@atlanai/sdk 0.1.1 → 0.2.2

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

@@ -0,0 +1,98 @@
+"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.instanceOfEvalScorerCreate = instanceOfEvalScorerCreate;
+exports.EvalScorerCreateFromJSON = EvalScorerCreateFromJSON;
+exports.EvalScorerCreateFromJSONTyped = EvalScorerCreateFromJSONTyped;
+exports.EvalScorerCreateToJSON = EvalScorerCreateToJSON;
+exports.EvalScorerCreateToJSONTyped = EvalScorerCreateToJSONTyped;
+const EvalScorerKind_1 = require("./EvalScorerKind");
+const EvalScorerScope_1 = require("./EvalScorerScope");
+/**
+ * Check if a given object implements the EvalScorerCreate interface.
+ */
+function instanceOfEvalScorerCreate(value) {
+    if (!('name' in value) || value['name'] === undefined)
+        return false;
+    if ((!('workspaceId' in value) && !('workspace_id' in value)) || (value['workspaceId'] === undefined && value['workspace_id'] === undefined))
+        return false;
+    if (!('scope' in value) || value['scope'] === undefined)
+        return false;
+    if ((!('scorerKind' in value) && !('scorer_kind' in value)) || (value['scorerKind'] === undefined && value['scorer_kind'] === undefined))
+        return false;
+    return true;
+}
+function EvalScorerCreateFromJSON(json) {
+    return EvalScorerCreateFromJSONTyped(json, false);
+}
+function EvalScorerCreateFromJSONTyped(json, ignoreDiscriminator) {
+    if (json == null) {
+        return json;
+    }
+    return {
+        'sourceCreatedAt': json['source_created_at'] == null ? undefined : json['source_created_at'],
+        'sourceCreatedBy': json['source_created_by'] == null ? undefined : json['source_created_by'],
+        'sourceId': json['source_id'] == null ? undefined : json['source_id'],
+        'sourceLastModifiedAt': json['source_last_modified_at'] == null ? undefined : json['source_last_modified_at'],
+        'sourceLastModifiedBy': json['source_last_modified_by'] == null ? undefined : json['source_last_modified_by'],
+        'sourceMetadata': json['source_metadata'] === undefined ? undefined : json['source_metadata'] === null ? null : json['source_metadata'],
+        'sourceSyncedAt': json['source_synced_at'] == null ? undefined : json['source_synced_at'],
+        'sourceType': json['source_type'] == null ? undefined : json['source_type'],
+        'sourceUri': json['source_uri'] == null ? undefined : json['source_uri'],
+        'sourceVersion': json['source_version'] == null ? undefined : json['source_version'],
+        'description': json['description'] == null ? undefined : json['description'],
+        'displayName': json['display_name'] == null ? undefined : json['display_name'],
+        'name': json['name'],
+        'ownerId': json['owner_id'] == null ? undefined : json['owner_id'],
+        'projectId': json['project_id'] == null ? undefined : json['project_id'],
+        'repoId': json['repo_id'] == null ? undefined : json['repo_id'],
+        'workspaceId': json['workspace_id'],
+        'outputs': json['outputs'] === undefined ? undefined : json['outputs'] === null ? null : json['outputs'],
+        'scope': (0, EvalScorerScope_1.EvalScorerScopeFromJSON)(json['scope']),
+        'scorerKind': (0, EvalScorerKind_1.EvalScorerKindFromJSON)(json['scorer_kind']),
+        'spec': json['spec'] === undefined ? undefined : json['spec'] === null ? null : json['spec'],
+    };
+}
+function EvalScorerCreateToJSON(json) {
+    return EvalScorerCreateToJSONTyped(json, false);
+}
+function EvalScorerCreateToJSONTyped(value, ignoreDiscriminator = false) {
+    if (value == null) {
+        return value;
+    }
+    return {
+        'source_created_at': value['sourceCreatedAt'],
+        'source_created_by': value['sourceCreatedBy'],
+        'source_id': value['sourceId'],
+        'source_last_modified_at': value['sourceLastModifiedAt'],
+        'source_last_modified_by': value['sourceLastModifiedBy'],
+        'source_metadata': value['sourceMetadata'],
+        'source_synced_at': value['sourceSyncedAt'],
+        'source_type': value['sourceType'],
+        'source_uri': value['sourceUri'],
+        'source_version': value['sourceVersion'],
+        'description': value['description'],
+        'display_name': value['displayName'],
+        'name': value['name'],
+        'owner_id': value['ownerId'],
+        'project_id': value['projectId'],
+        'repo_id': value['repoId'],
+        'workspace_id': value['workspaceId'],
+        'outputs': value['outputs'],
+        'scope': (0, EvalScorerScope_1.EvalScorerScopeToJSON)(value['scope']),
+        'scorer_kind': (0, EvalScorerKind_1.EvalScorerKindToJSON)(value['scorerKind']),
+        'spec': value['spec'],
+    };
+}

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

@@ -0,0 +1,32 @@
+/**
+ * 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.
+ */
+/**
+ * What implements a scorer. Promoted to the indexed `scorer_kind` `CHECK`
+ * column; doc-comment-free per variant.
+ *
+ * `llm_judge` and `human` describe evaluations the platform can eventually
+ * host (a judge is an LLM call; a review is a form); `code` is cataloged here
+ * but **executed externally, permanently** — sandboxing arbitrary grader code
+ * is a different product.
+ * @export
+ * @enum {string}
+ */
+export declare enum EvalScorerKind {
+    LlmJudge = "llm_judge",
+    Code = "code",
+    Human = "human"
+}
+export declare function instanceOfEvalScorerKind(value: any): boolean;
+export declare function EvalScorerKindFromJSON(json: any): EvalScorerKind;
+export declare function EvalScorerKindFromJSONTyped(json: any, ignoreDiscriminator: boolean): EvalScorerKind;
+export declare function EvalScorerKindToJSON(value?: EvalScorerKind | null): any;
+export declare function EvalScorerKindToJSONTyped(value: any, ignoreDiscriminator: boolean): EvalScorerKind;

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

@@ -0,0 +1,60 @@
+"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.EvalScorerKind = void 0;
+exports.instanceOfEvalScorerKind = instanceOfEvalScorerKind;
+exports.EvalScorerKindFromJSON = EvalScorerKindFromJSON;
+exports.EvalScorerKindFromJSONTyped = EvalScorerKindFromJSONTyped;
+exports.EvalScorerKindToJSON = EvalScorerKindToJSON;
+exports.EvalScorerKindToJSONTyped = EvalScorerKindToJSONTyped;
+/**
+ * What implements a scorer. Promoted to the indexed `scorer_kind` `CHECK`
+ * column; doc-comment-free per variant.
+ *
+ * `llm_judge` and `human` describe evaluations the platform can eventually
+ * host (a judge is an LLM call; a review is a form); `code` is cataloged here
+ * but **executed externally, permanently** — sandboxing arbitrary grader code
+ * is a different product.
+ * @export
+ * @enum {string}
+ */
+var EvalScorerKind;
+(function (EvalScorerKind) {
+    EvalScorerKind["LlmJudge"] = "llm_judge";
+    EvalScorerKind["Code"] = "code";
+    EvalScorerKind["Human"] = "human";
+})(EvalScorerKind || (exports.EvalScorerKind = EvalScorerKind = {}));
+function instanceOfEvalScorerKind(value) {
+    for (const key in EvalScorerKind) {
+        if (Object.prototype.hasOwnProperty.call(EvalScorerKind, key)) {
+            if (EvalScorerKind[key] === value) {
+                return true;
+            }
+        }
+    }
+    return false;
+}
+function EvalScorerKindFromJSON(json) {
+    return EvalScorerKindFromJSONTyped(json, false);
+}
+function EvalScorerKindFromJSONTyped(json, ignoreDiscriminator) {
+    return json;
+}
+function EvalScorerKindToJSON(value) {
+    return value;
+}
+function EvalScorerKindToJSONTyped(value, ignoreDiscriminator) {
+    return value;
+}

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

@@ -0,0 +1,67 @@
+/**
+ * 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 { EvalScorerKind } from './EvalScorerKind';
+import type { EvalScorerScope } from './EvalScorerScope';
+/**
+ * The `scorer` artifact kind: a named, versioned evaluation definition — an
+ * LLM judge rubric, a cataloged code grader, or a human review form.
+ *
+ * Versioned by the artifact line: editing the rubric is a new version of the
+ * same scorer, and every `score` pins the version that judged it, so a rubric
+ * edit never rewrites what old verdicts meant.
+ * @export
+ * @interface EvalScorerMeta
+ */
+export interface EvalScorerMeta {
+    /**
+     * The score keys this scorer emits and their ranges/types, e.g.
+     * `{"accuracy": {"type": "numeric", "min": 0, "max": 1}}` — schema
+     * governance for consumers, so a console can render columns without
+     * guessing.
+     * @type {{ [key: string]: any; }}
+     * @memberof EvalScorerMeta
+     */
+    outputs?: {
+        [key: string]: any;
+    } | null;
+    /**
+     *
+     * @type {EvalScorerScope}
+     * @memberof EvalScorerMeta
+     */
+    scope: EvalScorerScope;
+    /**
+     *
+     * @type {EvalScorerKind}
+     * @memberof EvalScorerMeta
+     */
+    scorerKind: EvalScorerKind;
+    /**
+     * The definition, by kind: an `llm_judge` carries its rubric prompt, judge
+     * model, and variable mapping; a `code` scorer carries a repo/entrypoint
+     * pointer (the code itself is never executed here); a `human` scorer
+     * carries its review form schema. Round-tripped verbatim.
+     * @type {{ [key: string]: any; }}
+     * @memberof EvalScorerMeta
+     */
+    spec?: {
+        [key: string]: any;
+    } | null;
+}
+/**
+ * Check if a given object implements the EvalScorerMeta interface.
+ */
+export declare function instanceOfEvalScorerMeta(value: object): value is EvalScorerMeta;
+export declare function EvalScorerMetaFromJSON(json: any): EvalScorerMeta;
+export declare function EvalScorerMetaFromJSONTyped(json: any, ignoreDiscriminator: boolean): EvalScorerMeta;
+export declare function EvalScorerMetaToJSON(json: any): EvalScorerMeta;
+export declare function EvalScorerMetaToJSONTyped(value?: EvalScorerMeta | null, ignoreDiscriminator?: boolean): any;

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

@@ -0,0 +1,60 @@
+"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.instanceOfEvalScorerMeta = instanceOfEvalScorerMeta;
+exports.EvalScorerMetaFromJSON = EvalScorerMetaFromJSON;
+exports.EvalScorerMetaFromJSONTyped = EvalScorerMetaFromJSONTyped;
+exports.EvalScorerMetaToJSON = EvalScorerMetaToJSON;
+exports.EvalScorerMetaToJSONTyped = EvalScorerMetaToJSONTyped;
+const EvalScorerKind_1 = require("./EvalScorerKind");
+const EvalScorerScope_1 = require("./EvalScorerScope");
+/**
+ * Check if a given object implements the EvalScorerMeta interface.
+ */
+function instanceOfEvalScorerMeta(value) {
+    if (!('scope' in value) || value['scope'] === undefined)
+        return false;
+    if ((!('scorerKind' in value) && !('scorer_kind' in value)) || (value['scorerKind'] === undefined && value['scorer_kind'] === undefined))
+        return false;
+    return true;
+}
+function EvalScorerMetaFromJSON(json) {
+    return EvalScorerMetaFromJSONTyped(json, false);
+}
+function EvalScorerMetaFromJSONTyped(json, ignoreDiscriminator) {
+    if (json == null) {
+        return json;
+    }
+    return {
+        'outputs': json['outputs'] === undefined ? undefined : json['outputs'] === null ? null : json['outputs'],
+        'scope': (0, EvalScorerScope_1.EvalScorerScopeFromJSON)(json['scope']),
+        'scorerKind': (0, EvalScorerKind_1.EvalScorerKindFromJSON)(json['scorer_kind']),
+        'spec': json['spec'] === undefined ? undefined : json['spec'] === null ? null : json['spec'],
+    };
+}
+function EvalScorerMetaToJSON(json) {
+    return EvalScorerMetaToJSONTyped(json, false);
+}
+function EvalScorerMetaToJSONTyped(value, ignoreDiscriminator = false) {
+    if (value == null) {
+        return value;
+    }
+    return {
+        'outputs': value['outputs'],
+        'scope': (0, EvalScorerScope_1.EvalScorerScopeToJSON)(value['scope']),
+        'scorer_kind': (0, EvalScorerKind_1.EvalScorerKindToJSON)(value['scorerKind']),
+        'spec': value['spec'],
+    };
+}

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

@@ -0,0 +1,75 @@
+/**
+ * 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.
+ */
+/**
+ * The editable surface of a scorer.
+ *
+ * `scorer_kind` and `scope` are absent on purpose: they are the scorer's
+ * identity, and changing them on the same line would reinterpret every score
+ * citing it (the update hook pins them regardless).
+ *
+ * **Editing `spec`/`outputs` does not currently mint a version.** The plane's
+ * PATCH edits the head in place — `version_ordinal` stays where it was, and
+ * only source-reconcile appends versions today — so a rubric edit silently
+ * changes what every score pinned to that ordinal actually cited. The pin is
+ * therefore honest about *which ordinal* judged, but that ordinal's content is
+ * mutable, which is not the guarantee this kind is supposed to give. Until
+ * update-mints-version lands, treat a material rubric change as a **new
+ * scorer**, not a patch. Tracked as an ADR; the alternative fix is to refuse
+ * `spec`/`outputs` edits outright.
+ * @export
+ * @interface EvalScorerPatch
+ */
+export interface EvalScorerPatch {
+    /**
+     * Set or clear the description. Explicit `null` clears it, per RFC 7396.
+     * @type {string}
+     * @memberof EvalScorerPatch
+     */
+    description?: string | null;
+    /**
+     * Human-readable name.
+     * @type {string}
+     * @memberof EvalScorerPatch
+     */
+    displayName?: string | null;
+    /**
+     * Stable machine-readable name.
+     * @type {string}
+     * @memberof EvalScorerPatch
+     */
+    name?: string | null;
+    /**
+     * Replace or clear the declared outputs wholesale. Explicit `null` clears
+     * it.
+     * @type {{ [key: string]: any; }}
+     * @memberof EvalScorerPatch
+     */
+    outputs?: {
+        [key: string]: any;
+    } | null;
+    /**
+     * Replace or clear the definition wholesale. Explicit `null` clears it.
+     * @type {{ [key: string]: any; }}
+     * @memberof EvalScorerPatch
+     */
+    spec?: {
+        [key: string]: any;
+    } | null;
+}
+/**
+ * Check if a given object implements the EvalScorerPatch interface.
+ */
+export declare function instanceOfEvalScorerPatch(value: object): value is EvalScorerPatch;
+export declare function EvalScorerPatchFromJSON(json: any): EvalScorerPatch;
+export declare function EvalScorerPatchFromJSONTyped(json: any, ignoreDiscriminator: boolean): EvalScorerPatch;
+export declare function EvalScorerPatchToJSON(json: any): EvalScorerPatch;
+export declare function EvalScorerPatchToJSONTyped(value?: EvalScorerPatch | null, ignoreDiscriminator?: boolean): any;

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

@@ -0,0 +1,56 @@
+"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.instanceOfEvalScorerPatch = instanceOfEvalScorerPatch;
+exports.EvalScorerPatchFromJSON = EvalScorerPatchFromJSON;
+exports.EvalScorerPatchFromJSONTyped = EvalScorerPatchFromJSONTyped;
+exports.EvalScorerPatchToJSON = EvalScorerPatchToJSON;
+exports.EvalScorerPatchToJSONTyped = EvalScorerPatchToJSONTyped;
+/**
+ * Check if a given object implements the EvalScorerPatch interface.
+ */
+function instanceOfEvalScorerPatch(value) {
+    return true;
+}
+function EvalScorerPatchFromJSON(json) {
+    return EvalScorerPatchFromJSONTyped(json, false);
+}
+function EvalScorerPatchFromJSONTyped(json, ignoreDiscriminator) {
+    if (json == null) {
+        return json;
+    }
+    return {
+        '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'],
+        'name': json['name'] === undefined ? undefined : json['name'] === null ? null : json['name'],
+        'outputs': json['outputs'] === undefined ? undefined : json['outputs'] === null ? null : json['outputs'],
+        'spec': json['spec'] === undefined ? undefined : json['spec'] === null ? null : json['spec'],
+    };
+}
+function EvalScorerPatchToJSON(json) {
+    return EvalScorerPatchToJSONTyped(json, false);
+}
+function EvalScorerPatchToJSONTyped(value, ignoreDiscriminator = false) {
+    if (value == null) {
+        return value;
+    }
+    return {
+        'description': value['description'],
+        'display_name': value['displayName'],
+        'name': value['name'],
+        'outputs': value['outputs'],
+        'spec': value['spec'],
+    };
+}

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

@@ -0,0 +1,32 @@
+/**
+ * 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.
+ */
+/**
+ * What a scorer needs to read. Promoted to the `scope` `CHECK` column;
+ * doc-comment-free per variant.
+ *
+ * An exact-match grader reads only the result's input/output (`result`); the
+ * claw critic reads the whole trajectory (`trace`); a conversation review
+ * reads the transcript (`session`). Declared so a runner (and, later, the
+ * hosted judge) knows what evidence to fetch before scoring.
+ * @export
+ * @enum {string}
+ */
+export declare enum EvalScorerScope {
+    Result = "result",
+    Session = "session",
+    Trace = "trace"
+}
+export declare function instanceOfEvalScorerScope(value: any): boolean;
+export declare function EvalScorerScopeFromJSON(json: any): EvalScorerScope;
+export declare function EvalScorerScopeFromJSONTyped(json: any, ignoreDiscriminator: boolean): EvalScorerScope;
+export declare function EvalScorerScopeToJSON(value?: EvalScorerScope | null): any;
+export declare function EvalScorerScopeToJSONTyped(value: any, ignoreDiscriminator: boolean): EvalScorerScope;

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

@@ -0,0 +1,60 @@
+"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.EvalScorerScope = void 0;
+exports.instanceOfEvalScorerScope = instanceOfEvalScorerScope;
+exports.EvalScorerScopeFromJSON = EvalScorerScopeFromJSON;
+exports.EvalScorerScopeFromJSONTyped = EvalScorerScopeFromJSONTyped;
+exports.EvalScorerScopeToJSON = EvalScorerScopeToJSON;
+exports.EvalScorerScopeToJSONTyped = EvalScorerScopeToJSONTyped;
+/**
+ * What a scorer needs to read. Promoted to the `scope` `CHECK` column;
+ * doc-comment-free per variant.
+ *
+ * An exact-match grader reads only the result's input/output (`result`); the
+ * claw critic reads the whole trajectory (`trace`); a conversation review
+ * reads the transcript (`session`). Declared so a runner (and, later, the
+ * hosted judge) knows what evidence to fetch before scoring.
+ * @export
+ * @enum {string}
+ */
+var EvalScorerScope;
+(function (EvalScorerScope) {
+    EvalScorerScope["Result"] = "result";
+    EvalScorerScope["Session"] = "session";
+    EvalScorerScope["Trace"] = "trace";
+})(EvalScorerScope || (exports.EvalScorerScope = EvalScorerScope = {}));
+function instanceOfEvalScorerScope(value) {
+    for (const key in EvalScorerScope) {
+        if (Object.prototype.hasOwnProperty.call(EvalScorerScope, key)) {
+            if (EvalScorerScope[key] === value) {
+                return true;
+            }
+        }
+    }
+    return false;
+}
+function EvalScorerScopeFromJSON(json) {
+    return EvalScorerScopeFromJSONTyped(json, false);
+}
+function EvalScorerScopeFromJSONTyped(json, ignoreDiscriminator) {
+    return json;
+}
+function EvalScorerScopeToJSON(value) {
+    return value;
+}
+function EvalScorerScopeToJSONTyped(value, ignoreDiscriminator) {
+    return value;
+}

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

@@ -0,0 +1,33 @@
+/**
+ * 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.
+ */
+/**
+ * Which kind an experiment's subject is. Promoted to the `subject_kind`
+ * `CHECK` column; doc-comment-free per variant.
+ *
+ * The wire values are **exactly the kind names**, for the same reason the agent
+ * extension's session pair works this way: `create.pre` passes the
+ * discriminator straight into a kind-scoped registry `get`, so a mismatched
+ * `(subject_kind, subject_id)` pair 404s on its own. `skill` is a member from
+ * day one — evaluating a skill is as first-class as evaluating an agent.
+ * @export
+ * @enum {string}
+ */
+export declare enum EvalSubjectKind {
+    Agent = "agent",
+    Harness = "harness",
+    Skill = "skill"
+}
+export declare function instanceOfEvalSubjectKind(value: any): boolean;
+export declare function EvalSubjectKindFromJSON(json: any): EvalSubjectKind;
+export declare function EvalSubjectKindFromJSONTyped(json: any, ignoreDiscriminator: boolean): EvalSubjectKind;
+export declare function EvalSubjectKindToJSON(value?: EvalSubjectKind | null): any;
+export declare function EvalSubjectKindToJSONTyped(value: any, ignoreDiscriminator: boolean): EvalSubjectKind;

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

@@ -0,0 +1,61 @@
+"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.EvalSubjectKind = void 0;
+exports.instanceOfEvalSubjectKind = instanceOfEvalSubjectKind;
+exports.EvalSubjectKindFromJSON = EvalSubjectKindFromJSON;
+exports.EvalSubjectKindFromJSONTyped = EvalSubjectKindFromJSONTyped;
+exports.EvalSubjectKindToJSON = EvalSubjectKindToJSON;
+exports.EvalSubjectKindToJSONTyped = EvalSubjectKindToJSONTyped;
+/**
+ * Which kind an experiment's subject is. Promoted to the `subject_kind`
+ * `CHECK` column; doc-comment-free per variant.
+ *
+ * The wire values are **exactly the kind names**, for the same reason the agent
+ * extension's session pair works this way: `create.pre` passes the
+ * discriminator straight into a kind-scoped registry `get`, so a mismatched
+ * `(subject_kind, subject_id)` pair 404s on its own. `skill` is a member from
+ * day one — evaluating a skill is as first-class as evaluating an agent.
+ * @export
+ * @enum {string}
+ */
+var EvalSubjectKind;
+(function (EvalSubjectKind) {
+    EvalSubjectKind["Agent"] = "agent";
+    EvalSubjectKind["Harness"] = "harness";
+    EvalSubjectKind["Skill"] = "skill";
+})(EvalSubjectKind || (exports.EvalSubjectKind = EvalSubjectKind = {}));
+function instanceOfEvalSubjectKind(value) {
+    for (const key in EvalSubjectKind) {
+        if (Object.prototype.hasOwnProperty.call(EvalSubjectKind, key)) {
+            if (EvalSubjectKind[key] === value) {
+                return true;
+            }
+        }
+    }
+    return false;
+}
+function EvalSubjectKindFromJSON(json) {
+    return EvalSubjectKindFromJSONTyped(json, false);
+}
+function EvalSubjectKindFromJSONTyped(json, ignoreDiscriminator) {
+    return json;
+}
+function EvalSubjectKindToJSON(value) {
+    return value;
+}
+function EvalSubjectKindToJSONTyped(value, ignoreDiscriminator) {
+    return value;
+}