runbios-sdk 0.2.14-dev.249 → 0.2.14-dev.253

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-dev.249";
+export declare const VERSION = "0.2.14-dev.253";
 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 = '0.2.14-dev.249';
+export const VERSION = '0.2.14-dev.253';
 export class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     models;

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>;
     /**
@@ -795,7 +796,7 @@ 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);
@@ -1100,7 +1101,7 @@ 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

@@ -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;
@@ -3457,7 +3457,7 @@ 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 +3473,11 @@ 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 +3521,10 @@ 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 +3533,11 @@ 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 +4149,7 @@ 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

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "runbios-sdk",
-  "version": "0.2.14-dev.249",
+  "version": "0.2.14-dev.253",
   "description": "Official TypeScript SDK for the Run BiOS training and deployment platform API",
   "type": "module",
   "main": "./dist/index.js",