runbios-sdk 0.2.14 → 0.2.15-dev.256

package/dist/client.js

@@ -340,7 +340,7 @@ export class HttpClient {
     constructor(config) {
         // Default host stays api.runbios.ai for now; cutover to api.runbios.ai is
         // planned once its DNS exists.
-        this.baseUrl = (config.baseUrl || envBaseUrl() || 'https://api.runbios.ai').replace(/\/+$/, '');
+        this.baseUrl = (config.baseUrl || envBaseUrl() || 'https://api-dev.runbios.ai').replace(/\/+$/, '');
         this.apiKey = config.apiKey ?? envApiKey();
         this.accessToken = config.accessToken;
         this.orgId = config.orgId;

package/dist/index.d.ts

@@ -36,7 +36,7 @@ import { Loop } from './resources/loop.js';
  * SDK version. Sent as part of the User-Agent header.
  * Must match package.json "version" -- enforced by a contract test.
  */
-export declare const VERSION = "0.2.14";
+export declare const VERSION = "0.2.15-dev.256";
 export declare class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     readonly models: Models;
@@ -75,7 +75,7 @@ export { Training } from './resources/training.js';
 export { Wallet } from './resources/wallet.js';
 export { GPU, type GPURecommendation } from './resources/gpu.js';
 export { Inference, validateChatRequest, parseSSE, buildInferenceRequest, contextSizingBasis, type ContextSizingBasis, CONTEXT_DEFAULT_CEILING, CONTEXT_EDITABLE_FLOOR, type ChatMessage, type FunctionTool, type ChatCompletionParams, type ChatCompletionResponse, type ChatCompletionChunk, type InferenceVerbParams, type CompletionParams, type EmbeddingParams, type RerankParams, type AnthropicMessageParams, type ServerlessUsageWindow, type ServerlessTimeseriesMetric, type ServerlessWorkspaceLimits, type ServerlessUsageEnvelope, type ServerlessUsageOverviewResponse, type ServerlessUsageRowsResponse, type ServerlessUsageTimeseriesResponse, type ServerlessUsageDailyResponse, type ServerlessSavingsTotals, type ServerlessSavingsPeriod, type ServerlessUsageSavingsResponse, } from './resources/inference.js';
-export type { BiOSConfig, PaginatedResponse, ApiErrorBody, AvailableGpuAlternative, CapacityMinimumRequirement, InferenceBookingAccepted, Model, ModelSearchParams, ModelSearchResponse, ModelDetailResponse, InferenceModel, InferenceModelListResponse, ModelConfig, Dataset, DatasetListParams, DatasetListResponse, DatasetUploadParams, DatasetPreview, DatasetPreviewParams, DatasetImportHFParams, DatasetRegisterHFParams, DatasetHubSearchParams, DatasetHubPreviewParams, DatasetValidation, DatasetFormatVariant, DatasetFormatSpec, DatasetFormatSpecs, DatasetStorageUsage, TrainingMethod, RLHFAlgorithm, AdapterType, TrainingJobStatus, TrainingStatusFilter, TrainingCreateParams, TrainingListParams, TrainingListResponse, TrainingJob, TrainingMetrics, TrainingResourceMetrics, TrainingDeviceMetrics, MetricPoint, MetricGraphConfig, TrainingCheckpoint, TrainingLogs, TrainingLogEntry, TrainingStopResponse, TrainingResumeResponse, CanonicalTrainingRequest, TrainingPreflightDataset, TrainingPreflightWarning, TrainingPreflightResponse, TrainingCapabilityChoice, TrainingConfigFieldCapability, TrainingCapabilities, GPUChoice, WalletBalance, Transaction, TransactionListResponse, TransactionListParams, GPUInfo, GPUPricingResponse, GPUOptionsParams, TrainingRecommendParams, TrainingAdvisorVerdict, TrainingAdvisorRecommendation, TrainingAdvisorUserShape, TrainingAdvisorJustification, TrainingAdvisorBasis, GPUOption, GPUOptionSuggestion, GPUOptionsResponse, InferenceStatus, InferenceCreateParams, InferenceUpdateParams, InferenceDeployment, InferenceDeploymentSummary, InferenceCreateResponse, InferenceListResponse, InferenceUpdateResponse, InferenceLifecycleResponse, InferenceDeleteResponse, InferenceGPUOptionMarket, InferenceGPUOption, InferenceGPUAlternative, InferenceGPUOptionsParams, InferenceGPUOptionsResponse, InferenceCanonicalRequest, InferencePreflightWarning, InferencePreflightResponse, AdapterCompatibility, AdapterCompatibilityResponse, AdapterCompatibilityParams, SupportedArchitecture, ArchitectureScope, SupportedArchitecturesResponse, ApiKey, ApiKeyScope, ApiKeyIntrospection, Organization, OrgMember, OrgInvite, Workspace, Integration, IntegrationCreateParams, StorageUsage, StorageObject, PromptTokensDetails, ChatCompletionUsage, TrainingCadence, TrainingCombinator, ServingKind, TrainingRulePausedReason, LoopTrainingMethod, TrainType, GradersScope, TrainingTrigger, TrainingRecipe, TrainingRunState, TrainingVerdict, TrainingDecision, NotifyState, EvaluationStatus, EvaluationItemStatus, EvaluationWinner, EvaluationWarningCode, ConsentVia, TrainingToolCall, TrainingMessage, TrainingRuleServing, TrainingGPURung, TrainingRule, TrainingRuleConsent, TrainingRulePreflightRefusal, TrainingRuleKeyRef, TrainingRuleKeyCheck, TrainingRulePreflight, TrainingRuleBuildSpec, TrainingRuleTriggerInput, TrainingRuleTrainingInput, TrainingRuleDeployInput, TrainingRuleMoneyInput, TrainingRuleEvaluationInput, TrainingRulePromotionInput, TrainingRuleAcceptTerms, TrainingRulePreflightRequest, TrainingRuleCreateRequest, TrainingRuleUpdateRequest, TrainingRuleConsentRequest, TrainingRuleListParams, TrainingRunPromoteRequest, TrainingRunRejectRequest, TrainingRunRollbackRequest, TrainingRunCancelRequest, TrainingRunListParams, TrainingRun, TrainingRunSummary, TrainingRunEvent, TrainingRunLinks, TrainingRunActions, EvaluationRef, EvaluationDecoding, EvaluationJudgeDimension, EvaluationDimensionScore, EvaluationGraderScore, EvaluationWarning, EvaluationMargin, JudgeAgreementDimension, JudgeAgreement, Evaluation, EvaluationItemSide, EvaluationItem, EvaluationItemListParams, JudgeAgreementParams, AgentSettings, AgentSettingsRequest, BenchmarkStatus, BenchmarkSourceKind, BenchmarkRunStatus, BenchmarkScoring, BenchmarkDimensionScore, Benchmark, BenchmarkItem, BenchmarkRun, BenchmarkHistoryPoint, BenchmarkSource, BenchmarkCreateParams, BenchmarkListParams, BenchmarkItemListParams, BenchmarkHistoryParams, BenchmarkRetireParams, TrainingRuleBenchmarkRequest, InferenceAlias, InferenceAliasRequest, TrainingRuleListResponse, TrainingRuleResponse, TrainingRuleMutationResponse, TrainingRuleDeleteResponse, TrainingRunListResponse, TrainingRunResponse, TrainingRunActionResponse, EvaluationItemsResponse, AgentSettingsResponse, InferenceAliasListResponse, InferenceAliasResponse, InferenceAliasDeleteResponse, BenchmarkListResponse, BenchmarkItemsResponse, BenchmarkHistoryResponse, TrainingRuleRunRefusalCode, TrainingRuleMonthlyLimitRefusal, LoopDatasetDeleteRefusalCode, PipelineStatus, PipelineRecipeExploration, PipelineActiveRun, PipelineVersion, Pipeline, PipelineListResponse, PipelineResponse, } from './types.js';
+export type { BiOSConfig, PaginatedResponse, ApiErrorBody, AvailableGpuAlternative, CapacityMinimumRequirement, InferenceBookingAccepted, Model, ModelSearchParams, ModelSearchResponse, ModelDetailResponse, InferenceModel, InferenceModelListResponse, ModelConfig, Dataset, DatasetListParams, DatasetListResponse, DatasetUploadParams, DatasetPreview, DatasetPreviewParams, DatasetImportHFParams, DatasetRegisterHFParams, DatasetHubSearchParams, DatasetHubPreviewParams, DatasetValidation, DatasetFormatVariant, DatasetFormatSpec, DatasetFormatSpecs, DatasetStorageUsage, TrainingMethod, RLHFAlgorithm, AdapterType, TrainingJobStatus, TrainingStatusFilter, TrainingCreateParams, TrainingListParams, TrainingListResponse, TrainingJob, TrainingMetrics, TrainingResourceMetrics, TrainingDeviceMetrics, MetricPoint, MetricGraphConfig, TrainingCheckpoint, TrainingLogs, TrainingLogEntry, TrainingStopResponse, TrainingResumeResponse, CanonicalTrainingRequest, TrainingPreflightDataset, TrainingPreflightWarning, TrainingPreflightResponse, TrainingCapabilityChoice, TrainingConfigFieldCapability, TrainingCapabilities, GPUChoice, WalletBalance, Transaction, TransactionListResponse, TransactionListParams, GPUInfo, GPUPricingResponse, GPUOptionsParams, TrainingRecommendParams, TrainingAdvisorVerdict, TrainingAdvisorRecommendation, TrainingAdvisorUserShape, TrainingAdvisorJustification, TrainingAdvisorBasis, GPUOption, GPUOptionSuggestion, GPUOptionsResponse, InferenceStatus, InferenceCreateParams, InferenceUpdateParams, InferenceDeployment, InferenceDeploymentSummary, InferenceCreateResponse, InferenceListResponse, InferenceUpdateResponse, InferenceLifecycleResponse, InferenceDeleteResponse, InferenceGPUOptionMarket, InferenceGPUOption, InferenceGPUAlternative, InferenceGPUOptionsParams, InferenceGPUOptionsResponse, InferenceCanonicalRequest, InferencePreflightWarning, InferencePreflightResponse, AdapterCompatibility, AdapterCompatibilityResponse, AdapterCompatibilityParams, SupportedArchitecture, ArchitectureScope, SupportedArchitecturesResponse, ApiKey, ApiKeyScope, ApiKeyIntrospection, Organization, OrgMember, OrgInvite, Workspace, Integration, IntegrationCreateParams, StorageUsage, StorageObject, PromptTokensDetails, ChatCompletionUsage, TrainingCadence, TrainingCombinator, ServingKind, TrainingRulePausedReason, LoopTrainingMethod, TrainType, GradersScope, TrainingTrigger, TrainingRecipe, TrainingRunState, TrainingVerdict, TrainingDecision, NotifyState, EvaluationStatus, EvaluationItemStatus, EvaluationWinner, EvaluationWarningCode, ConsentVia, TrainingToolCall, TrainingMessage, TrainingRuleServing, TrainingGPURung, TrainingRule, TrainingRuleConsent, TrainingRulePreflightRefusal, TrainingRuleKeyRef, TrainingRuleKeyCheck, TrainingRulePreflight, TrainingRuleBuildSpec, TrainingRuleTriggerInput, TrainingRuleTrainingInput, TrainingRuleDeployInput, TrainingRuleMoneyInput, TrainingRuleEvaluationInput, TrainingRulePromotionInput, TrainingRuleAcceptTerms, TrainingRulePreflightRequest, TrainingRuleCreateRequest, TrainingRuleUpdateRequest, TrainingRuleConsentRequest, TrainingRuleListParams, TrainingRunPromoteRequest, TrainingRunRejectRequest, TrainingRunRollbackRequest, TrainingRunCancelRequest, TrainingRunListParams, TrainingRun, TrainingRunSummary, TrainingRunEvent, TrainingRunLinks, TrainingRunActions, EvaluationRef, EvaluationDecoding, EvaluationJudgeDimension, EvaluationDimensionScore, EvaluationGraderScore, EvaluationWarning, EvaluationMargin, JudgeAgreementDimension, JudgeAgreement, Evaluation, EvaluationItemSide, EvaluationItem, EvaluationItemListParams, JudgeAgreementParams, AgentSettings, AgentSettingsRequest, BenchmarkStatus, BenchmarkSourceKind, BenchmarkRunStatus, BenchmarkScoring, BenchmarkDimensionScore, Benchmark, BenchmarkItem, BenchmarkRun, BenchmarkHistoryPoint, BenchmarkSource, BenchmarkCreateParams, BenchmarkListParams, BenchmarkItemListParams, BenchmarkHistoryParams, BenchmarkRetireParams, TrainingRuleBenchmarkRequest, InferenceAlias, InferenceAliasRequest, TrainingRuleListResponse, TrainingRuleResponse, TrainingRuleMutationResponse, TrainingRuleDeleteResponse, TrainingRunListResponse, TrainingRunResponse, TrainingRunActionResponse, EvaluationItemsResponse, AgentSettingsResponse, InferenceAliasListResponse, InferenceAliasResponse, InferenceAliasDeleteResponse, BenchmarkListResponse, BenchmarkItemsResponse, BenchmarkHistoryResponse, TrainingRuleRunRefusalCode, TrainingRuleSaveRefusalCode, TrainingRuleMonthlyLimitRefusal, LoopDatasetDeleteRefusalCode, PipelineStatus, PipelineRecipeExploration, PipelineActiveRun, PipelineVersion, Pipeline, PipelineListResponse, PipelineResponse, } from './types.js';
 /** The closed set of run states a training run never leaves. */
 export { TERMINAL_RUN_STATES } from './types.js';
 /** The most versions a pipeline may be set to make. */

package/dist/index.js

@@ -36,7 +36,7 @@ import { Loop } from './resources/loop.js';
  * SDK version. Sent as part of the User-Agent header.
  * Must match package.json "version" -- enforced by a contract test.
  */
-export const VERSION = '0.2.14';
+export const VERSION = '0.2.15-dev.256';
 export class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     models;

package/dist/resources/inference.js

@@ -336,7 +336,7 @@ export class Inference {
         this.key = config.inferenceKey || envInferenceKey() || envApiKey();
         // Same default host as HttpClient (api.runbios.ai); api.runbios.ai cutover
         // is planned once its DNS exists — update both call sites together.
-        this.baseUrl = (config.baseUrl || envBaseUrl() || 'https://api.runbios.ai').replace(/\/+$/, '');
+        this.baseUrl = (config.baseUrl || envBaseUrl() || 'https://api-dev.runbios.ai').replace(/\/+$/, '');
         this.timeout = config.timeout ?? 900_000;
         this._http = http;
     }

package/dist/resources/loop.d.ts

@@ -296,7 +296,8 @@ export declare class Loop {
      *
      * `min_new_rows` counts only work reviewed SINCE THE LAST BUILD. Counting
      * the whole corpus would fire the rule every interval forever, because a
-     * total that has crossed a threshold stays across it.
+     * total that has crossed a threshold stays across it. It is at least 100,
+     * and 100 when absent: a set built from fewer is too small to learn from.
      */
     createBuildRule(params: LoopBuildRuleParams): Promise<LoopBuildRule>;
     /**
@@ -575,6 +576,10 @@ export declare class Loop {
      * `model_revision` is empty unless training-service actually pinned a
      * commit; test `unreachable.length > 0` to tell a degraded pass from a
      * refusal.
+     *
+     * A `TRAINING_NOT_ENABLED` refusal is neither: training is in beta and this
+     * organisation does not hold the grant. It is in `refusals`, never in
+     * `unreachable`, and nothing in the request can lift it.
      */
     preflightTrainingRule(params: TrainingRulePreflightRequest): Promise<TrainingRulePreflight>;
     /**
@@ -597,6 +602,10 @@ export declare class Loop {
      * `terms_version` they were shown. Do not invent ceilings or price caps on
      * their behalf.
      *
+     * Refused `409 TRAINING_NOT_ENABLED`, with nothing saved, while training is
+     * in beta and this organisation does not hold the grant: do not retry, ask
+     * the platform team to enable it. See `TrainingRuleSaveRefusalCode`.
+     *
      * @example
      * ```ts
      * const estimate = await client.loop.preflightTrainingRule(draft);
@@ -638,6 +647,12 @@ export declare class Loop {
      * `consent_required`, and carries a fresh `preflight` with the new figures.
      * Pass `expected_revision` to be refused with `REVISION_MISMATCH` rather
      * than overwrite an edit somebody else made in the meantime.
+     *
+     * Switching a rule on, switching it into preference training or carrying
+     * `accept_terms` re-runs the preflight, and can be refused
+     * `409 TRAINING_NOT_ENABLED` like a create. A rule paused
+     * `training_not_enabled` is started again by a save of it once the
+     * organisation has been enabled.
      */
     updateTrainingRule(id: string, params: TrainingRuleUpdateRequest): Promise<TrainingRuleMutationResponse>;
     /**
@@ -654,7 +669,8 @@ export declare class Loop {
      * {@link preflightTrainingRule} or from the `preflight` on the update reply,
      * and send the `revision` those figures belong to. A stale revision is
      * refused with `409`, which is the point: it means the amounts moved again
-     * after they were read.
+     * after they were read. `409 TRAINING_NOT_ENABLED` means training is in beta
+     * and this organisation does not hold the grant.
      */
     consentTrainingRule(id: string, params: TrainingRuleConsentRequest): Promise<TrainingRule>;
     /**
@@ -795,7 +811,7 @@ export declare class Loop {
      * replayed against both models on every run that finishes its comparison,
      * and reported as two absolute numbers on a scale that does not move.
      *
-     * 10 to 200 conversations. A named conversation with nothing to ask a model
+     * 20 to 200 conversations. A named conversation with nothing to ask a model
      * is refused rather than skipped, because pinning 47 of the 50 you chose is
      * the set being wrong from the first day and you would never find out.
      *

package/dist/resources/loop.js

@@ -372,7 +372,8 @@ export class Loop {
      *
      * `min_new_rows` counts only work reviewed SINCE THE LAST BUILD. Counting
      * the whole corpus would fire the rule every interval forever, because a
-     * total that has crossed a threshold stays across it.
+     * total that has crossed a threshold stays across it. It is at least 100,
+     * and 100 when absent: a set built from fewer is too small to learn from.
      */
     async createBuildRule(params) {
         const res = await this._http.fetchPost('/api/loop/build-rules', params);
@@ -784,6 +785,10 @@ export class Loop {
      * `model_revision` is empty unless training-service actually pinned a
      * commit; test `unreachable.length > 0` to tell a degraded pass from a
      * refusal.
+     *
+     * A `TRAINING_NOT_ENABLED` refusal is neither: training is in beta and this
+     * organisation does not hold the grant. It is in `refusals`, never in
+     * `unreachable`, and nothing in the request can lift it.
      */
     async preflightTrainingRule(params) {
         return this._http.fetchPost('/api/loop/training-rules/preflight', params);
@@ -808,6 +813,10 @@ export class Loop {
      * `terms_version` they were shown. Do not invent ceilings or price caps on
      * their behalf.
      *
+     * Refused `409 TRAINING_NOT_ENABLED`, with nothing saved, while training is
+     * in beta and this organisation does not hold the grant: do not retry, ask
+     * the platform team to enable it. See `TrainingRuleSaveRefusalCode`.
+     *
      * @example
      * ```ts
      * const estimate = await client.loop.preflightTrainingRule(draft);
@@ -872,6 +881,12 @@ export class Loop {
      * `consent_required`, and carries a fresh `preflight` with the new figures.
      * Pass `expected_revision` to be refused with `REVISION_MISMATCH` rather
      * than overwrite an edit somebody else made in the meantime.
+     *
+     * Switching a rule on, switching it into preference training or carrying
+     * `accept_terms` re-runs the preflight, and can be refused
+     * `409 TRAINING_NOT_ENABLED` like a create. A rule paused
+     * `training_not_enabled` is started again by a save of it once the
+     * organisation has been enabled.
      */
     async updateTrainingRule(id, params) {
         return this._http.fetchPut(`/api/loop/training-rules/${encodeURIComponent(id)}`, params);
@@ -890,7 +905,8 @@ export class Loop {
      * {@link preflightTrainingRule} or from the `preflight` on the update reply,
      * and send the `revision` those figures belong to. A stale revision is
      * refused with `409`, which is the point: it means the amounts moved again
-     * after they were read.
+     * after they were read. `409 TRAINING_NOT_ENABLED` means training is in beta
+     * and this organisation does not hold the grant.
      */
     async consentTrainingRule(id, params) {
         const res = await this._http.fetchPost(`/api/loop/training-rules/${encodeURIComponent(id)}/consent`, params);
@@ -1100,7 +1116,7 @@ export class Loop {
      * replayed against both models on every run that finishes its comparison,
      * and reported as two absolute numbers on a scale that does not move.
      *
-     * 10 to 200 conversations. A named conversation with nothing to ask a model
+     * 20 to 200 conversations. A named conversation with nothing to ask a model
      * is refused rather than skipped, because pinning 47 of the 50 you chose is
      * the set being wrong from the first day and you would never find out.
      *

package/dist/types.d.ts

@@ -8,13 +8,13 @@ export interface BiOSConfig {
     orgId?: string;
     /** Workspace ID. Optional when using API keys (resolved from the key). Can override for multi-workspace keys. */
     workspaceId?: string;
-    /** Base URL for the API. Falls back to the RUNBIOS_BASE_URL environment variable (legacy: BIOS_BASE_URL), then the canonical https://api.runbios.ai hostname. */
+    /** Base URL for the API. Falls back to the RUNBIOS_BASE_URL environment variable (legacy: BIOS_BASE_URL), then the canonical https://api-dev.runbios.ai hostname. */
     baseUrl?: string;
     /** Request timeout in milliseconds. Defaults to 30000. */
     timeout?: number;
     /** Default per-deployment inference key. Can be overridden per inference call. */
     inferenceKey?: string;
-    /** Inference base URL. Defaults to baseUrl, then https://api.runbios.ai. */
+    /** Inference base URL. Defaults to baseUrl, then https://api-dev.runbios.ai. */
     inferenceBaseUrl?: string;
     /** End-to-end inference timeout in milliseconds. Defaults to 15 minutes. */
     inferenceTimeout?: number;
@@ -2704,9 +2704,9 @@ export interface LoopBuildRuleParams {
     method: string;
     /**
      * How many rows reviewed SINCE THE LAST BUILD must exist before this fires
-     * again. Minimum 10. Counting the whole corpus instead would fire the rule
-     * every interval forever, because a total that has crossed a threshold stays
-     * across it.
+     * again. Minimum 100, and 100 when absent. Counting the whole corpus instead
+     * would fire the rule every interval forever, because a total that has
+     * crossed a threshold stays across it.
      */
     min_new_rows?: number;
     enabled?: boolean;
@@ -3146,7 +3146,13 @@ export type TrainingCombinator = 'and' | 'or';
 export type ServingKind = 'serverless_slug' | 'deployment';
 export type TrainingRulePausedReason = 'consent_invalid' | 'insufficient_funds' | 'authorizer_not_member' | 'model_not_trainable' | 'platform_capacity'
 /** A peer was unreachable when the rule was saved, so it has not been validated yet. */
- | 'validation_pending';
+ | 'validation_pending'
+/**
+ * Training is in beta and the rule's organisation does not hold the grant: a
+ * run was refused with the platform's pre-launch gate. Nothing on the rule is
+ * wrong; the platform team enables training, then a member saves the rule.
+ */
+ | 'training_not_enabled';
 /**
  * How a training rule trains, on the wire of the automatic-training routes.
  *
@@ -3457,7 +3463,7 @@ export interface TrainingRuleBuildSpec {
  */
 export interface LoopBuildRuleUpdateParams {
     enabled?: boolean;
-    /** Rows reviewed SINCE THE LAST BUILD before this fires again. Minimum 10. */
+    /** Rows reviewed SINCE THE LAST BUILD before this fires again. Minimum 100. */
     min_new_rows?: number;
     /** The same selection `createDataset` takes, replayed verbatim. */
     spec?: LoopDatasetCreateParams;
@@ -3473,7 +3479,11 @@ export interface TrainingRuleTriggerInput {
     /** null drops the weekday, which is what a weekly rule moved to daily needs. */
     cadence_weekday?: number | null;
     combinator?: TrainingCombinator;
-    /** null removes the row floor, leaving the schedule as the only trigger. */
+    /**
+     * Conversations reviewed since the last version before the rule fires on
+     * rows. At least 100. null removes the row floor, leaving the schedule as the
+     * only trigger.
+     */
     min_new_rows?: number | null;
     /** 1 to 10. Absent leaves it as it is; absent on a create means five. */
     max_versions?: number;
@@ -3517,6 +3527,10 @@ export interface TrainingRuleMoneyInput {
     eval_ceiling_cents?: number;
     /** null removes the monthly cap. Absent leaves it exactly where it stands. */
     monthly_ceiling_cents?: number | null;
+    /**
+     * The most held-back conversations one comparison scores, 10 to 500 and
+     * never below `min_holdout_rows`. Absent on a create means 100.
+     */
     eval_max_rows?: number;
 }
 export interface TrainingRuleEvaluationInput {
@@ -3525,6 +3539,11 @@ export interface TrainingRuleEvaluationInput {
     /** null drops the override, returning to the judge's own advisory model. */
     judge_model?: string | null;
     graders_scope?: GradersScope;
+    /**
+     * The fewest held-back conversations that may decide a comparison. At least
+     * 20, and 20 when absent on a create. A set that would hold back fewer is
+     * not built.
+     */
     min_holdout_rows?: number;
     eval_max_tokens?: number;
 }
@@ -4136,7 +4155,7 @@ export interface BenchmarkSource {
     split?: string;
 }
 /**
- * The pin: 10 to 200 conversations, and a judge with a rubric to measure them.
+ * The pin: 20 to 200 conversations, and a judge with a rubric to measure them.
  *
  * A conversation with nothing to ask a model is refused rather than skipped.
  * Pinning 47 of the 50 somebody chose is the set being wrong from the first
@@ -4216,6 +4235,25 @@ export interface TrainingRuleBenchmarkRequest {
  *   `train_rows`, `holdout_rows` and the `floor` it needed.
  */
 export type TrainingRuleRunRefusalCode = 'CONSENT_REQUIRED' | 'RULE_PAUSED' | 'RUN_ACTIVE' | 'VERSION_LIMIT_REACHED' | 'MONTHLY_LIMIT_REACHED' | 'AGENT_OFF' | 'NOT_ENOUGH_ROWS';
+/**
+ * The machine codes a SAVE of a training rule is refused with when the
+ * preflight it runs refuses: `createTrainingRule`, an `updateTrainingRule`
+ * that switches a rule on or into preference training or carries
+ * `accept_terms`, and `consentTrainingRule`. Every other preflight refusal is
+ * a `400` carrying the peer's own sentence; `preflightTrainingRule` answers
+ * `200` with the same codes in `refusals`.
+ *
+ * - `TRAINING_NOT_ENABLED` (409): training is in beta and this organisation
+ *   does not hold the grant. Nothing is saved and nothing in the request can
+ *   lift it, so do not retry: ask the platform team to enable training for
+ *   the organisation. It is answered ahead of every other refusal. A rule that
+ *   meets it at fire time is paused `training_not_enabled` instead (see
+ *   {@link TrainingRulePausedReason}).
+ * - `METHOD_NOT_ENABLED` (422): a training method the platform has not enabled.
+ * - `CEILING_TOO_LOW` (422): a ceiling cannot buy the minimum hours at the
+ *   accepted price cap.
+ */
+export type TrainingRuleSaveRefusalCode = 'TRAINING_NOT_ENABLED' | 'METHOD_NOT_ENABLED' | 'CEILING_TOO_LOW';
 /**
  * The machine codes `deleteDataset` can be refused with.
  *

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "runbios-sdk",
-  "version": "0.2.14",
+  "version": "0.2.15-dev.256",
   "description": "Official TypeScript SDK for the Run BiOS training and deployment platform API",
   "type": "module",
   "main": "./dist/index.js",