@tangle-network/agent-eval 0.199.0 → 0.200.0

package/dist/rollout/index.d.ts

@@ -1,5 +1,5 @@
 import { C as assertMinted, D as isRolloutLine, E as gateGamedOutcome, O as isTrainableSplit, S as ToolDef, T as assertRolloutLine, _ as RolloutProvenance, a as MintedRolloutLine, b as RolloutStep, c as ROLLOUT_ROLES, d as RolloutArtifacts, f as RolloutCapture, g as RolloutPolicy, h as RolloutOutcome, i as GatedEvidence, k as validateRolloutLine, l as ROLLOUT_SCHEMA, m as RolloutLine, n as ChatRole, o as MintedRolloutOutcome, p as RolloutCostBlock, r as ChatToolCall, s as ROLLOUT_CAPTURES, t as ChatMessage, u as ROLLOUT_SPLITS, v as RolloutRole, w as assertMintedLines, x as RolloutTask, y as RolloutSplit } from "../schema-B5LtYvn9.js";
 import { a as mintRolloutRows, c as ScorePreference, d as observedSplitScore, f as scoreOrigin, i as RolloutSearchLineage, l as isRealnessGated, m as trainingScore, n as MintRolloutResult, o as unmintableReasons, p as trainingReward, r as RolloutScrubber, s as ScoreOrigin, t as MintRolloutOptions, u as observedScore } from "../mint-C9Lojt1V.js";
 import { a as SftRow, c as VerifiersTokenUsage, d as toRftItem, f as toRftItems, h as toVerifiersRolloutOutputs, i as SftExportOptions, l as toJsonl, m as toVerifiersRolloutOutput, n as RewardRow, p as toSftRows, r as RftItem, s as VerifiersRolloutOutput, t as RealnessLabels, u as toRewardRows } from "../exporters-XMCYJlqT.js";
-import { $ as HarborFinalMetrics, A as measureFormatGate, B as DEFAULT_CLAUDE_PROJECTS_DIR, C as FORMAT_GATE_DISPOSITION, Ct as gatedEvidenceOf, D as ReleaseRowRef, E as GateReport, F as openOpencodeDb, G as readRolloutJournal, H as findClaudeTranscripts, I as readOpencodeSessionMessages, J as ATIF_SCHEMA_VERSION, K as readRolloutLedger, L as ClaudeTranscript, M as DEFAULT_OPENCODE_DB, N as OpencodeSessionRow, O as assertGateReport, P as findOpencodeSessionsByDirectory, Q as HarborContentPart, R as ClaudeTranscriptRef, S as EmittedEvidence, St as gateErrors, T as GateDisposition, U as readClaudeTranscript, V as claudeProjectSlug, W as appendRolloutLines, X as HARBOR_IMPORT_GAP, Y as FromHarborOptions, Z as HarborAgent, _ as defaultRolloutScrubber, _t as GateCheckDisposition, a as buildHfDataset, at as HarborStepSource, b as scrubRolloutLine, bt as GateEntryPoint, c as runRolloutReleaseCli, ct as HarborTrajectory, d as RELEASE_FORMATS, dt as toHarborTrajectories, et as HarborImageSource, f as ReleaseFormat, ft as toHarborTrajectory, g as addScrubCounts, gt as GateCheck, h as ScrubRule, ht as GATE_POLICIES, i as ScrubReport, it as HarborStep, j as releaseRowRefs, k as gatedRolloutIds, l as DatasetCardInputs, lt as fromHarborTrajectory, m as ScrubCounts, mt as GATE_CHECK_IDS, n as BuildSummary, nt as HarborObservation, o as parseRolloutReleaseArgs, ot as HarborSubagentTrajectoryRef, p as buildDatasetCard, pt as GATE_CHECKS, q as writeRolloutLedger, r as RolloutReleaseCliArgs, rt as HarborObservationResult, s as planPushCommand, st as HarborToolCall, t as BuildOptions, tt as HarborMetrics, u as FORMAT_FILES, ut as relabelImportedSplit, v as emptyScrubCounts, vt as GateCheckId, w as FormatGateCounts, x as scrubText, xt as GatePolicy, y as scrubLines, yt as GateCheckedOutcome, z as ClaudeUsageTotals } from "../index-D4pC7Tvd.js";
+import { $ as HarborFinalMetrics, A as measureFormatGate, B as DEFAULT_CLAUDE_PROJECTS_DIR, C as FORMAT_GATE_DISPOSITION, Ct as gatedEvidenceOf, D as ReleaseRowRef, E as GateReport, F as openOpencodeDb, G as readRolloutJournal, H as findClaudeTranscripts, I as readOpencodeSessionMessages, J as ATIF_SCHEMA_VERSION, K as readRolloutLedger, L as ClaudeTranscript, M as DEFAULT_OPENCODE_DB, N as OpencodeSessionRow, O as assertGateReport, P as findOpencodeSessionsByDirectory, Q as HarborContentPart, R as ClaudeTranscriptRef, S as EmittedEvidence, St as gateErrors, T as GateDisposition, U as readClaudeTranscript, V as claudeProjectSlug, W as appendRolloutLines, X as HARBOR_IMPORT_GAP, Y as FromHarborOptions, Z as HarborAgent, _ as defaultRolloutScrubber, _t as GateCheckDisposition, a as buildHfDataset, at as HarborStepSource, b as scrubRolloutLine, bt as GateEntryPoint, c as runRolloutReleaseCli, ct as HarborTrajectory, d as RELEASE_FORMATS, dt as toHarborTrajectories, et as HarborImageSource, f as ReleaseFormat, ft as toHarborTrajectory, g as addScrubCounts, gt as GateCheck, h as ScrubRule, ht as GATE_POLICIES, i as ScrubReport, it as HarborStep, j as releaseRowRefs, k as gatedRolloutIds, l as DatasetCardInputs, lt as fromHarborTrajectory, m as ScrubCounts, mt as GATE_CHECK_IDS, n as BuildSummary, nt as HarborObservation, o as parseRolloutReleaseArgs, ot as HarborSubagentTrajectoryRef, p as buildDatasetCard, pt as GATE_CHECKS, q as writeRolloutLedger, r as RolloutReleaseCliArgs, rt as HarborObservationResult, s as planPushCommand, st as HarborToolCall, t as BuildOptions, tt as HarborMetrics, u as FORMAT_FILES, ut as relabelImportedSplit, v as emptyScrubCounts, vt as GateCheckId, w as FormatGateCounts, x as scrubText, xt as GatePolicy, y as scrubLines, yt as GateCheckedOutcome, z as ClaudeUsageTotals } from "../index-D4pC7Tvd2.js";
 export { ATIF_SCHEMA_VERSION, type BuildOptions, type BuildSummary, type ChatMessage, type ChatRole, type ChatToolCall, type ClaudeTranscript, type ClaudeTranscriptRef, type ClaudeUsageTotals, DEFAULT_CLAUDE_PROJECTS_DIR, DEFAULT_OPENCODE_DB, type DatasetCardInputs, type EmittedEvidence, FORMAT_FILES, FORMAT_GATE_DISPOSITION, type FormatGateCounts, type FromHarborOptions, GATE_CHECKS, GATE_CHECK_IDS, GATE_POLICIES, type GateCheck, type GateCheckDisposition, type GateCheckId, type GateCheckedOutcome, type GateDisposition, type GateEntryPoint, type GatePolicy, type GateReport, type GatedEvidence, HARBOR_IMPORT_GAP, type HarborAgent, type HarborContentPart, type HarborFinalMetrics, type HarborImageSource, type HarborMetrics, type HarborObservation, type HarborObservationResult, type HarborStep, type HarborStepSource, type HarborSubagentTrajectoryRef, type HarborToolCall, type HarborTrajectory, type MintRolloutOptions, type MintRolloutResult, type MintedRolloutLine, type MintedRolloutOutcome, type OpencodeSessionRow, RELEASE_FORMATS, ROLLOUT_CAPTURES, ROLLOUT_ROLES, ROLLOUT_SCHEMA, ROLLOUT_SPLITS, type RealnessLabels, type ReleaseFormat, type ReleaseRowRef, type RewardRow, type RftItem, type RolloutArtifacts, type RolloutCapture, type RolloutCostBlock, type RolloutLine, type RolloutOutcome, type RolloutPolicy, type RolloutProvenance, type RolloutReleaseCliArgs, type RolloutRole, type RolloutScrubber, type RolloutSearchLineage, type RolloutSplit, type RolloutStep, type RolloutTask, type ScoreOrigin, type ScorePreference, type ScrubCounts, type ScrubReport, type ScrubRule, type SftExportOptions, type SftRow, type ToolDef, type VerifiersRolloutOutput, type VerifiersTokenUsage, addScrubCounts, appendRolloutLines, assertGateReport, assertMinted, assertMintedLines, assertRolloutLine, buildDatasetCard, buildHfDataset, claudeProjectSlug, defaultRolloutScrubber, emptyScrubCounts, findClaudeTranscripts, findOpencodeSessionsByDirectory, fromHarborTrajectory, gateErrors, gateGamedOutcome, gatedEvidenceOf, gatedRolloutIds, isRealnessGated, isRolloutLine, isTrainableSplit, measureFormatGate, mintRolloutRows, observedScore, observedSplitScore, openOpencodeDb, parseRolloutReleaseArgs, planPushCommand, readClaudeTranscript, readOpencodeSessionMessages, readRolloutJournal, readRolloutLedger, relabelImportedSplit, releaseRowRefs, runRolloutReleaseCli, scoreOrigin, scrubLines, scrubRolloutLine, scrubText, toHarborTrajectories, toHarborTrajectory, toJsonl, toRewardRows, toRftItem, toRftItems, toSftRows, toVerifiersRolloutOutput, toVerifiersRolloutOutputs, trainingReward, trainingScore, unmintableReasons, validateRolloutLine, writeRolloutLedger };

package/dist/search/index.d.ts

@@ -1,6 +1,6 @@
 import { A as SearchLedgerHash, C as SearchEdgeOperator, G as SearchSplit, H as SearchProposerKind, I as SearchNodeStatus, L as SearchOpenedEvent, T as SearchEstimateMethod, W as SearchSourceRef, i as SearchArtifactRef, j as SearchModelIdentity, l as SearchCandidateSurface, m as SearchClaim, t as NodeEstimate, y as SearchCloseReason } from "../search-ledger-types-p2xKfiDJ.js";
-import { $ as SearchNode, it as SearchStateView } from "../types-EBUav9Bu.js";
-import { $ as EXPANSION_OPERATORS, L as SearchPolicy, M as SearchUnitExtension, at as OperatorYieldRow, ct as LensResult, dt as rankingSplit, et as ExpansionOperator, ft as summarizeSamples, it as OperatorYieldOptions, lt as LensSignal, nt as OperatorOutcomeCounts, ot as OperatorYieldSignal, p as SearchRunResult, r as SearchArtifactCodec, rt as OperatorYieldData, s as SearchLane, st as operatorYield, t as RunSearchOptions, tt as MIN_OUTCOMES_FOR_WEIGHT, ut as SampleSummary, z as SearchPolicyView } from "../search-kernel-DA2-mS4N.js";
+import { $ as SearchNode, it as SearchStateView } from "../types-DfvQOe91.js";
+import { $ as EXPANSION_OPERATORS, L as SearchPolicy, M as SearchUnitExtension, at as OperatorYieldRow, ct as LensResult, dt as rankingSplit, et as ExpansionOperator, ft as summarizeSamples, it as OperatorYieldOptions, lt as LensSignal, nt as OperatorOutcomeCounts, ot as OperatorYieldSignal, p as SearchRunResult, r as SearchArtifactCodec, rt as OperatorYieldData, s as SearchLane, st as operatorYield, t as RunSearchOptions, tt as MIN_OUTCOMES_FOR_WEIGHT, ut as SampleSummary, z as SearchPolicyView } from "../search-kernel-C9CXvFR7.js";
 import { AgentProfileResourceRef } from "@tangle-network/agent-interface";
 //#region src/search/lenses/edit-credit.d.ts
 /** Signal name a policy reads. */

package/dist/search/index.js

@@ -1,6 +1,6 @@
 import { a as hashCanonical } from "../canonical-BsZVgBhW.js";
-import { _ as rankingSplit, h as operatorYield, m as MIN_OUTCOMES_FOR_WEIGHT, n as runSearch, p as EXPANSION_OPERATORS, v as summarizeSamples } from "../search-kernel-CumYAjA3.js";
-import { A as editCreditText, C as draftOnPlateau, D as EDIT_CREDIT_ESTIMATOR, E as front, O as EDIT_CREDIT_SIGNAL, S as vectorEmbedding, T as screenedNodes, _ as lineageEdits, a as skillCalibration, b as surfaceDigestEdits, c as META_SEARCH_SIGNAL, d as objectiveKey, f as renderMetaSearchText, g as lineEditDistance, h as landscape, i as nextUnitExtension, k as editCredit, l as metaSearch, m as formatLandscape, n as taskMatrix, o as skillManifold, p as searchPolicyGenome, r as formatSkillManifold, s as META_SEARCH_SCORE_SOURCE, t as tree, u as metaSearchScore, v as nearestNeighbours, w as searchPlateau, x as surfaceTextEdits, y as profileTextLines } from "../tree-DxGfhISA.js";
+import { _ as rankingSplit, h as operatorYield, m as MIN_OUTCOMES_FOR_WEIGHT, n as runSearch, p as EXPANSION_OPERATORS, v as summarizeSamples } from "../search-kernel-AVgIjXaI.js";
+import { A as editCreditText, C as draftOnPlateau, D as EDIT_CREDIT_ESTIMATOR, E as front, O as EDIT_CREDIT_SIGNAL, S as vectorEmbedding, T as screenedNodes, _ as lineageEdits, a as skillCalibration, b as surfaceDigestEdits, c as META_SEARCH_SIGNAL, d as objectiveKey, f as renderMetaSearchText, g as lineEditDistance, h as landscape, i as nextUnitExtension, k as editCredit, l as metaSearch, m as formatLandscape, n as taskMatrix, o as skillManifold, p as searchPolicyGenome, r as formatSkillManifold, s as META_SEARCH_SCORE_SOURCE, t as tree, u as metaSearchScore, v as nearestNeighbours, w as searchPlateau, x as surfaceTextEdits, y as profileTextLines } from "../tree-Ki_gW7z8.js";
 //#region src/search/nested-search.ts
 /** The deterministic id of the inner search one outer cell attempt runs. */
 function nestedSearchId(containment) {

package/dist/{search-kernel-CumYAjA3.js → search-kernel-AVgIjXaI.js}

@@ -6,7 +6,7 @@ import { n as DECISION_PAIRED_DELTA_STATISTIC } from "./paired-tests-C8iCsioC.js
 import { i as pairedDeltaTest, r as minimumPairsForPairedDeltaTest, t as decidePairedPromotion } from "./paired-promotion-decision-D5B1Cxc2.js";
 import { a as SearchLedgerError } from "./search-ledger-file-DY0AXX3W.js";
 import { f as redactText } from "./redact-Dk1ZdJEm.js";
-import { f as searchUnitScores, l as searchCellId, t as FileSearchLedger } from "./search-ledger-BnUytR8X.js";
+import { f as searchUnitScores, l as searchCellId, t as FileSearchLedger } from "./search-ledger-CL0-mXQ8.js";
 import { t as acquireSingleRunLock } from "./single-run-lock-fVu3MZWH.js";
 //#region src/pareto.ts
 /** Does candidate A weakly dominate B — strictly better on at least one objective and no worse on any? */
@@ -396,15 +396,27 @@ function summarizeSamples(samples, seed) {
 * per-node mean is already the node's improvement over the root, oriented so
 * larger is always better regardless of the objective's direction — this
 * lens adds no second orientation rule of its own. A child's sample is
-* `posteriorMean / knownCostUsd`, and only enters the sample when both are
-* known: the node shares 2 or more units with the root (one shared unit is
-* no estimate, §6.4), every one of its cells has a known cost
-* (`SearchSpend.unknownCostCells === 0`, so an unpriced cell never floors a
-* node to an artificially cheap, inflated yield), and it was not decided
-* `invalid` (a judge integrity or admission failure: its score is not
-* evidence of improvement, and `editCredit` excludes such nodes the same
-* way). A $0-known-cost node is excluded from yield (division by zero is not
-* a yield), not treated as free.
+* `posteriorMean / screenCostUsd`: the gain over the known cost of producing
+* and screening the node, which is the operator's doing. That cost is the
+* node's share of its proposal (the candidate-generation operation's cost
+* split evenly over every child edge the operation recorded, re-proposals
+* included) plus its cells allocated before its first decision: the first
+* rung and the proposer's train feedback, which the allocator plans together
+* at admission. Cells a later rung or the claim adds come after an
+* `advanced` or `finalist` decision; they are the allocator's and the
+* claim's choice, not the operator's, so an operator whose children advance
+* is not charged for the measurement their success earned. A seed edge
+* proposed nothing, so the root's cost is its cells alone.
+*
+* A child enters the sample only when both numbers are known: the node
+* shares 2 or more units with the root (one shared unit is no estimate,
+* §6.4); the edge names a recorded proposal operation whose cost is known;
+* every screen cell has run and has a known cost (an unpriced or unrecorded
+* part never floors a node to an artificially cheap, inflated yield); and it
+* was not decided `invalid` (a judge integrity or admission failure: its
+* score is not evidence of improvement, and `editCredit` excludes such nodes
+* the same way). A $0-known-cost node is excluded from yield (division by
+* zero is not a yield), not treated as free.
 */
 /** The operators an expansion may choose (`SearchExpansion.operator` in
 * `search-policy.ts`): every edge operator except `seed` (the search's own
@@ -420,6 +432,7 @@ const EXPANSION_OPERATORS = [
 * weight for that operator. Same threshold `NodeEstimate.method` uses for
 * `insufficient` (`minimumPairsForPairedDeltaTest(0.95)` = 6). */
 const MIN_OUTCOMES_FOR_WEIGHT = 6;
+const YIELD_METHOD = "per registered node, searchPosterior's mean gain over the root divided by the known cost of producing and screening the node: its even share of the proposal operation's cost over the operation's child edges, plus its cells allocated before its first decision (first rung and train feedback; later rungs and the claim excluded); no sample when the node was decided invalid, shares fewer than 2 units with the root, or any part of that cost is unknown, unrecorded or zero";
 function operatorYield(state, options = {}) {
 	const split = options.split ?? rankingSplit(state);
 	const posteriorByNode = new Map(state.header && state.rootNodeId !== null ? searchPosterior(state, { split }).nodes.map((node) => [node.nodeId, node]) : []);
@@ -446,7 +459,13 @@ function operatorYield(state, options = {}) {
 		}
 		return existing;
 	};
-	for (const edge of state.edges()) {
+	const edges = state.edges();
+	const childEdges = /* @__PURE__ */ new Map();
+	for (const edge of edges) {
+		const operationId = edge.proposer?.operationId;
+		if (operationId) childEdges.set(operationId, (childEdges.get(operationId) ?? 0) + 1);
+	}
+	for (const edge of edges) {
 		const child = state.node(edge.childNodeId);
 		if (!child) continue;
 		const bucket = group(edge.operator);
@@ -457,12 +476,12 @@ function operatorYield(state, options = {}) {
 		}
 		bucket.outcomes[statusKey(child.status)] += 1;
 		const posterior = posteriorByNode.get(child.nodeId);
-		const costKnown = child.spend.unknownCostCells === 0;
-		if (child.status === "invalid" || !posterior || posterior.mean === null || posterior.pairs < 2 || !costKnown || child.spend.knownUsd <= 0) {
+		const costUsd = screenCostUsd(state, child, edge, childEdges);
+		if (child.status === "invalid" || !posterior || posterior.mean === null || posterior.pairs < 2 || costUsd === null || costUsd <= 0) {
 			bucket.excluded += 1;
 			continue;
 		}
-		bucket.samples.push(posterior.mean / child.spend.knownUsd);
+		bucket.samples.push(posterior.mean / costUsd);
 	}
 	const rows = [...grouped.entries()].sort(([a], [b]) => a < b ? -1 : a > b ? 1 : 0).map(([operator, bucket]) => ({
 		operator,
@@ -480,6 +499,7 @@ function operatorYield(state, options = {}) {
 	return {
 		data: {
 			split,
+			method: YIELD_METHOD,
 			rows
 		},
 		signal: {
@@ -488,6 +508,27 @@ function operatorYield(state, options = {}) {
 		}
 	};
 }
+/** The known cost of producing and screening `node`, which `edge` registered;
+* null when any part of it is unknown. The screen ends at the node's first
+* decision: the allocator records `advanced` before it allocates a further
+* rung, and the claim decides `finalist` before it allocates test cells. */
+function screenCostUsd(state, node, edge, childEdges) {
+	let usd = 0;
+	if (edge.proposer !== null) {
+		const { operationId } = edge.proposer;
+		if (operationId === null) return null;
+		const operation = state.operation(operationId);
+		if (!operation?.recorded || !operation.costKnown) return null;
+		usd += operation.spentUsd / childEdges.get(operationId);
+	}
+	const screenEnd = node.decisions[0]?.sequence ?? Number.POSITIVE_INFINITY;
+	for (const cell of state.cells({ nodeId: node.nodeId })) {
+		if (cell.allocatedSequence > screenEnd) continue;
+		if (!cell.costKnown || cell.attempts === 0 && cell.cancelled === null) return null;
+		usd += cell.spentUsd;
+	}
+	return usd;
+}
 function statusKey(status) {
 	return status ?? "undecided";
 }
@@ -2424,4 +2465,4 @@ function errorMessage(error) {
 //#endregion
 export { searchEstimateMethod as C, paretoFrontier as D, dominates as E, paretoFrontierWithCrowding as O, searchCellSetDigest as S, seedFromDigest as T, rankingSplit as _, searchPolicyView as a, estimateNode as b, planSearchClaim as c, pairedPromotionPower as d, requiredPairsForPairedPromotion as f, INSUFFICIENT_FROM as g, operatorYield as h, searchExpansionIndex as i, searchClaimReserveUsd as l, MIN_OUTCOMES_FOR_WEIGHT as m, runSearch as n, SEARCH_CLAIM_RULE as o, EXPANSION_OPERATORS as p, searchDivergence as r, decideSearchClaim as s, SEARCH_KERNEL_SOURCE as t, verifySearchClaim as u, summarizeSamples as v, searchPosterior as w, estimateNodeFromCells as x, SEARCH_ESTIMATOR as y };
 
-//# sourceMappingURL=search-kernel-CumYAjA3.js.map
+//# sourceMappingURL=search-kernel-AVgIjXaI.js.map