dataset_id'] == null ? undefined : json['dataset_id'],
+        'displayInput': json['display_input'] === undefined ? undefined : json['display_input'] === null ? null : json['display_input'],
         'expected': json['expected'] === undefined ? undefined : json['expected'] === null ? null : json['expected'],
         'extra': json['extra'] === undefined ? undefined : json['extra'] === null ? null : json['extra'],
         'input': json['input'] == null ? undefined : json['input'],
@@ -114,6 +115,7 @@ function EvalDatasetRecordArtifactViewToJSONTyped(value, ignoreDiscriminator = f
         return value;
     }
     return {
+        'app': value['app'],
         'artifact_status': value['artifactStatus'],
         'commit_ts': value['commitTs'],
         'content_hash_hex': value['contentHashHex'],
@@ -121,7 +123,6 @@ function EvalDatasetRecordArtifactViewToJSONTyped(value, ignoreDiscriminator = f
         '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'],
@@ -149,6 +150,7 @@ function EvalDatasetRecordArtifactViewToJSONTyped(value, ignoreDiscriminator = f
         'workspace_id': value['workspaceId'],
         'categories': value['categories'],
         'dataset_id': value['datasetId'],
+        'display_input': value['displayInput'],
         'expected': value['expected'],
         'extra': value['extra'],
         'input': value['input'],

package/dist/raw/eval/models/EvalDatasetRecordBulkCreate.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/EvalDatasetRecordBulkCreate.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/EvalDatasetRecordCreate.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/EvalDatasetRecordCreate.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/EvalDatasetRecordMeta.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
  *
@@ -37,6 +37,12 @@ export interface EvalDatasetRecordMeta {
      * @memberof EvalDatasetRecordMeta
      */
     datasetId: string;
+    /**
+     *
+     * @type {any}
+     * @memberof EvalDatasetRecordMeta
+     */
+    displayInput?: any | null;
     /**
      * The expected outcome, when the case has one. A labeled example carries
      * it; a raw traffic sample often does not.

package/dist/raw/eval/models/EvalDatasetRecordMeta.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
  *
@@ -39,6 +39,7 @@ function EvalDatasetRecordMetaFromJSONTyped(json, ignoreDiscriminator) {
     return {
         'categories': json['categories'] == null ? undefined : json['categories'],
         'datasetId': json['dataset_id'],
+        'displayInput': json['display_input'] === undefined ? undefined : json['display_input'] === null ? null : json['display_input'],
         'expected': json['expected'] === undefined ? undefined : json['expected'] === null ? null : json['expected'],
         'extra': json['extra'] === undefined ? undefined : json['extra'] === null ? null : json['extra'],
         'input': json['input'],
@@ -57,6 +58,7 @@ function EvalDatasetRecordMetaToJSONTyped(value, ignoreDiscriminator = false) {
     return {
         'categories': value['categories'],
         'dataset_id': value['datasetId'],
+        'display_input': value['displayInput'],
         'expected': value['expected'],
         'extra': value['extra'],
         'input': value['input'],

package/dist/raw/eval/models/EvalDatasetRecordPatch.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/EvalDatasetRecordPatch.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
  *