@tangle-network/agent-eval 0.110.0 → 0.111.0

package/dist/meta-eval/index.d.ts

@@ -3,7 +3,7 @@ export { D as DeploymentOutcome, F as FileSystemOutcomeStore, a as FileSystemOut
 export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from '../rubric-predictive-validity-C-fMteAW.js';
 import { C as ContinuousAgreement, a as CalibrationResult, b as ContinuousCalibrationResult, c as CandidateScore, G as GoldenItem } from '../judge-calibration-7C-IDmKr.js';
 import { S as SeriesConvergenceOptions, a as SeriesConvergenceResult } from '../series-convergence-D5OWMBg6.js';
-import { C as CorpusAgreementReport } from '../statistics-D88peojY.js';
+import { C as CorpusAgreementReport } from '../statistics-CDfpwIdX.js';
 import '../store-BsVi7ncX.js';
 import '../schema-SGWcK9wa.js';
 import '../run-record-DksGsfgv.js';

package/dist/meta-eval/index.js

@@ -11,11 +11,11 @@ import {
 } from "../chunk-3RF76KTD.js";
 import {
   rubricPredictiveValidity
-} from "../chunk-X6NIXVOD.js";
+} from "../chunk-3LDPUYAC.js";
 import {
   pearsonR,
   spearmanR
-} from "../chunk-XMSYF4A7.js";
+} from "../chunk-VC43KQCK.js";
 import {
   aggregateLlm,
   llmSpans

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.110.0",
+    "version": "0.111.0",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.0.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/pipelines/index.js

@@ -4,11 +4,11 @@ import {
   classifyFailure,
   compareToBaseline,
   computeToolUseMetrics
-} from "../chunk-PLOMR3HP.js";
+} from "../chunk-YWGKQARF.js";
 import {
   interRaterReliability,
   pearsonR
-} from "../chunk-XMSYF4A7.js";
+} from "../chunk-VC43KQCK.js";
 import {
   aggregateLlm,
   argHash,

package/dist/{provenance-DMvsfknv.d.ts → provenance-KhY8ESVM.d.ts}

@@ -1,7 +1,7 @@
 import { S as Scenario, g as Gate, G as GateResult, n as GateContext, C as CampaignResult, p as Mutator, f as SurfaceProposer, M as MutableSurface, h as GateDecision } from './types-v--ctu-b.js';
-import { e as RedTeamCase, D as Direction, b as RunCampaignOptions } from './gepa-BUNP3606.js';
+import { e as RedTeamCase, D as Direction, b as RunCampaignOptions } from './gepa-T8T215nw.js';
 import { R as RunRecord } from './run-record-DksGsfgv.js';
-import { a as PairedBootstrapResult } from './statistics-D88peojY.js';
+import { a as PairedBootstrapResult } from './statistics-CDfpwIdX.js';
 import { HostedClient, TraceSpanEvent } from './hosted/index.js';
 import { C as CampaignStorage } from './storage-Dw_f7WMt.js';
 

package/dist/reporting.d.ts

@@ -1,7 +1,7 @@
 export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from './rubric-predictive-validity-C-fMteAW.js';
 export { B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, f as ReleaseConfidenceScorecard, g as ReleaseConfidenceStatus, h as ReleaseConfidenceThresholds, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-oBfOz8ku.js';
 export { I as InterimReleaseConfidence, a as InterimReleaseConfidenceInput, P as PairedEvalueOptions, b as PairedEvalueSequence, c as PairedEvalueStep, S as SequentialDecision, e as evaluateInterimReleaseConfidence, p as pairedEvalueSequence } from './sequential-5iSVfzl2.js';
-export { P as PairedBootstrapOptions, a as PairedBootstrapResult, b as benjaminiHochberg, p as pairedBootstrap, w as wilcoxonSignedRank } from './statistics-D88peojY.js';
+export { P as PairedBootstrapOptions, a as PairedBootstrapResult, b as benjaminiHochberg, p as pairedBootstrap, w as wilcoxonSignedRank } from './statistics-CDfpwIdX.js';
 export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-Bz-0-t8v.js';
 import './run-record-DksGsfgv.js';
 import '@tangle-network/agent-interface';

package/dist/reporting.js

@@ -4,10 +4,10 @@ import {
   evaluateReleaseConfidence,
   judgeReplayGate,
   renderReleaseReport
-} from "./chunk-6HFYGEZ3.js";
+} from "./chunk-RT2AFUXM.js";
 import {
   rubricPredictiveValidity
-} from "./chunk-X6NIXVOD.js";
+} from "./chunk-3LDPUYAC.js";
 import {
   evaluateInterimReleaseConfidence,
   pairedEvalueSequence
@@ -18,12 +18,12 @@ import {
   paretoChart,
   researchReport,
   summaryTable
-} from "./chunk-6MFFKPZ4.js";
+} from "./chunk-EMKORATZ.js";
 import {
   benjaminiHochberg,
   pairedBootstrap,
   wilcoxonSignedRank
-} from "./chunk-XMSYF4A7.js";
+} from "./chunk-VC43KQCK.js";
 import "./chunk-VSMTAMNK.js";
 import "./chunk-ONWEPEDO.js";
 import "./chunk-PZ5AY32C.js";

package/dist/rl.js

@@ -10,24 +10,24 @@ import {
 } from "./chunk-3RF76KTD.js";
 import {
   runEvalCampaign
-} from "./chunk-6PL5MGDL.js";
+} from "./chunk-WZOIFFG2.js";
 import {
   detectRewardHacking,
   extractVerifiableReward,
   extractVerifiableRewardsFromRecords,
   filterDeterministicallyRewarded
-} from "./chunk-N22ZO7FV.js";
+} from "./chunk-7QPK6W7R.js";
 import {
   rubricPredictiveValidity
-} from "./chunk-X6NIXVOD.js";
+} from "./chunk-3LDPUYAC.js";
 import {
   evaluateInterimReleaseConfidence
 } from "./chunk-MAZ26DC7.js";
-import "./chunk-6MFFKPZ4.js";
+import "./chunk-EMKORATZ.js";
 import {
   benjaminiHochberg,
   wilcoxonSignedRank
-} from "./chunk-XMSYF4A7.js";
+} from "./chunk-VC43KQCK.js";
 import {
   observationsFromRunRecords,
   thompsonCurriculum,

package/dist/{run-campaign-PNHAJIT2.js → run-campaign-HJQEDCJQ.js}

@@ -1,12 +1,12 @@
 import {
   planCampaignRun,
   runCampaign
-} from "./chunk-3PFZBGMR.js";
-import "./chunk-XMSYF4A7.js";
+} from "./chunk-XNCF3JU3.js";
+import "./chunk-VC43KQCK.js";
 import "./chunk-ONWEPEDO.js";
 import "./chunk-PZ5AY32C.js";
 export {
   planCampaignRun,
   runCampaign
 };
-//# sourceMappingURL=run-campaign-PNHAJIT2.js.map
+//# sourceMappingURL=run-campaign-HJQEDCJQ.js.map

package/dist/{statistics-D88peojY.d.ts → statistics-CDfpwIdX.d.ts}

@@ -253,6 +253,18 @@ declare function bonferroni(pValues: number[], alpha?: number): {
     adjusted: number[];
     significant: boolean[];
 };
+/**
+ * Holm step-down family-wise error adjustment.
+ *
+ * P-values are sorted from smallest to largest, multiplied by their remaining
+ * hypothesis count, and made monotonically non-decreasing before being mapped
+ * back to input order. This uniformly dominates plain Bonferroni while keeping
+ * strong family-wise error control under arbitrary dependence.
+ */
+declare function holm(pValues: readonly number[], alpha?: number): {
+    adjusted: number[];
+    significant: boolean[];
+};
 /**
  * Benjamini–Hochberg false discovery rate. Returns adjusted q-values and
  * significance at the target FDR; handles ties and preserves q monotonicity.
@@ -446,4 +458,4 @@ declare function eProcess(opts?: EProcessOptions): EProcess;
  *  gate verdicts is non-reproducible by construction). */
 declare function mulberry32(seed: number): () => number;
 
-export { mulberry32 as A, normalizeScores as B, type CorpusAgreementReport as C, pairedMde as D, type EProcessState as E, pairedRiskDifference as F, pairedTTest as G, partialCredit as H, passAtK as I, pearsonR as J, ranks as K, requiredSampleSize as L, type McNemarResult as M, spearmanR as N, weightedComposite as O, type PairedBootstrapOptions as P, weightedMean as Q, type RiskDifferenceResult as R, wilson as S, type WeightedCompositeInput as W, type PairedBootstrapResult as a, benjaminiHochberg as b, type CliffsMagnitude as c, type CorpusAgreementOptions as d, type CorpusAgreementPerDimension as e, type CorpusScoreRecord as f, type EProcess as g, type EProcessOptions as h, type EProcessStep as i, type ProportionInterval as j, type WeightedCompositeResult as k, bonferroni as l, cliffsDelta as m, cohensD as n, confidenceInterval as o, pairedBootstrap as p, corpusInterRaterAgreement as q, corpusInterRaterAgreementFromJudgeScores as r, eProcess as s, interRaterReliability as t, interpretCliffs as u, mannWhitneyU as v, wilcoxonSignedRank as w, mcnemar as x, mcnemarPower as y, mcnemarRequiredN as z };
+export { mcnemarRequiredN as A, mulberry32 as B, type CorpusAgreementReport as C, normalizeScores as D, type EProcessState as E, pairedMde as F, pairedRiskDifference as G, pairedTTest as H, partialCredit as I, passAtK as J, pearsonR as K, ranks as L, type McNemarResult as M, requiredSampleSize as N, spearmanR as O, type PairedBootstrapOptions as P, weightedComposite as Q, type RiskDifferenceResult as R, weightedMean as S, wilson as T, type WeightedCompositeInput as W, type PairedBootstrapResult as a, benjaminiHochberg as b, type CliffsMagnitude as c, type CorpusAgreementOptions as d, type CorpusAgreementPerDimension as e, type CorpusScoreRecord as f, type EProcess as g, type EProcessOptions as h, type EProcessStep as i, type ProportionInterval as j, type WeightedCompositeResult as k, bonferroni as l, cliffsDelta as m, cohensD as n, confidenceInterval as o, pairedBootstrap as p, corpusInterRaterAgreement as q, corpusInterRaterAgreementFromJudgeScores as r, eProcess as s, holm as t, interRaterReliability as u, interpretCliffs as v, wilcoxonSignedRank as w, mannWhitneyU as x, mcnemar as y, mcnemarPower as z };

package/docs/improvement-glossary.md

@@ -61,25 +61,26 @@ Start at the top row and only move down when the row's *"reach for it when"* mat
 | `gepaProposer` | You want the strong default: reflective full-surface prompt rewrites, grounded in findings, keeping a Pareto frontier of complementary winners. | prompt string | **Yes — `improve({ surface: 'prompt' })` default.** Proven live. |
 | `skillOptProposer` | You are editing a structured `SKILL.md`/runbook and want small anchored add/delete/replace patches that preserve earlier rules. | skill/prompt string | **Yes — `improve({ surface: 'skills' })` default.** Not yet proven live. |
 | `parameterSweepProposer` | The likely fix is a config knob, not words — `retrieval.k`, `temperature`, `max_tokens`. You give it candidate patches; it applies them to a JSON surface. | JSON config string | Yes, but you supply the candidate list. |
-| `fapoProposer` | You want *evidence to decide when to escalate*: try prompt edits first, move to parameters, then to structural code — one scoped change per cycle, only escalating when the cheaper level is exhausted. **This is the composite/orchestration proposer** (there is no separate `compositeProposer`). | whatever its level proposers return | Exported; you wire the level proposers. |
+| `fapoProposer` | You want *evidence to decide when to escalate*: try prompt edits first, move to parameters, then to structural code — one scoped change per cycle, only escalating when the cheaper level is exhausted. | whatever its level proposers return | Exported; you wire the level proposers. |
+| `compositeProposer` | You want several proposers to share one candidate-generation budget in the same round. It allocates the population by declared weights, preserves member provenance, deduplicates surfaces, and isolates a member failure unless every member fails. | whatever its member proposers return | Exported; you wire the member proposers. |
 | `aceProposer` | You are accumulating hard-won lessons into a playbook and must **never** summarize an old lesson away (append-only, provenance-tagged). | playbook string | Exported. |
 | `memoryCurationProposer` | Same as ACE but you want a compact, deduped, re-ranked memory instead of append-only growth. | memory string | Exported. |
 | `evolutionaryProposer` | You want blind population search (mutate → measure → select) with no reflection over findings — a cheap control or a baseline to beat. | any string | Exported. |
 | `traceAnalystProposer` | Bench-only: race our trace-analysis evidence engine head-to-head inside `compareProposers`. | prompt string | Bench-only. |
 | `haloProposer` | Bench-only: race the external `halo-engine` analysis against ours. | prompt string | Bench-only, external. |
 
-Default path: `gepaProposer` for prompts; add `parameterSweepProposer` when a config knob is the suspect; wrap both in `fapoProposer` when you want the loop to decide *when* to escalate from words to knobs to code.
+Default path: `gepaProposer` for prompts; add `parameterSweepProposer` when a config knob is the suspect; wrap levels in `fapoProposer` when the loop should decide *when* to escalate, or use `compositeProposer` when multiple proposer families must split one fixed population budget.
 
-## Composing proposers — the only three ways, with runnable code
+## Composing proposers — four distinct shapes
 
-There is deliberately **no** `chainProposer`/`compositeProposer` primitive.
-Composition happens in exactly three shapes:
+Choose the shape that matches the experiment:
 
-1. **Escalate** — `fapoProposer` wraps a prompt + parameter + structural proposer into one, spending the cheapest level first.
-2. **Race** — `compareProposers` runs N proposers on one holdout and returns a per-proposer lift CI + pairwise "who won".
-3. **Plug in** — hand any composed proposer to the gated loop via `runImprovementLoop({ proposer })`, or the one-liner `improve({ surface, generator })` in `@tangle-network/agent-runtime`.
+1. **Portfolio** — `compositeProposer` splits one generation's population across member proposers by fixed weights and returns one provenance-labelled pool.
+2. **Escalate** — `fapoProposer` wraps prompt + parameter + structural levels into one proposer and spends on the cheapest level until evidence says to escalate.
+3. **Race** — `compareProposers` gives proposers separate loops, then re-scores their winners on one holdout and returns per-proposer lift intervals plus pairwise results.
+4. **Plug in** — hand any proposer to `runImprovementLoop({ proposer })`, or use `improve({ surface, generator })` in `@tangle-network/agent-runtime`.
 
-### 1 + 3 — compose by escalation, then run the gated loop
+### 2 + 4 — compose by escalation, then run the improvement loop
 
 ```ts
 import {
@@ -93,8 +94,8 @@ import {
 const llm = { baseUrl: process.env.TANGLE_BASE_URL, apiKey: process.env.TANGLE_API_KEY }
 const model = 'deepseek-v4-flash'
 
-// Compose: prompt edits first (GEPA), escalate to a config knob only when
-// prompt-level search plateaus. `fapoProposer` IS the composite proposer.
+// Compose by escalation: prompt edits first (GEPA), then a config knob only
+// when prompt-level search plateaus. This is distinct from a peer portfolio.
 const proposer = fapoProposer({
   scope: { allowedLevels: ['prompt', 'parameter'] },   // no structural/code tier here
   promptProposer: gepaProposer({ llm, model, target: 'agent system prompt' }),
@@ -147,7 +148,7 @@ const out = await improve(profile, findings, {
 if (out.shipped) deploy(out.profile)  // out.lift is the held-out winner − baseline
 ```
 
-### 2 — race proposers head-to-head for a lift CI
+### 3 — race proposers head-to-head for a lift CI
 
 ```ts
 import {
@@ -196,7 +197,7 @@ Every entrant is re-scored on the **same** holdout with the **same** judges, so
 - Do not put eval logic inside a proposer — scoring lives in `dispatch` + `judges`, proposing lives in the proposer.
 - Do not let a proposer read held-out judge scores — `ProposeContext` makes that a compile error on purpose; a proposer that games the acceptance axis is an oracle, not an optimizer.
 - Do not read `lift` without `result.power`/MDE — a "+4" on a valset too small to detect +4 is noise wearing a number.
-- Do not look for a `compositeProposer` — `fapoProposer` is the composite; `compareProposers` is the race; `runImprovementLoop`/`improve` is the plug.
+- Do not confuse `compositeProposer` with `fapoProposer`: the former allocates one fixed population across peers, while the latter escalates through ordered levels from cheaper to more structural changes.
 
 ### neutralizationGate — the placebo / content-causality control
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.110.0",
+  "version": "0.111.0",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {