runbios-sdk 0.3.0-dev.327 → 0.3.0-dev.332

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 = "0.3.0-dev.327";
+export declare const VERSION = "0.3.0-dev.332";
 export declare class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     readonly models: Models;

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 = '0.3.0-dev.327';
+export const VERSION = '0.3.0-dev.332';
 export class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     models;

package/dist/resources/loop.d.ts

@@ -382,7 +382,8 @@ export declare class Loop {
      * it made, winner or not, each the shape `attempts` has on the pipeline
      * (with its `explanation` in plain words). `limit` is 50 when absent and at
      * most 200. Narrow by `result` (`became_version`, `not_better`,
-     * `inconclusive`, `failed`, `stopped`, `running` -- which includes one
+     * `inconclusive` -- tested, could not decide --, `not_tested` -- ended
+     * before it was tested --, `failed`, `stopped`, `running` -- which includes one
      * waiting for a decision -- or `waiting_for_review`) and by `model` (`main`,
      * or `second` for the second model's attempts); `total` follows the
      * filters.

package/dist/resources/loop.js

@@ -633,7 +633,8 @@ export class Loop {
      * it made, winner or not, each the shape `attempts` has on the pipeline
      * (with its `explanation` in plain words). `limit` is 50 when absent and at
      * most 200. Narrow by `result` (`became_version`, `not_better`,
-     * `inconclusive`, `failed`, `stopped`, `running` -- which includes one
+     * `inconclusive` -- tested, could not decide --, `not_tested` -- ended
+     * before it was tested --, `failed`, `stopped`, `running` -- which includes one
      * waiting for a decision -- or `waiting_for_review`) and by `model` (`main`,
      * or `second` for the second model's attempts); `total` follows the
      * filters.

package/dist/types.d.ts

@@ -2967,9 +2967,22 @@ export interface LoopMetricPoint {
     judge_candidate_mean: number | null;
     /** The trainer's loss on its validation slice, as {@link Evaluation.trainer_eval_loss}. */
     trainer_eval_loss: number | null;
-    /** The bar this attempt had to clear, frozen with its report. */
+    /**
+     * The pipeline's margin frozen with this attempt's report: the most the
+     * mean-change bar asks. The bar it actually had to clear is
+     * {@link LoopMetricPoint.mean_change_needed}.
+     */
     promote_margin: number | null;
     promote_min_win_rate: number | null;
+    /**
+     * The average score change this attempt had to clear, frozen with its
+     * report: `promote_margin`, lowered toward a quarter of the room left above
+     * what it was compared with near the top, never below what the comparison
+     * can tell from noise ({@link EvaluationMargin.mean_change_needed}). On a
+     * report written before that bar existed, or one that compared nothing, it
+     * is `promote_margin`. Null when the attempt has no report.
+     */
+    mean_change_needed: number | null;
     checkpoint_id: string | null;
     checkpoint_step: number | null;
     attempt_no: number | null;
@@ -3516,6 +3529,14 @@ export interface EvaluationMargin {
     promote_margin: number;
     promote_min_win_rate: number;
     min_holdout_rows: number;
+    /**
+     * The average score change this comparison actually asked of the new model:
+     * `promote_margin`, lowered toward a quarter of the room left above what it
+     * was compared with once that is smaller, never below what the
+     * comparison's size can tell from noise. Null when nothing was compared,
+     * and on a report written before, whose bar was `promote_margin`.
+     */
+    mean_change_needed: number | null;
 }
 /** The comparison report: the candidate against what serves today, same rows. */
 export interface Evaluation {
@@ -3788,15 +3809,30 @@ export interface BenchmarkDimensionScore {
 /**
  * What a pipeline's history, or a benchmark's pinned set, takes up in storage
  * right now. Conversations, answers and scoring, and the training rows of a
- * pipeline's curated sets, are kept in storage only, never in the database. Not billed; it counts toward the workspace's storage
- * quota and is deleted with the pipeline (a benchmark's with the workspace).
- * `where` is null while nothing is stored.
+ * pipeline's curated sets, are kept in storage only, never in the database.
+ * Only the part of a pipeline's figure kept in the platform's storage is ever
+ * billed, as Conscious Loop storage at the storage rate with no daily minimum
+ * (each day's cost is added up and charged once it reaches a whole cent), and
+ * only where `billed` says so; deleting the pipeline stops it. The part in the
+ * workspace's own storage and the index in the database never are, and
+ * neither is a benchmark's pinned set, which is kept until the workspace is
+ * deleted (`billed` is always false on a benchmark). It counts toward the
+ * workspace's storage quota and is deleted with the pipeline (a benchmark's
+ * with the workspace). `where` is null while nothing is stored.
  */
 export type LoopStorageWhere = 'platform' | 'your_storage' | 'both';
 export interface LoopStorage {
     bytes: number;
     objects: number;
     where: LoopStorageWhere | null;
+    /**
+     * True when part of a pipeline's figure is in the platform's storage and
+     * this environment charges for that; false while nothing of it is charged
+     * (all of it in the workspace's own storage, nothing stored, an environment
+     * that only measures it for now, or a benchmark's pinned set, which is never
+     * billed).
+     */
+    billed: boolean;
 }
 /**
  * The frozen definition: the oracle, the scoring rule and the fingerprint of
@@ -4357,8 +4393,12 @@ export interface PipelineVersion {
  * learned from (`rebuild`).
  */
 export type PipelineAttemptTrigger = 'data' | 'schedule' | 'manual' | 'challenger' | 'rebuild';
-/** What became of an attempt. */
-export type PipelineAttemptOutcome = 'became_version' | 'not_better' | 'inconclusive' | 'stopped' | 'failed' | 'running' | 'waiting_for_review';
+/**
+ * What became of an attempt. `inconclusive` is a held-back test that ran and
+ * could not decide; `not_tested` is an attempt that ended before it was tested
+ * (its comparison machine never started, or it expired first).
+ */
+export type PipelineAttemptOutcome = 'became_version' | 'not_better' | 'inconclusive' | 'not_tested' | 'stopped' | 'failed' | 'running' | 'waiting_for_review';
 /** One attempt: every training run a pipeline makes, winner or not. */
 export interface PipelineAttempt {
     attempt: number;
@@ -4882,7 +4922,7 @@ export interface PipelineResponse {
     pipeline: Pipeline;
 }
 /** What became of an attempt, as `listPipelineAttempts` filters it. */
-export type PipelineAttemptResultFilter = 'became_version' | 'not_better' | 'inconclusive' | 'failed' | 'stopped' | 'running' | 'waiting_for_review';
+export type PipelineAttemptResultFilter = 'became_version' | 'not_better' | 'inconclusive' | 'not_tested' | 'failed' | 'stopped' | 'running' | 'waiting_for_review';
 /** Which model's attempts: the pipeline's own (`main`) or its second model's (`second`). */
 export type PipelineAttemptModelFilter = 'main' | 'second';
 export interface PipelineAttemptListParams {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "runbios-sdk",
-  "version": "0.3.0-dev.327",
+  "version": "0.3.0-dev.332",
   "description": "Official TypeScript SDK for the Run BiOS training and deployment platform API",
   "type": "module",
   "main": "./dist/index.js",