@atlanai/sdk 0.2.1 → 0.2.2

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

@@ -0,0 +1,57 @@
+/**
+ * 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. 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
+ *
+ *
+ * NOTE: This class is auto generated by OpenAPI Generator (https://openapi-generator.tech).
+ * https://openapi-generator.tech
+ * Do not edit the class manually.
+ */
+import type { EvalDatasetSnapshotRecord } from './EvalDatasetSnapshotRecord';
+/**
+ * The immutable dataset materialization stamped onto an experiment.
+ * @export
+ * @interface EvalDatasetSnapshot
+ */
+export interface EvalDatasetSnapshot {
+    /**
+     * Value of the `content_hash_hex` field.
+     * @type {string}
+     * @memberof EvalDatasetSnapshot
+     */
+    contentHashHex?: string | null;
+    /**
+     * Stable identifier of the related dataset resource.
+     * @type {string}
+     * @memberof EvalDatasetSnapshot
+     */
+    datasetId: string;
+    /**
+     * Value of the `input_key` field.
+     * @type {string}
+     * @memberof EvalDatasetSnapshot
+     */
+    inputKey: string;
+    /**
+     * Value of the `records` field.
+     * @type {Array<EvalDatasetSnapshotRecord>}
+     * @memberof EvalDatasetSnapshot
+     */
+    records: Array<EvalDatasetSnapshotRecord>;
+    /**
+     * Value of the `version_ordinal` field.
+     * @type {number}
+     * @memberof EvalDatasetSnapshot
+     */
+    versionOrdinal: number;
+}
+/**
+ * Check if a given object implements the EvalDatasetSnapshot interface.
+ */
+export declare function instanceOfEvalDatasetSnapshot(value: object): value is EvalDatasetSnapshot;
+export declare function EvalDatasetSnapshotFromJSON(json: any): EvalDatasetSnapshot;
+export declare function EvalDatasetSnapshotFromJSONTyped(json: any, ignoreDiscriminator: boolean): EvalDatasetSnapshot;
+export declare function EvalDatasetSnapshotToJSON(json: any): EvalDatasetSnapshot;
+export declare function EvalDatasetSnapshotToJSONTyped(value?: EvalDatasetSnapshot | null, ignoreDiscriminator?: boolean): any;

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

@@ -0,0 +1,65 @@
+"use strict";
+/* tslint:disable */
+/* 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. 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
+ *
+ *
+ * NOTE: This class is auto generated by OpenAPI Generator (https://openapi-generator.tech).
+ * https://openapi-generator.tech
+ * Do not edit the class manually.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.instanceOfEvalDatasetSnapshot = instanceOfEvalDatasetSnapshot;
+exports.EvalDatasetSnapshotFromJSON = EvalDatasetSnapshotFromJSON;
+exports.EvalDatasetSnapshotFromJSONTyped = EvalDatasetSnapshotFromJSONTyped;
+exports.EvalDatasetSnapshotToJSON = EvalDatasetSnapshotToJSON;
+exports.EvalDatasetSnapshotToJSONTyped = EvalDatasetSnapshotToJSONTyped;
+const EvalDatasetSnapshotRecord_1 = require("./EvalDatasetSnapshotRecord");
+/**
+ * Check if a given object implements the EvalDatasetSnapshot interface.
+ */
+function instanceOfEvalDatasetSnapshot(value) {
+    if ((!('datasetId' in value) && !('dataset_id' in value)) || (value['datasetId'] === undefined && value['dataset_id'] === undefined))
+        return false;
+    if ((!('inputKey' in value) && !('input_key' in value)) || (value['inputKey'] === undefined && value['input_key'] === undefined))
+        return false;
+    if (!('records' in value) || value['records'] === undefined)
+        return false;
+    if ((!('versionOrdinal' in value) && !('version_ordinal' in value)) || (value['versionOrdinal'] === undefined && value['version_ordinal'] === undefined))
+        return false;
+    return true;
+}
+function EvalDatasetSnapshotFromJSON(json) {
+    return EvalDatasetSnapshotFromJSONTyped(json, false);
+}
+function EvalDatasetSnapshotFromJSONTyped(json, ignoreDiscriminator) {
+    if (json == null) {
+        return json;
+    }
+    return {
+        'contentHashHex': json['content_hash_hex'] === undefined ? undefined : json['content_hash_hex'] === null ? null : json['content_hash_hex'],
+        'datasetId': json['dataset_id'],
+        'inputKey': json['input_key'],
+        'records': (json['records'].map(EvalDatasetSnapshotRecord_1.EvalDatasetSnapshotRecordFromJSON)),
+        'versionOrdinal': json['version_ordinal'],
+    };
+}
+function EvalDatasetSnapshotToJSON(json) {
+    return EvalDatasetSnapshotToJSONTyped(json, false);
+}
+function EvalDatasetSnapshotToJSONTyped(value, ignoreDiscriminator = false) {
+    if (value == null) {
+        return value;
+    }
+    return {
+        'content_hash_hex': value['contentHashHex'],
+        'dataset_id': value['datasetId'],
+        'input_key': value['inputKey'],
+        'records': (value['records'].map(EvalDatasetSnapshotRecord_1.EvalDatasetSnapshotRecordToJSON)),
+        'version_ordinal': value['versionOrdinal'],
+    };
+}

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

@@ -0,0 +1,99 @@
+/**
+ * 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. 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
+ *
+ *
+ * NOTE: This class is auto generated by OpenAPI Generator (https://openapi-generator.tech).
+ * https://openapi-generator.tech
+ * Do not edit the class manually.
+ */
+import type { EvalRecordSource } from './EvalRecordSource';
+/**
+ * One frozen dataset row inside an experiment snapshot.
+ * @export
+ * @interface EvalDatasetSnapshotRecord
+ */
+export interface EvalDatasetSnapshotRecord {
+    /**
+     * Value of the `categories` field.
+     * @type {Array<string>}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    categories?: Array<string>;
+    /**
+     * Value of the `content_hash_hex` field.
+     * @type {string}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    contentHashHex?: string | null;
+    /**
+     *
+     * @type {any}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    displayInput?: any | null;
+    /**
+     * Value of the `expected` field.
+     * @type {{ [key: string]: any | null; }}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    expected?: {
+        [key: string]: any | null;
+    };
+    /**
+     * Value of the `extra` field.
+     * @type {{ [key: string]: any | null; }}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    extra?: {
+        [key: string]: any | null;
+    };
+    /**
+     * Stable identifier of this resource.
+     * @type {string}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    id: string;
+    /**
+     * Value of the `input` field.
+     * @type {{ [key: string]: any | null; }}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    input: {
+        [key: string]: any | null;
+    };
+    /**
+     * Value of the `label` field.
+     * @type {string}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    label?: string | null;
+    /**
+     *
+     * @type {EvalRecordSource}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    sourceKind: EvalRecordSource;
+    /**
+     * Value of the `source_ref` field.
+     * @type {string}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    sourceRef?: string | null;
+    /**
+     * Value of the `version_ordinal` field.
+     * @type {number}
+     * @memberof EvalDatasetSnapshotRecord
+     */
+    versionOrdinal: number;
+}
+/**
+ * Check if a given object implements the EvalDatasetSnapshotRecord interface.
+ */
+export declare function instanceOfEvalDatasetSnapshotRecord(value: object): value is EvalDatasetSnapshotRecord;
+export declare function EvalDatasetSnapshotRecordFromJSON(json: any): EvalDatasetSnapshotRecord;
+export declare function EvalDatasetSnapshotRecordFromJSONTyped(json: any, ignoreDiscriminator: boolean): EvalDatasetSnapshotRecord;
+export declare function EvalDatasetSnapshotRecordToJSON(json: any): EvalDatasetSnapshotRecord;
+export declare function EvalDatasetSnapshotRecordToJSONTyped(value?: EvalDatasetSnapshotRecord | null, ignoreDiscriminator?: boolean): any;

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

@@ -0,0 +1,77 @@
+"use strict";
+/* tslint:disable */
+/* 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. 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
+ *
+ *
+ * NOTE: This class is auto generated by OpenAPI Generator (https://openapi-generator.tech).
+ * https://openapi-generator.tech
+ * Do not edit the class manually.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.instanceOfEvalDatasetSnapshotRecord = instanceOfEvalDatasetSnapshotRecord;
+exports.EvalDatasetSnapshotRecordFromJSON = EvalDatasetSnapshotRecordFromJSON;
+exports.EvalDatasetSnapshotRecordFromJSONTyped = EvalDatasetSnapshotRecordFromJSONTyped;
+exports.EvalDatasetSnapshotRecordToJSON = EvalDatasetSnapshotRecordToJSON;
+exports.EvalDatasetSnapshotRecordToJSONTyped = EvalDatasetSnapshotRecordToJSONTyped;
+const EvalRecordSource_1 = require("./EvalRecordSource");
+/**
+ * Check if a given object implements the EvalDatasetSnapshotRecord interface.
+ */
+function instanceOfEvalDatasetSnapshotRecord(value) {
+    if (!('id' in value) || value['id'] === undefined)
+        return false;
+    if (!('input' in value) || value['input'] === undefined)
+        return false;
+    if ((!('sourceKind' in value) && !('source_kind' in value)) || (value['sourceKind'] === undefined && value['source_kind'] === undefined))
+        return false;
+    if ((!('versionOrdinal' in value) && !('version_ordinal' in value)) || (value['versionOrdinal'] === undefined && value['version_ordinal'] === undefined))
+        return false;
+    return true;
+}
+function EvalDatasetSnapshotRecordFromJSON(json) {
+    return EvalDatasetSnapshotRecordFromJSONTyped(json, false);
+}
+function EvalDatasetSnapshotRecordFromJSONTyped(json, ignoreDiscriminator) {
+    if (json == null) {
+        return json;
+    }
+    return {
+        'categories': json['categories'] == null ? undefined : json['categories'],
+        'contentHashHex': json['content_hash_hex'] === undefined ? undefined : json['content_hash_hex'] === null ? null : json['content_hash_hex'],
+        'displayInput': json['display_input'] === undefined ? undefined : json['display_input'] === null ? null : json['display_input'],
+        'expected': json['expected'] == null ? undefined : json['expected'],
+        'extra': json['extra'] == null ? undefined : json['extra'],
+        'id': json['id'],
+        'input': json['input'],
+        'label': json['label'] === undefined ? undefined : json['label'] === null ? null : json['label'],
+        'sourceKind': (0, EvalRecordSource_1.EvalRecordSourceFromJSON)(json['source_kind']),
+        'sourceRef': json['source_ref'] === undefined ? undefined : json['source_ref'] === null ? null : json['source_ref'],
+        'versionOrdinal': json['version_ordinal'],
+    };
+}
+function EvalDatasetSnapshotRecordToJSON(json) {
+    return EvalDatasetSnapshotRecordToJSONTyped(json, false);
+}
+function EvalDatasetSnapshotRecordToJSONTyped(value, ignoreDiscriminator = false) {
+    if (value == null) {
+        return value;
+    }
+    return {
+        'categories': value['categories'],
+        'content_hash_hex': value['contentHashHex'],
+        'display_input': value['displayInput'],
+        'expected': value['expected'],
+        'extra': value['extra'],
+        'id': value['id'],
+        'input': value['input'],
+        'label': value['label'],
+        'source_kind': (0, EvalRecordSource_1.EvalRecordSourceToJSON)(value['sourceKind']),
+        'source_ref': value['sourceRef'],
+        'version_ordinal': value['versionOrdinal'],
+    };
+}

package/dist/raw/eval/models/EvalExperimentArtifactView.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
  *
@@ -13,12 +13,19 @@ import type { GatewayArtifactIdentityRevealView } from './GatewayArtifactIdentit
 import type { EvalSubjectKind } from './EvalSubjectKind';
 import type { EvalExperimentStatus } from './EvalExperimentStatus';
 import type { RegistryRelationshipsView } from './RegistryRelationshipsView';
+import type { EvalDatasetSnapshot } from './EvalDatasetSnapshot';
 /**
  * Published representation of ExperimentArtifact.
  * @export
  * @interface EvalExperimentArtifactView
  */
 export interface EvalExperimentArtifactView {
+    /**
+     * Value of the `app` field.
+     * @type {string}
+     * @memberof EvalExperimentArtifactView
+     */
+    app?: string;
     /**
      * Value of the `artifact_status` field.
      * @type {string}
@@ -61,12 +68,6 @@ export interface EvalExperimentArtifactView {
      * @memberof EvalExperimentArtifactView
      */
     displayName?: string;
-    /**
-     * Value of the `extension` field.
-     * @type {string}
-     * @memberof EvalExperimentArtifactView
-     */
-    extension?: string;
     /**
      * Kind-prefixed Crockford-base32 handle (`<kind>_<base32>`).
      * @type {string}
@@ -105,19 +106,28 @@ export interface EvalExperimentArtifactView {
      * 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 EvalExperimentArtifactView
      */
     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 EvalExperimentArtifactView
      */
@@ -225,13 +235,16 @@ export interface EvalExperimentArtifactView {
      */
     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 EvalExperimentArtifactView
      */
     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 EvalExperimentArtifactView
      */
@@ -267,6 +280,12 @@ export interface EvalExperimentArtifactView {
      * @memberof EvalExperimentArtifactView
      */
     datasetId?: string | null;
+    /**
+     * Value of the `dataset_snapshot` field.
+     * @type {EvalDatasetSnapshot}
+     * @memberof EvalExperimentArtifactView
+     */
+    datasetSnapshot?: EvalDatasetSnapshot | null;
     /**
      *
      * @type {EvalExperimentStatus}

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
  *