@tangle-network/agent-eval 1.0.18 → 1.0.19
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +5 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/analyst/index.d.ts +3 -3
- package/dist/analyst/index.js +7 -7
- package/dist/app-grade/index.d.ts +10 -5
- package/dist/app-grade/index.d.ts.map +1 -1
- package/dist/app-grade/index.js +11 -7
- package/dist/app-grade/index.js.map +1 -1
- package/dist/{benchmark-command-Bqqvv9Tk.js → benchmark-command-DSJQeYom.js} +8 -8
- package/dist/{benchmark-command-Bqqvv9Tk.js.map → benchmark-command-DSJQeYom.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.d.ts +4 -4
- package/dist/campaign/index.js +10 -10
- package/dist/{campaign-DzLlT0So.js → campaign-BABipFel.js} +10 -10
- package/dist/{campaign-DzLlT0So.js.map → campaign-BABipFel.js.map} +1 -1
- package/dist/{campaign-evidence-CcA7a67m.js → campaign-evidence-DONEWvVY.js} +2 -2
- package/dist/{campaign-evidence-CcA7a67m.js.map → campaign-evidence-DONEWvVY.js.map} +1 -1
- package/dist/cli.js +6 -6
- package/dist/contract/index.d.ts +7 -7
- package/dist/contract/index.js +6 -6
- package/dist/{default-registry-BNv7eB1C.d.ts → default-registry-BbXr18AT.d.ts} +2 -2
- package/dist/{default-registry-BNv7eB1C.d.ts.map → default-registry-BbXr18AT.d.ts.map} +1 -1
- package/dist/{define-agent-eval-DKVYlm-T.d.ts → define-agent-eval-D_DQMH04.d.ts} +5 -5
- package/dist/{define-agent-eval-DKVYlm-T.d.ts.map → define-agent-eval-D_DQMH04.d.ts.map} +1 -1
- package/dist/{define-agent-eval-CjF2nTA3.js → define-agent-eval-Q4FdXGj7.js} +5 -5
- package/dist/{define-agent-eval-CjF2nTA3.js.map → define-agent-eval-Q4FdXGj7.js.map} +1 -1
- package/dist/{dspy-rlm-engine-_MEe2jIO.js → dspy-rlm-engine-CqQc0R2l.js} +2 -2
- package/dist/{dspy-rlm-engine-_MEe2jIO.js.map → dspy-rlm-engine-CqQc0R2l.js.map} +1 -1
- package/dist/{evaluation-DYAIkPQB.d.ts → evaluation-SDXKL6Fx.d.ts} +2 -2
- package/dist/{evaluation-DYAIkPQB.d.ts.map → evaluation-SDXKL6Fx.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/experiment/index.js +1 -1
- package/dist/{external-optimizer-process-za-SsIap.js → external-optimizer-process-CfrY-nJT.js} +2 -2
- package/dist/{external-optimizer-process-za-SsIap.js.map → external-optimizer-process-CfrY-nJT.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-B44kOfZb.js → external-optimizer-subprocess-BtJk8UcU.js} +2 -2
- package/dist/{external-optimizer-subprocess-B44kOfZb.js.map → external-optimizer-subprocess-BtJk8UcU.js.map} +1 -1
- package/dist/{findings-store-DWiBvJVM.js → findings-store-BnYN57-h.js} +3 -3
- package/dist/{findings-store-DWiBvJVM.js.map → findings-store-BnYN57-h.js.map} +1 -1
- package/dist/hosted/index.js +1 -1
- package/dist/{index-Dt6E4BHX.d.ts → index-AmfM3usI.d.ts} +2 -2
- package/dist/{index-Dt6E4BHX.d.ts.map → index-AmfM3usI.d.ts.map} +1 -1
- package/dist/{index-BJM4rwP2.d.ts → index-BPmj-_ta.d.ts} +4 -4
- package/dist/{index-BJM4rwP2.d.ts.map → index-BPmj-_ta.d.ts.map} +1 -1
- package/dist/{index-BAPkHt5H.d.ts → index-D4lT7aFu.d.ts} +10 -10
- package/dist/{index-BAPkHt5H.d.ts.map → index-D4lT7aFu.d.ts.map} +1 -1
- package/dist/{index-chNyKroE.d.ts → index-DLVbfss9.d.ts} +6 -6
- package/dist/{index-chNyKroE.d.ts.map → index-DLVbfss9.d.ts.map} +1 -1
- package/dist/{index-f09MX_H_.d.ts → index-HuSMvqR2.d.ts} +2 -2
- package/dist/{index-f09MX_H_.d.ts.map → index-HuSMvqR2.d.ts.map} +1 -1
- package/dist/{index-B1Bm9oJA.d.ts → index-lRKqfyg4.d.ts} +3 -7
- package/dist/index-lRKqfyg4.d.ts.map +1 -0
- package/dist/index.d.ts +9 -9
- package/dist/index.js +6 -6
- package/dist/{journal-C1uLJi0w.js → journal-CTX5uK-n.js} +36 -26
- package/dist/journal-CTX5uK-n.js.map +1 -0
- package/dist/ledger-core/index.d.ts +1 -1
- package/dist/ledger-core/index.js +2 -2
- package/dist/{ledger-core-CfT-FzPf.js → ledger-core-24IxvmVY.js} +2 -2
- package/dist/{ledger-core-CfT-FzPf.js.map → ledger-core-24IxvmVY.js.map} +1 -1
- package/dist/{llm-judge-CHLvjbVR.js → llm-judge-VD_MI6kd.js} +8 -8
- package/dist/{llm-judge-CHLvjbVR.js.map → llm-judge-VD_MI6kd.js.map} +1 -1
- package/dist/{matrix-Cy3oy7Rq.d.ts → matrix-Yg1lHIGI.d.ts} +2 -2
- package/dist/{matrix-Cy3oy7Rq.d.ts.map → matrix-Yg1lHIGI.d.ts.map} +1 -1
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/{pareto-B34N-vnn.js → pareto-a3EOmjfe.js} +4 -4
- package/dist/{pareto-B34N-vnn.js.map → pareto-a3EOmjfe.js.map} +1 -1
- package/dist/{proposal-findings-CZJkS_GY.js → proposal-findings-BfzzZ_lN.js} +2 -2
- package/dist/{proposal-findings-CZJkS_GY.js.map → proposal-findings-BfzzZ_lN.js.map} +1 -1
- package/dist/rl.d.ts +1 -1
- package/dist/rollout/index.d.ts +2 -2
- package/dist/search/index.d.ts +2 -2
- package/dist/search/index.js +2 -2
- package/dist/{search-kernel-kwO-sgCq.d.ts → search-kernel-DjqaPJqL.d.ts} +3 -3
- package/dist/{search-kernel-kwO-sgCq.d.ts.map → search-kernel-DjqaPJqL.d.ts.map} +1 -1
- package/dist/{search-ledger-CV_bQQed.js → search-ledger-Dk-yk5Li.js} +4 -4
- package/dist/{search-ledger-CV_bQQed.js.map → search-ledger-Dk-yk5Li.js.map} +1 -1
- package/dist/{search-ledger-file-DTBD6X-y.js → search-ledger-file-BhOdgYIt.js} +3 -3
- package/dist/{search-ledger-file-DTBD6X-y.js.map → search-ledger-file-BhOdgYIt.js.map} +1 -1
- package/dist/{search-shipper-CA3ooBbC.js → search-shipper-Dm-_Dtvy.js} +3 -3
- package/dist/{search-shipper-CA3ooBbC.js.map → search-shipper-Dm-_Dtvy.js.map} +1 -1
- package/dist/{search-summary-DBMjdecs.js → search-summary-RJ5zXjHN.js} +3 -3
- package/dist/{search-summary-DBMjdecs.js.map → search-summary-RJ5zXjHN.js.map} +1 -1
- package/dist/{single-run-lock-BGlW3OSk.js → single-run-lock-CUj2tjxT.js} +2 -2
- package/dist/{single-run-lock-BGlW3OSk.js.map → single-run-lock-CUj2tjxT.js.map} +1 -1
- package/dist/{skillopt-optimization-method-D72XZZfR.js → skillopt-optimization-method-D9Hlmhjg.js} +8 -8
- package/dist/{skillopt-optimization-method-D72XZZfR.js.map → skillopt-optimization-method-D9Hlmhjg.js.map} +1 -1
- package/dist/{statistical-heldout-_dXihc9T.d.ts → statistical-heldout-C-tgpUuu.d.ts} +2 -2
- package/dist/{statistical-heldout-_dXihc9T.d.ts.map → statistical-heldout-C-tgpUuu.d.ts.map} +1 -1
- package/dist/{systemone-CBvL1gx-.d.ts → systemone-CpDppX8s.d.ts} +3 -3
- package/dist/{systemone-CBvL1gx-.d.ts.map → systemone-CpDppX8s.d.ts.map} +1 -1
- package/dist/systemone.d.ts +2 -2
- package/dist/{tree-DPwgM2LX.js → tree-DIHp6UQ0.js} +3 -3
- package/dist/{tree-DPwgM2LX.js.map → tree-DIHp6UQ0.js.map} +1 -1
- package/dist/{types-D91tnhzf.d.ts → types-DsuOcYsE.d.ts} +2 -2
- package/dist/{types-D91tnhzf.d.ts.map → types-DsuOcYsE.d.ts.map} +1 -1
- package/dist/{types-CBkDo7ja.d.ts → types-wDBk14lj.d.ts} +2 -2
- package/dist/{types-CBkDo7ja.d.ts.map → types-wDBk14lj.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +2 -2
- package/package.json +1 -1
- package/dist/index-B1Bm9oJA.d.ts.map +0 -1
- package/dist/journal-C1uLJi0w.js.map +0 -1
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { u as CostProvenance } from "./cost-ledger-Czro9wmo.js";
|
|
2
|
-
import { C as JudgeDimension, w as JudgeScore } from "./types-
|
|
2
|
+
import { C as JudgeDimension, w as JudgeScore } from "./types-wDBk14lj.js";
|
|
3
3
|
import { o as MatrixResult } from "./index-DwUGehqX.js";
|
|
4
4
|
import { AgentProfile } from "@tangle-network/agent-interface";
|
|
5
5
|
interface MultishotMessage {
|
|
@@ -381,4 +381,4 @@ interface RunMultishotMatrixResult {
|
|
|
381
381
|
}
|
|
382
382
|
declare function runMultishotMatrix<TPersona extends MultishotPersona>(opts: RunMultishotMatrixOptions<TPersona>): Promise<RunMultishotMatrixResult>;
|
|
383
383
|
export { MultishotTransportRequest as A, MultishotMessage as C, MultishotToolDefinition as D, MultishotShape as E, MultishotTransportToolCall as M, assertMultishotShotResult as N, MultishotToolExecutor as O, MultishotFatalToolError as S, MultishotResult as T, renderDimensions as _, MultishotCellOutput as a, MultishotArtifact as b, RunMultishotMatrixResult as c, MultishotShot as d, RunMultishotOptions as f, MultishotJudgeConfig as g, JudgeRunResult as h, ConversationJudgeInput as i, MultishotTransportResponse as j, MultishotTransport as k, computeCellComposite as l, DEFAULT_JUDGE_MODEL as m, CellCompositeInput as n, MultishotJudges as o, runMultishot as p, CellCompositeScore as r, RunMultishotMatrixOptions as s, ArtifactJudgeInput as t, runMultishotMatrix as u, renderJsonFooter as v, MultishotPersona as w, MultishotDriverEmptyError as x, runJudge as y };
|
|
384
|
-
//# sourceMappingURL=matrix-
|
|
384
|
+
//# sourceMappingURL=matrix-Yg1lHIGI.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"matrix-
|
|
1
|
+
{"version":3,"file":"matrix-Yg1lHIGI.d.ts","names":[],"sources":["../src/multishot/types.ts","../src/multishot/judges.ts","../src/multishot/multishot.ts","../src/multishot/matrix.ts"],"mappings":";;;;UAIiB;EACf;EACA;EACA;EACA,YAAY;IAAQ;IAAY;IAAc,MAAM;;;UAGrC;EACf;EACA;EACA;IAAc;IAAc,MAAM;;EAClC;;UAGe;EACf,YAAY;EACZ,WAAW;EACX;EACA;;;EAGA;;;;;;;;EAQA,iBAAiB;;UAGF;EACf;EACA;IACE;IACA;IACA,YAAY;;;;;;UAOC;EACf;EACA,UAAU,MAAM;EAChB,QAAQ;EACR;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA;IAAY;IAAc;;;UAGX;EACf;IAAW;IAAyB,aAAa;;EACjD;IAAU;IAAwB;;;;EAGlC;;;;EAIA;;;;;;;;KASU,sBACV,KAAK,8BACF,QAAQ;KAED,yBACV,MAAM,yBACN;;EAEE,WAAW;EACX,SAAS;MAER;EAAU;EAAiB;;UAEf;;EAEf;;GAEC;;;;;;;;UASc,eAAe,iBAAiB;;EAE/C,eAAe,SAAS;;;EAGxB,2BAA2B,SAAS;;cAGzB,kCAAkC;WACjB;EAAA,YAAA;;cAMjB,gCAAgC;EAC/B,YAAA;;;;;;;;;;;;;;;;;;;iBAgCE,0BAA0B,yBAAyB,SAAS;cCvI/D;UAEI,qBAAqB;;EAEpC;;;EAGA,WAAW;;EAEX;;EAEA,YAAY;;EAEZ;;;EAGA,cAAc,OAAO;;EAErB;;UAGe;;EAEf,OAAO;;EAEP,MAAM;;iBAGc,SAAS,QAC7B,OAAO,qBAAqB,SAC5B,OAAO,SACN,QAAQ;;;iBAsIK,iBAAiB,eAAe;;iBAKhC,iBAAiB,eAAe;UCxK/B,oBAAoB,iBAAiB;EACpD,SAAS;EACT,SAAS;;;EAGT,QAAQ,eAAe;;EAEvB,QAAQ;;EAER,gBAAgB,eAAe;;;;EAI/B,mBAAmB;EACnB;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,gBAAgB;;;;EAIhB,iBAAiB;;;EAGjB,gBAAgB;EAChB,SAAS;;;;;;;;;;;KAYC,cAAc,iBAAiB,qBACzC,MAAM,oBAAoB,cACvB,QAAQ;;;;;;;;;;;;;;;;;iBAkBS,aAAa,iBAAiB,kBAClD,MAAM,oBAAoB,YACzB,QAAQ;UClEM,uBAAuB,iBAAiB;EACvD,YAAY;EACZ,SAAS;;UAGM,mBAAmB,iBAAiB;EACnD,UAAU;EACV,SAAS;;UAGM,gBAAgB,iBAAiB;;EAEhD,cAAc,qBAAqB,uBAAuB;;EAE1D,aAAa,qBAAqB,mBAAmB;;EAErD,iBAAiB,qBAAqB,mBAAmB;;EAEzD;;EAEA;;UAGe;EACf;EACA,cAAc;EACd;IACE,aAAa,MAAM;MAAe;MAAc;;IAChD;;EAEF;IACE,aAAa,MAAM;MAAe;MAAc;;IAChD;;;UAIa,0BAA0B,iBAAiB;;EAE1D,UAAU;IAAQ;IAAY,OAAO;;;EAErC,UAAU;;;EAGV,QAAQ,eAAe;;EAEvB,QAAQ,gBAAgB;;EAExB,QAAQ;;EAER,gBAAgB,eAAe;;EAE/B,mBAAmB;;EAEnB;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA,gBAAgB;;EAEhB,iBAAiB;;;EAGjB,gBAAgB;;;;;;;;;;;;;;;EAehB,UAAU,cAAc;;;;;UAMT;EACf;EACA;EACA;;UAoBe;EACf,cAAc;;EAEd,cAAc,cAAc;;EAE5B,iBAAiB,cAAc;;;;;;;iBAQjB,qBAAqB,OAAO;EAC1C;EACA;EACA;EACA;;UAuBe;EACf,QAAQ,aAAa;;iBAGD,mBAAmB,iBAAiB,kBACxD,MAAM,0BAA0B,YAC/B,QAAQ"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { A as MultishotTransportRequest, D as MultishotToolDefinition, T as MultishotResult, c as RunMultishotMatrixResult, f as RunMultishotOptions, s as RunMultishotMatrixOptions, w as MultishotPersona } from "../../matrix-
|
|
1
|
+
import { A as MultishotTransportRequest, D as MultishotToolDefinition, T as MultishotResult, c as RunMultishotMatrixResult, f as RunMultishotOptions, s as RunMultishotMatrixOptions, w as MultishotPersona } from "../../matrix-Yg1lHIGI.js";
|
|
2
2
|
interface CompareOptions {
|
|
3
3
|
/** Stop after this many mismatches. A structural divergence high in the
|
|
4
4
|
* tree would otherwise report every leaf below it. */
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { C as JudgeDimension, w as JudgeScore } from "../types-
|
|
2
|
-
import { A as MultishotTransportRequest, C as MultishotMessage, D as MultishotToolDefinition, E as MultishotShape, M as MultishotTransportToolCall, N as assertMultishotShotResult, O as MultishotToolExecutor, S as MultishotFatalToolError, T as MultishotResult, _ as renderDimensions, a as MultishotCellOutput, b as MultishotArtifact, c as RunMultishotMatrixResult, d as MultishotShot, f as RunMultishotOptions, g as MultishotJudgeConfig, h as JudgeRunResult, i as ConversationJudgeInput, j as MultishotTransportResponse, k as MultishotTransport, l as computeCellComposite, m as DEFAULT_JUDGE_MODEL, n as CellCompositeInput, o as MultishotJudges, p as runMultishot, r as CellCompositeScore, s as RunMultishotMatrixOptions, t as ArtifactJudgeInput, u as runMultishotMatrix, v as renderJsonFooter, w as MultishotPersona, x as MultishotDriverEmptyError, y as runJudge } from "../matrix-
|
|
1
|
+
import { C as JudgeDimension, w as JudgeScore } from "../types-wDBk14lj.js";
|
|
2
|
+
import { A as MultishotTransportRequest, C as MultishotMessage, D as MultishotToolDefinition, E as MultishotShape, M as MultishotTransportToolCall, N as assertMultishotShotResult, O as MultishotToolExecutor, S as MultishotFatalToolError, T as MultishotResult, _ as renderDimensions, a as MultishotCellOutput, b as MultishotArtifact, c as RunMultishotMatrixResult, d as MultishotShot, f as RunMultishotOptions, g as MultishotJudgeConfig, h as JudgeRunResult, i as ConversationJudgeInput, j as MultishotTransportResponse, k as MultishotTransport, l as computeCellComposite, m as DEFAULT_JUDGE_MODEL, n as CellCompositeInput, o as MultishotJudges, p as runMultishot, r as CellCompositeScore, s as RunMultishotMatrixOptions, t as ArtifactJudgeInput, u as runMultishotMatrix, v as renderJsonFooter, w as MultishotPersona, x as MultishotDriverEmptyError, y as runJudge } from "../matrix-Yg1lHIGI.js";
|
|
3
3
|
import { AgentProfile } from "@tangle-network/agent-interface";
|
|
4
4
|
/** Prompt and completion tokens one leg reported. */
|
|
5
5
|
interface MultishotLegTokens {
|
package/dist/openapi.json
CHANGED
|
@@ -2,7 +2,7 @@
|
|
|
2
2
|
"openapi": "3.1.0",
|
|
3
3
|
"info": {
|
|
4
4
|
"title": "@tangle-network/agent-eval — wire protocol",
|
|
5
|
-
"version": "1.0.
|
|
5
|
+
"version": "1.0.19",
|
|
6
6
|
"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: 1.1.0. Bumps on breaking changes to request/response schemas.",
|
|
7
7
|
"contact": {
|
|
8
8
|
"name": "Tangle Network",
|
|
@@ -5,10 +5,10 @@ import { t as confidenceInterval } from "./descriptive-B3Eb7tX0.js";
|
|
|
5
5
|
import { s as wilson } from "./paired-binary-CMzmSSd5.js";
|
|
6
6
|
import { n as DECISION_PAIRED_DELTA_STATISTIC } from "./paired-tests-Buh6feJ4.js";
|
|
7
7
|
import { i as pairedDeltaTest, r as minimumPairsForPairedDeltaTest, t as decidePairedPromotion } from "./paired-promotion-decision-CiueaTTj.js";
|
|
8
|
-
import { a as SearchLedgerError } from "./search-ledger-file-
|
|
8
|
+
import { a as SearchLedgerError } from "./search-ledger-file-BhOdgYIt.js";
|
|
9
9
|
import { c as redactText } from "./redact-zp_ILC4l.js";
|
|
10
|
-
import { l as searchUnitScores, p as USD_TOLERANCE, t as FileSearchLedger, u as searchCellId } from "./search-ledger-
|
|
11
|
-
import { t as acquireSingleRunLock } from "./single-run-lock-
|
|
10
|
+
import { l as searchUnitScores, p as USD_TOLERANCE, t as FileSearchLedger, u as searchCellId } from "./search-ledger-Dk-yk5Li.js";
|
|
11
|
+
import { t as acquireSingleRunLock } from "./single-run-lock-CUj2tjxT.js";
|
|
12
12
|
// src/statistics/normalize-signed-zero.ts
|
|
13
13
|
/** Normalize IEEE negative zero before numerical evidence reaches canonical JSON. */
|
|
14
14
|
function normalizeSignedZero(value) {
|
|
@@ -2609,4 +2609,4 @@ function paretoFrontierWithCrowding(candidates, objectives) {
|
|
|
2609
2609
|
}
|
|
2610
2610
|
export { normalizeSignedZero as A, SEARCH_ESTIMATOR as C, searchEstimateMethod as D, searchCellSetDigest as E, searchPosterior as O, requiredPairsForPairedPromotion as S, estimateNodeFromCells as T, planSearchClaim as _, searchDivergence as a, verifySearchClaim as b, MIN_OUTCOMES_FOR_WEIGHT as c, rankingSplit as d, summarizeSamples as f, decideSearchClaim as g, SEARCH_CLAIM_RULE as h, runSearch as i, seedFromDigest as k, operatorYield as l, searchExpansionIndex as m, paretoFrontier as n, searchPolicyView as o, SEARCH_KERNEL_SOURCE as p, paretoFrontierWithCrowding as r, EXPANSION_OPERATORS as s, dominates as t, INSUFFICIENT_FROM as u, searchClaimDecision as v, estimateNode as w, pairedPromotionPower as x, searchClaimReserveUsd as y };
|
|
2611
2611
|
|
|
2612
|
-
//# sourceMappingURL=pareto-
|
|
2612
|
+
//# sourceMappingURL=pareto-a3EOmjfe.js.map
|