@tangle-network/agent-eval 0.144.0 → 0.144.3
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 +36 -0
- package/dist/analyst/index.d.ts +88 -23
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +4 -4
- package/dist/{analyze-runs-BScZvqMV.js → analyze-runs-DWIvOAGk.js} +2 -2
- package/dist/{analyze-runs-BScZvqMV.js.map → analyze-runs-DWIvOAGk.js.map} +1 -1
- package/dist/{benchmark-DxaZfy0w.d.ts → benchmark-CP6kWfj8.d.ts} +3 -3
- package/dist/{benchmark-DxaZfy0w.d.ts.map → benchmark-CP6kWfj8.d.ts.map} +1 -1
- package/dist/{benchmark-command-4c7N_rlw.js → benchmark-command-CQPKRUr-.js} +332 -111
- package/dist/benchmark-command-CQPKRUr-.js.map +1 -0
- package/dist/benchmarks/index.d.ts +1 -1
- package/dist/benchmarks/index.js +1 -1
- package/dist/{benchmarks-05UGZ8sZ.js → benchmarks-CkG1bWFa.js} +4 -4
- package/dist/{benchmarks-05UGZ8sZ.js.map → benchmarks-CkG1bWFa.js.map} +1 -1
- package/dist/campaign/index.d.ts +6 -6
- package/dist/campaign/index.js +4 -4
- package/dist/{campaign-BKOtvRAB.js → campaign-DjGFyPxH.js} +74 -40
- package/dist/campaign-DjGFyPxH.js.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/{client-Cgl6KasJ.d.ts → client-Bbht4xxl.d.ts} +4 -4
- package/dist/{client-Cgl6KasJ.d.ts.map → client-Bbht4xxl.d.ts.map} +1 -1
- package/dist/{completion-verifier-D15NHYSk.d.ts → completion-verifier-EJERfFwF.d.ts} +3 -3
- package/dist/{completion-verifier-D15NHYSk.d.ts.map → completion-verifier-EJERfFwF.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +8 -8
- package/dist/contract/index.js +10 -8
- package/dist/contract/index.js.map +1 -1
- package/dist/control.d.ts +2 -2
- package/dist/control.js +1 -1
- package/dist/{default-registry-DfHJEwYh.js → default-registry-SOyHB6qG.js} +2 -2
- package/dist/{default-registry-DfHJEwYh.js.map → default-registry-SOyHB6qG.js.map} +1 -1
- package/dist/{default-registry-D3uqKbo6.d.ts → default-registry-iXfu2trt.d.ts} +5 -5
- package/dist/{default-registry-D3uqKbo6.d.ts.map → default-registry-iXfu2trt.d.ts.map} +1 -1
- package/dist/{dspy-rlm-engine-CBFwlyaY.js → dspy-rlm-engine-IRCG8kdi.js} +61 -105
- package/dist/dspy-rlm-engine-IRCG8kdi.js.map +1 -0
- package/dist/{eval-campaign-YdkpWWoT.js → eval-campaign-lZcDIwQM.js} +3 -3
- package/dist/{eval-campaign-YdkpWWoT.js.map → eval-campaign-lZcDIwQM.js.map} +1 -1
- package/dist/{exact-types-B0lJV3tu.d.ts → exact-types-BygCBR4L.d.ts} +2 -2
- package/dist/{exact-types-B0lJV3tu.d.ts.map → exact-types-BygCBR4L.d.ts.map} +1 -1
- package/dist/{external-optimizer-contracts-iK0yu4AR.d.ts → external-optimizer-contracts-CdmX2K2S.d.ts} +47 -10
- package/dist/external-optimizer-contracts-CdmX2K2S.d.ts.map +1 -0
- package/dist/{feedback-trajectory-BCHqzLh3.d.ts → feedback-trajectory-CSIkRLQX.d.ts} +3 -3
- package/dist/{feedback-trajectory-BCHqzLh3.d.ts.map → feedback-trajectory-CSIkRLQX.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +3 -3
- package/dist/{index-DgbFl4cv.d.ts → index-BuHs_OnD.d.ts} +17 -12
- package/dist/{index-DgbFl4cv.d.ts.map → index-BuHs_OnD.d.ts.map} +1 -1
- package/dist/{index-D2enbqA0.d.ts → index-CNOCxBLh.d.ts} +2 -2
- package/dist/{index-D2enbqA0.d.ts.map → index-CNOCxBLh.d.ts.map} +1 -1
- package/dist/{index-DFLVtPZ9.d.ts → index-DGIzNtRv.d.ts} +2 -2
- package/dist/{index-DFLVtPZ9.d.ts.map → index-DGIzNtRv.d.ts.map} +1 -1
- package/dist/{index-DtMpBKVF.d.ts → index-D_qTihaQ.d.ts} +5 -5
- package/dist/{index-DtMpBKVF.d.ts.map → index-D_qTihaQ.d.ts.map} +1 -1
- package/dist/index.d.ts +20 -20
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +17 -17
- package/dist/{insight-report-Bh_8ksel.d.ts → insight-report-D5m1z0_n.d.ts} +3 -3
- package/dist/{insight-report-Bh_8ksel.d.ts.map → insight-report-D5m1z0_n.d.ts.map} +1 -1
- package/dist/ledger-core/index.js +1 -1
- package/dist/{ledger-core-Dxz0Rkwa.js → ledger-core-DXZIqu17.js} +3 -3
- package/dist/ledger-core-DXZIqu17.js.map +1 -0
- package/dist/{llm-client-B3WXSH5Y.js → llm-client-D3EoChAU.js} +7 -7
- package/dist/llm-client-D3EoChAU.js.map +1 -0
- package/dist/meta-eval/index.d.ts +1 -1
- package/dist/{mint-Ctwk079K.js → mint-DD-0oQTA.js} +2 -2
- package/dist/{mint-Ctwk079K.js.map → mint-DD-0oQTA.js.map} +1 -1
- package/dist/multishot/index.d.ts +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{propose-review-control-DLXz4FCX.js → propose-review-control-BciUCZoh.js} +2 -2
- package/dist/{propose-review-control-DLXz4FCX.js.map → propose-review-control-BciUCZoh.js.map} +1 -1
- package/dist/{release-report-B_bQOHM-.d.ts → release-report-BZRWdq_t.d.ts} +3 -3
- package/dist/{release-report-B_bQOHM-.d.ts.map → release-report-BZRWdq_t.d.ts.map} +1 -1
- package/dist/{release-report-B5XPBvAU.js → release-report-Sl0xfkFv.js} +2 -2
- package/dist/{release-report-B5XPBvAU.js.map → release-report-Sl0xfkFv.js.map} +1 -1
- package/dist/{replay-DjUfTrHD.js → replay-CqOsGjzU.js} +2 -9
- package/dist/replay-CqOsGjzU.js.map +1 -0
- package/dist/{replay-BuJM6kLh.d.ts → replay-DQ-55DC_.d.ts} +4 -4
- package/dist/{replay-BuJM6kLh.d.ts.map → replay-DQ-55DC_.d.ts.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +1 -1
- package/dist/{researcher-C6lzl-rP.d.ts → researcher-BSiCoM1s.d.ts} +3 -3
- package/dist/{researcher-C6lzl-rP.d.ts.map → researcher-BSiCoM1s.d.ts.map} +1 -1
- package/dist/{reward-hacking-DjTi9HLb.js → reward-hacking-CyuzxKly.js} +2 -2
- package/dist/{reward-hacking-DjTi9HLb.js.map → reward-hacking-CyuzxKly.js.map} +1 -1
- package/dist/{reward-hacking-CEVVmy3h.d.ts → reward-hacking-DFgkEY4p.d.ts} +2 -2
- package/dist/{reward-hacking-CEVVmy3h.d.ts.map → reward-hacking-DFgkEY4p.d.ts.map} +1 -1
- package/dist/rl.d.ts +5 -5
- package/dist/rl.js +4 -4
- package/dist/rollout/index.d.ts +1 -1
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-8nj3mYvx.js → rollout-C2fD1cf4.js} +2 -2
- package/dist/{rollout-8nj3mYvx.js.map → rollout-C2fD1cf4.js.map} +1 -1
- package/dist/{rubric-predictive-validity-C2CthIfY.d.ts → rubric-predictive-validity-C4r4Y-q8.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-C2CthIfY.d.ts.map → rubric-predictive-validity-C4r4Y-q8.d.ts.map} +1 -1
- package/dist/{run-evidence-8Ou28QSa.d.ts → run-evidence-H1vRpIdT.d.ts} +3 -3
- package/dist/{run-evidence-8Ou28QSa.d.ts.map → run-evidence-H1vRpIdT.d.ts.map} +1 -1
- package/dist/{run-record-vRgqWmJw.js → run-record-CWN8-VsV.js} +40 -14
- package/dist/{run-record-vRgqWmJw.js.map → run-record-CWN8-VsV.js.map} +1 -1
- package/dist/{run-record-Tb3TTtUn.d.ts → run-record-ooo9FWns.d.ts} +5 -9
- package/dist/{run-record-Tb3TTtUn.d.ts.map → run-record-ooo9FWns.d.ts.map} +1 -1
- package/dist/{semantic-concept-judge-DJQtFr95.js → semantic-concept-judge-Do5aM9wP.js} +3 -3
- package/dist/{semantic-concept-judge-DJQtFr95.js.map → semantic-concept-judge-Do5aM9wP.js.map} +1 -1
- package/dist/{server-Cu4M3NSO.js → server-Df00sdwz.js} +3 -3
- package/dist/{server-Cu4M3NSO.js.map → server-Df00sdwz.js.map} +1 -1
- package/dist/{single-run-lock-t1si1ob7.js → single-run-lock-D5iN0Xzb.js} +719 -187
- package/dist/single-run-lock-D5iN0Xzb.js.map +1 -0
- package/dist/{skill-usage-BiVEU0QY.d.ts → skill-usage-DtpLou9L.d.ts} +28 -9
- package/dist/skill-usage-DtpLou9L.d.ts.map +1 -0
- package/dist/{skillopt-optimization-method-Ds8J1_K8.js → skillopt-optimization-method-C4FX42dy.js} +42 -210
- package/dist/skillopt-optimization-method-C4FX42dy.js.map +1 -0
- package/dist/{skillopt-optimization-method-B7o01OdX.d.ts → skillopt-optimization-method-CvSJGdm3.d.ts} +12 -8
- package/dist/{skillopt-optimization-method-B7o01OdX.d.ts.map → skillopt-optimization-method-CvSJGdm3.d.ts.map} +1 -1
- package/dist/{summary-report-o3eJ3gxG.d.ts → summary-report-BOM6dfP7.d.ts} +2 -2
- package/dist/{summary-report-o3eJ3gxG.d.ts.map → summary-report-BOM6dfP7.d.ts.map} +1 -1
- package/dist/{tool-groups-Bqy4A3QB.d.ts → tool-groups-CMmsgTzj.d.ts} +7 -7
- package/dist/{tool-groups-Bqy4A3QB.d.ts.map → tool-groups-CMmsgTzj.d.ts.map} +1 -1
- package/dist/traces.d.ts +4 -4
- package/dist/traces.js +1 -1
- package/dist/types-BjMFz88h.d.ts.map +1 -1
- package/dist/{types-KEqL1pZc.d.ts → types-DcJxgsLy.d.ts} +8 -6
- package/dist/{types-KEqL1pZc.d.ts.map → types-DcJxgsLy.d.ts.map} +1 -1
- package/dist/{types-D3jh6F98.d.ts → types-y8jrxXWd.d.ts} +2 -2
- package/dist/{types-D3jh6F98.d.ts.map → types-y8jrxXWd.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +1 -1
- package/dist/wire/index.js +1 -1
- package/docs/eval-surface-map.md +4 -0
- package/docs/trace-analysis.md +43 -32
- package/package.json +1 -1
- package/dist/benchmark-command-4c7N_rlw.js.map +0 -1
- package/dist/campaign-BKOtvRAB.js.map +0 -1
- package/dist/dspy-rlm-engine-CBFwlyaY.js.map +0 -1
- package/dist/external-optimizer-contracts-iK0yu4AR.d.ts.map +0 -1
- package/dist/ledger-core-Dxz0Rkwa.js.map +0 -1
- package/dist/llm-client-B3WXSH5Y.js.map +0 -1
- package/dist/replay-DjUfTrHD.js.map +0 -1
- package/dist/single-run-lock-t1si1ob7.js.map +0 -1
- package/dist/skill-usage-BiVEU0QY.d.ts.map +0 -1
- package/dist/skillopt-optimization-method-Ds8J1_K8.js.map +0 -1
package/dist/cli.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
import { o as runRolloutReleaseCli } from "./hf-dataset-XggBupCr.js";
|
|
3
|
-
import { n as runAnalystBenchmarkCommand } from "./benchmark-command-
|
|
4
|
-
import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-
|
|
3
|
+
import { n as runAnalystBenchmarkCommand } from "./benchmark-command-CQPKRUr-.js";
|
|
4
|
+
import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-Df00sdwz.js";
|
|
5
5
|
import { writeFileSync } from "node:fs";
|
|
6
6
|
//#region src/cli-config.ts
|
|
7
7
|
function resolveCliLlmConfig(env = process.env) {
|
|
@@ -1,6 +1,6 @@
|
|
|
1
|
-
import { l as RunTerminalOutcome } from "./run-record-
|
|
2
|
-
import { _ as GateDecision, j as MutableSurface } from "./types-
|
|
3
|
-
import { o as InsightReport } from "./insight-report-
|
|
1
|
+
import { l as RunTerminalOutcome } from "./run-record-ooo9FWns.js";
|
|
2
|
+
import { _ as GateDecision, j as MutableSurface } from "./types-DcJxgsLy.js";
|
|
3
|
+
import { o as InsightReport } from "./insight-report-D5m1z0_n.js";
|
|
4
4
|
//#region src/hosted/types.d.ts
|
|
5
5
|
declare const HOSTED_WIRE_VERSION: '2026-07-24.v1';
|
|
6
6
|
type HostedWireVersion = typeof HOSTED_WIRE_VERSION;
|
|
@@ -199,4 +199,4 @@ declare function hostedClientFromEnv(overrides?: Partial<HostedTenant> & {
|
|
|
199
199
|
}): HostedClient | undefined;
|
|
200
200
|
//#endregion
|
|
201
201
|
export { UnixNanoTimestamp as _, hostedTenantFromEnv as a, EvalRunGenerationSnapshot as c, HostedIngestHeaders as d, HostedWireVersion as f, TraceSpanEvent as g, IngestTracesRequest as h, hostedClientFromEnv as i, EvalRunStatus as l, IngestResponse as m, HostedTenant as n, EvalRunCellScore as o, IngestEvalRunsRequest as p, createHostedClient as r, EvalRunEvent as s, HostedClient as t, HOSTED_WIRE_VERSION as u };
|
|
202
|
-
//# sourceMappingURL=client-
|
|
202
|
+
//# sourceMappingURL=client-Bbht4xxl.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"client-
|
|
1
|
+
{"version":3,"file":"client-Bbht4xxl.d.ts","names":[],"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;cAgCa;KACD,2BAA2B;;UAKtB;;EAEf;;EAEA;;EAEA,yBAAyB;;EAEzB;;;KAMU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA,YAAY,eAAe;;EAE3B,iBAAiB;;EAEjB;;EAEA;;UAGe;;EAEf;;;EAGA;;;EAGA,UAAU;;EAEV,OAAO;;EAEP;;EAEA;;EAEA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,QAAQ;;EAER,WAAW;;EAEX,aAAa;;EAEb,eAAe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;EAMA,gBAAgB;;;;;;KASN;;;;;;UAOK;EACf;EACA;EACA;EACA;EACA,mBAAmB;EACnB,iBAAiB;EACjB,YAAY;EACZ,SAAS;IACP,cAAc;IACd;IACA,aAAa;;EAEf;IAAW;IAAgC;;;EAE3C;;EAEA;;EAEA;;EAEA;;UAKe;EACf,aAAa;EACb,QAAQ;;UAGO;EACf,aAAa;EACb,OAAO;;UAGQ;;EAEf;;EAEA,UAAU;IAAQ;IAAe;;;;;UC1JlB;;EAEf;;EAEA;;EAEA;;EAEA,mBAAmB;;EAEnB;;EAEA;;UAGe;EACf,cAAc,OAAO,cAAc,0BAA0B,QAAQ;EACrE,eAAe,QAAQ,gBAAgB,0BAA0B,QAAQ;EACzE,aAAa,OAAO,kBAAkB,0BAA0B,QAAQ;WAC/D,QAAQ;WACR,aAAa;;iBAyGR,mBAAmB,QAAQ,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsE1C,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C;iBAiBa,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C"}
|
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
import { t as DefaultVerdict } from "./verdict-Dps8_okt.js";
|
|
2
2
|
import { c as CostLedgerHandle } from "./cost-ledger-FuQvHxPm.js";
|
|
3
3
|
import { A as ChatClient, _t as RawProviderSink } from "./types-BjMFz88h.js";
|
|
4
|
-
import { c as AnalystRunInputs, i as AnalystFinding, l as AnalystRunResult, n as AnalystContext, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary } from "./types-
|
|
5
|
-
import { i as ExactAnalystRunEvent, o as ExactAnalystRunResult, u as ExactExecutionComponentIdentity } from "./exact-types-
|
|
4
|
+
import { c as AnalystRunInputs, i as AnalystFinding, l as AnalystRunResult, n as AnalystContext, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary } from "./types-y8jrxXWd.js";
|
|
5
|
+
import { i as ExactAnalystRunEvent, o as ExactAnalystRunResult, u as ExactExecutionComponentIdentity } from "./exact-types-BygCBR4L.js";
|
|
6
6
|
//#region src/analyst/registry.d.ts
|
|
7
7
|
interface AnalystHooks {
|
|
8
8
|
/** Legacy runs may mutate ctx; exact runs provide a frozen observational context. */
|
|
@@ -411,4 +411,4 @@ declare function createTokenRecallChecker(opts?: {
|
|
|
411
411
|
}): CorrectnessChecker;
|
|
412
412
|
//#endregion
|
|
413
413
|
export { ExactAnalystRunExecutionError as A, jsonHasKeys as C, AnalystRegistryOptions as D, AnalystRegistry as E, RegistryRunOpts as M, assertExactRegistryRunOpts as N, BudgetPolicy as O, containsAll as S, AnalystHooks as T, ValidationContext as _, ProducedProposal as a, byteLengthRange as b, SatisfiedBy as c, createLlmCorrectnessChecker as d, createTokenRecallChecker as f, ArtifactValidator as g, Artifact as h, LlmCorrectnessCheckerOpts as i, ExactRegistryRunOpts as j, ExactAnalystBudgetPolicy as k, TaskGold as l, verifyCompletion as m, CompletionVerdict as n, ProducedState as o, parseCorrectnessResponse as p, CorrectnessChecker as r, RequirementCheck as s, CompletionRequirement as t, completionVerdict as u, ValidationIssue as v, regexMatch as w, composeValidators as x, ValidationResult as y };
|
|
414
|
-
//# sourceMappingURL=completion-verifier-
|
|
414
|
+
//# sourceMappingURL=completion-verifier-EJERfFwF.d.ts.map
|
package/dist/{completion-verifier-D15NHYSk.d.ts.map → completion-verifier-EJERfFwF.d.ts.map}
RENAMED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"completion-verifier-
|
|
1
|
+
{"version":3,"file":"completion-verifier-EJERfFwF.d.ts","names":[],"sources":["../src/analyst/registry.ts","../src/artifact-validator.ts","../src/completion-verifier.ts"],"mappings":";;;;;;UAqDiB;;EAEf,iBAAiB;IACf,SAAS;IACT,KAAK;IACL;aACS;;EAEX,gBAAgB;IACd,SAAS;IACT,SAAS;IACT,UAAU;IACV;aACS;;;;;;EAMX,SAAS;IACP,SAAS;IACT,OAAO;IACP;MACE,+BAA+B,QAAQ;;EAE3C,YAAY;IAAQ,QAAQ;aAA4B;;UAGzC;;EAEf;;EAEA,UAAU;;;;;;;EAOV,YAAY;IACV,SAAS;IACT;IACA;IACA;;;UAIa;;EAEf,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,QAAQ;;EAER,gBAAgB;;EAEhB,gBAAgB;;EAEhB,eAAe;;UAGA;;EAEf;;EAEA;;EAEA,SAAS;;EAET;;EAEA,SAAS;;EAET,aAAa;;EAEb;;EAEA,OAAO;;;;;;;;;EASP,gBAAgB,cAAc,kBAAkB,eAAe,cAAc;;;;;;EAM7E;;;KAIU;WAEG;WACA;;WAGA;WACA;;WAEA,SAAS,SAAS;;;;;;;;;;UAWhB;WACN;WACA,QAAQ;WACR;WACA,QAAQ;WACR,YAAY;WACZ,oBAAoB;WACpB;WACA,MAAM,SAAS;WACf,eACL,cAAc,kBACd,SAAS,eAAe,cAAc;WAEjC;WACA;WACA;WACA;;;cAqCE,sCAAsC;WACxC;WACA,QAAQ;EAEjB,YAAY,iBAAiB,QAAQ,uBAAuB,UAAU;;cAU3D;mBACM;mBACA;EAEjB,YAAY,UAAS;EAIrB,SAAS,SAAS;EAsBlB,QAAQ;IACN;IACA;IACA;IACA,MAAM;;EAUF,IACJ,eACA,QAAQ,kBACR,UAAS,kBACR,QAAQ;;EAUL,SACJ,eACA,QAAQ,kBACR,SAAS,uBACR,QAAQ;;EAQJ,eACL,eACA,QAAQ,kBACR,SAAS,uBACR,eAAe;;;;;;;;;;;;EAmBX,UACL,eACA,QAAQ,kBACR,UAAS,kBACR,eAAe;UAIV;UA8BA;UAwFO;UA2aP;UAaA;UAUA;;;iBAiGM,2BAA2B,yBAAyB,SAAS;;;;;;;;;;;;;;;;UC79B5D;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,WAAW;;UAGI;EACf;EACA;;EAEA,iBAAiB;;EAEjB,QAAQ;;UAGO;EACf;EACA;;EAEA;;UAGe;EACf;;EAEA;EACA,QAAQ;;EAER,WAAW;;UAGI;;EAEf;;EAEA;;EAEA,SAAS,UAAU,UAAU,SAAS,oBAAoB,QAAQ;;;;;;iBAWpD,kBACd,YAAY,qBACZ;EAAY;EAAe;IAC1B;;iBAgCa,WAAW,cAAc,SAAS,SAAS;;iBAkB3C,YAAY,cAAc,0BAA0B;;iBAuCpD,gBAAgB,cAAc,aAAa,cAAc;;iBAoBzD,YACd,cACA,oBACA;EAAY;IACX;;;;KCnJS;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,cAAc;;UAGC;EACf;EACA,cAAc;;UAGC;EACf;EACA;EACA;;EAEA;;;UAIe;;EAEf,WAAW;;EAEX,WAAW;;EAEX;;UAGe;EACf;EACA;;EAEA;;;;;;EAMA;;EAEA;;;;;;;;;EASA;;EAEA;;EAEA;;;;;UAMe,0BAA0B;EACzC;EACA,cAAc;;EAEd;;EAEA;;EAEA;;;;;;;;;iBAUc,kBAAkB;EAChC;EACA,cAAc;IACZ;;;;;;KAoCQ,sBACV,aAAa,uBACb,oBACG;EAAU;EAAkB;;;;;;;;;iBAqLX,iBACpB,MAAM,UACN,OAAO,eACP,kBAAkB,qBACjB,QAAQ;UAoFM;EACf;;EAEA,aAAa;EACb;EACA,WAAW;EACX,SAAS;;EAET;;;;;;EAMA;;;;;;EAMA,UAAU;;;;;;;;;iBAUI,yBAAyB;EAAgB;EAAkB;;;;;;;;iBAiC3D,4BACd,MAAM,YACN,OAAM,4BACL;;;;;;;;;;;;;;;iBA6Ia,yBACd;EAAQ;EAAoB;IAC3B"}
|
package/dist/contract/index.d.ts
CHANGED
|
@@ -1,13 +1,13 @@
|
|
|
1
1
|
import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "../cost-ledger-FuQvHxPm.js";
|
|
2
|
-
import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-
|
|
2
|
+
import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-ooo9FWns.js";
|
|
3
3
|
import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } from "../types-BjMFz88h.js";
|
|
4
|
-
import { h as ProposalFindingOrigin, i as AnalystFinding, m as ProposalFinding, v as makeProposalFinding } from "../types-
|
|
5
|
-
import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-
|
|
6
|
-
import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-
|
|
7
|
-
import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-
|
|
8
|
-
import { A as summarizeExecution, D as ExecutionReport, E as AnalyzeRunsOptions, O as SummarizeExecutionOptions, k as analyzeRuns } from "../tool-groups-
|
|
9
|
-
import { A as RunEvalOptions, An as LlmJudgeDimension, B as OptimizerModelBudget, Bt as ComparisonCost, C as RunImprovementLoopOptions, Cn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, D as RunOptimizationOptions, Dn as ReferenceEquivalenceScenario, E as PremeasuredOptimizationBaseline, En as ReferenceEquivalenceJudgeResult, F as GepaOptimizationMethodConfig, Ft as ExternalTextOptimizerContext, G as ObjectiveSource, Gt as OptimizationMethodProvenance, H as AxisVerdict, Ht as OptimizationMethodComparison, I as GepaOptimizationRecipe, It as ExternalTextOptimizerResult, J as PromotionPolicy, K as ParetoSignificanceGateOptions, Kt as OptimizationMethodResult, L as GepaRunnerCommand, Lt as ExternalOptimizationExample, M as GepaAdaptiveEngineRun, Mn as llmJudge, Mt as composeGate, N as GepaEngineOptions, Nt as externalTextOptimizationMethod, On as createReferenceEquivalenceJudge, P as GepaEngineRun, Pt as ExternalTextOptimizationMethodConfig, Qt as compareOptimizationMethods, R as gepaOptimizationMethod, Rt as ExternalTextEvaluationResponse, Sn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, T as runImprovementLoop, Tn as ReferenceEquivalenceJudgeOptions, U as BuildEvidenceVectorOptions, Ut as OptimizationMethodInput, V as AxisEvidence, Vt as OptimizationMethod, W as EvidenceVector, X as paretoPolicy, Xt as OptimizationTokenUsage, Y as buildEvidenceVector, Yt as OptimizationPackageSource, Z as paretoSignificanceGate, _n as campaignSplitDigest, an as RunCampaignOptions, dt as DefaultProductionGateCheck, fn as fsCampaignStorage, ft as DefaultProductionGateOptions, i as skillOptOptimizationMethod, j as runEval, jn as LlmJudgeOptions, kn as runReferenceEquivalenceJudge, lt as HeldOutGateOptions, mt as defaultProductionGate, n as SkillOptRunnerCommand, ot as PowerPreflight, p as LoopProvenanceRecord, pn as inMemoryCampaignStorage, pt as DefaultProductionRewardHackingOptions, q as PromotionObjective, r as SkillOptTrainerConfig, sn as runCampaign, t as SkillOptOptimizationMethodConfig, tn as CampaignCellFailureReceipt, un as CampaignStorage, ut as heldOutGate, w as RunImprovementLoopResult, wn as ReferenceEquivalenceJudgeInput, z as OpenAICompatibleOptimizerModel, zt as CompareOptimizationMethodsOptions } from "../skillopt-optimization-method-
|
|
10
|
-
import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-
|
|
4
|
+
import { h as ProposalFindingOrigin, i as AnalystFinding, m as ProposalFinding, v as makeProposalFinding } from "../types-y8jrxXWd.js";
|
|
5
|
+
import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-iXfu2trt.js";
|
|
6
|
+
import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-DcJxgsLy.js";
|
|
7
|
+
import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-D5m1z0_n.js";
|
|
8
|
+
import { A as summarizeExecution, D as ExecutionReport, E as AnalyzeRunsOptions, O as SummarizeExecutionOptions, k as analyzeRuns } from "../tool-groups-CMmsgTzj.js";
|
|
9
|
+
import { A as RunEvalOptions, An as LlmJudgeDimension, B as OptimizerModelBudget, Bt as ComparisonCost, C as RunImprovementLoopOptions, Cn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, D as RunOptimizationOptions, Dn as ReferenceEquivalenceScenario, E as PremeasuredOptimizationBaseline, En as ReferenceEquivalenceJudgeResult, F as GepaOptimizationMethodConfig, Ft as ExternalTextOptimizerContext, G as ObjectiveSource, Gt as OptimizationMethodProvenance, H as AxisVerdict, Ht as OptimizationMethodComparison, I as GepaOptimizationRecipe, It as ExternalTextOptimizerResult, J as PromotionPolicy, K as ParetoSignificanceGateOptions, Kt as OptimizationMethodResult, L as GepaRunnerCommand, Lt as ExternalOptimizationExample, M as GepaAdaptiveEngineRun, Mn as llmJudge, Mt as composeGate, N as GepaEngineOptions, Nt as externalTextOptimizationMethod, On as createReferenceEquivalenceJudge, P as GepaEngineRun, Pt as ExternalTextOptimizationMethodConfig, Qt as compareOptimizationMethods, R as gepaOptimizationMethod, Rt as ExternalTextEvaluationResponse, Sn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, T as runImprovementLoop, Tn as ReferenceEquivalenceJudgeOptions, U as BuildEvidenceVectorOptions, Ut as OptimizationMethodInput, V as AxisEvidence, Vt as OptimizationMethod, W as EvidenceVector, X as paretoPolicy, Xt as OptimizationTokenUsage, Y as buildEvidenceVector, Yt as OptimizationPackageSource, Z as paretoSignificanceGate, _n as campaignSplitDigest, an as RunCampaignOptions, dt as DefaultProductionGateCheck, fn as fsCampaignStorage, ft as DefaultProductionGateOptions, i as skillOptOptimizationMethod, j as runEval, jn as LlmJudgeOptions, kn as runReferenceEquivalenceJudge, lt as HeldOutGateOptions, mt as defaultProductionGate, n as SkillOptRunnerCommand, ot as PowerPreflight, p as LoopProvenanceRecord, pn as inMemoryCampaignStorage, pt as DefaultProductionRewardHackingOptions, q as PromotionObjective, r as SkillOptTrainerConfig, sn as runCampaign, t as SkillOptOptimizationMethodConfig, tn as CampaignCellFailureReceipt, un as CampaignStorage, ut as heldOutGate, w as RunImprovementLoopResult, wn as ReferenceEquivalenceJudgeInput, z as OpenAICompatibleOptimizerModel, zt as CompareOptimizationMethodsOptions } from "../skillopt-optimization-method-CvSJGdm3.js";
|
|
10
|
+
import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-Bbht4xxl.js";
|
|
11
11
|
import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-BYHIuO0e.js";
|
|
12
12
|
import { AgentCandidateBenchmarkCellRef, AgentCandidateBenchmarkSuiteInputs, AgentCandidateBenchmarkTask, AgentCandidateBenchmarkTaskMaterial, AgentCandidateBundle, AgentCandidateEvaluationPolicy, AgentCandidateExperiment, AgentCandidateExperimentMaterial, AgentCandidateExperimentMeasurement, AgentImprovementCost, AgentImprovementMeasuredComparison, AgentProfileImprovementExperiment, AgentProfileImprovementExperimentMaterial, AgentProfileImprovementMeasuredComparison, AgentProfileImprovementMeasurement, AgentProfileImprovementRunCell, AgentProfileImprovementRunReceipt, AgentProfileImprovementSuiteInputs, AgentProfileImprovementTask, AgentProfileImprovementTaskMaterial, CandidateExecutionEvidence, Sha256Digest } from "@tangle-network/agent-interface";
|
|
13
13
|
//#region src/contract/self-improve.d.ts
|
package/dist/contract/index.js
CHANGED
|
@@ -1,18 +1,18 @@
|
|
|
1
1
|
import { c as ValidationError } from "../errors-D-LKuDhb.js";
|
|
2
2
|
import { i as estimateCost, o as isModelPriced } from "../metrics-C9YY1OcL.js";
|
|
3
3
|
import { i as CostLedger } from "../cost-ledger-DMFxsLKr.js";
|
|
4
|
-
import { C as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-
|
|
4
|
+
import { C as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-SOyHB6qG.js";
|
|
5
5
|
import { LLM_MODEL_ATTR_KEYS, SPAN_KIND_ATTR_KEYS } from "../trace-attributes.js";
|
|
6
6
|
import { nt as classifyOtlpSpanRole, rt as isOtlpModelCall } from "../kind-factory-Bvwe3pup.js";
|
|
7
7
|
import { s as makeProposalFinding } from "../usage-receipt-CgxMEBZq.js";
|
|
8
|
-
import { i as inMemoryCampaignStorage, n as createRunCostLedger, r as fsCampaignStorage } from "../single-run-lock-
|
|
9
|
-
import { m as mapConcurrentRange } from "../ledger-core-
|
|
10
|
-
import { A as assertOptimizationResult, B as surfaceContentHash, Ct as decidePairedPromotion, M as compareOptimizationMethods, O as composeGate, S as defaultProductionGate, T as heldoutSignificance, V as surfaceHash, X as resolveRunDir, Y as runCampaign, _ as buildEvidenceVector, a as emitLoopProvenance, at as REFERENCE_EQUIVALENCE_JUDGE_VERSION, b as powerPreflight, d as runImprovementLoop, g as gepaOptimizationMethod, gt as llmJudge, it as REFERENCE_EQUIVALENCE_INPUT_LIMITS, k as externalTextOptimizationMethod, o as loopProvenanceArgsFromResult, ot as createReferenceEquivalenceJudge, p as runEval, st as runReferenceEquivalenceJudge, t as skillOptOptimizationMethod, tt as campaignSplitDigest, v as paretoPolicy, x as heldOutGate, y as paretoSignificanceGate } from "../skillopt-optimization-method-
|
|
8
|
+
import { i as inMemoryCampaignStorage, n as createRunCostLedger, r as fsCampaignStorage } from "../single-run-lock-D5iN0Xzb.js";
|
|
9
|
+
import { m as mapConcurrentRange } from "../ledger-core-DXZIqu17.js";
|
|
10
|
+
import { A as assertOptimizationResult, B as surfaceContentHash, Ct as decidePairedPromotion, M as compareOptimizationMethods, O as composeGate, S as defaultProductionGate, T as heldoutSignificance, V as surfaceHash, X as resolveRunDir, Y as runCampaign, _ as buildEvidenceVector, a as emitLoopProvenance, at as REFERENCE_EQUIVALENCE_JUDGE_VERSION, b as powerPreflight, d as runImprovementLoop, g as gepaOptimizationMethod, gt as llmJudge, it as REFERENCE_EQUIVALENCE_INPUT_LIMITS, k as externalTextOptimizationMethod, o as loopProvenanceArgsFromResult, ot as createReferenceEquivalenceJudge, p as runEval, st as runReferenceEquivalenceJudge, t as skillOptOptimizationMethod, tt as campaignSplitDigest, v as paretoPolicy, x as heldOutGate, y as paretoSignificanceGate } from "../skillopt-optimization-method-C4FX42dy.js";
|
|
11
11
|
import { E as pairedBootstrap } from "../statistics-ByxzSiOM.js";
|
|
12
|
-
import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-
|
|
12
|
+
import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-CWN8-VsV.js";
|
|
13
13
|
import { a as recordAggregateMeasurements, i as readTaskFailureLabels, o as summarizeExecutionMeasurements, s as summarizeTraceErrors, t as extractUsage } from "../extract-usage-7l1Xq5ti.js";
|
|
14
|
-
import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-
|
|
15
|
-
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "../reward-hacking-
|
|
14
|
+
import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-DWIvOAGk.js";
|
|
15
|
+
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "../reward-hacking-CyuzxKly.js";
|
|
16
16
|
import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-ChBKlTd_.js";
|
|
17
17
|
import { t as createHostedClient } from "../client-LIuo-KPv.js";
|
|
18
18
|
import { dirname, join } from "node:path";
|
|
@@ -436,7 +436,9 @@ function cellsToRunRecords(cells, candidateId, runId, surface, splitTag, fallbac
|
|
|
436
436
|
const promptHash = surfaceContentHash(surface);
|
|
437
437
|
const configHash = surfaceContentHash(candidateId);
|
|
438
438
|
return cells.map((cell) => {
|
|
439
|
-
const
|
|
439
|
+
const receiptModels = cell.resolvedModels ?? (cell.resolvedModel ? [cell.resolvedModel] : []);
|
|
440
|
+
if (receiptModels.length > 1) throw new ValidationError(`selfImprove cell ${cell.cellId} used multiple agent models: ${receiptModels.join(", ")}`);
|
|
441
|
+
const model = receiptModels[0] ?? fallbackModel;
|
|
440
442
|
if (!model) throw new ValidationError(`selfImprove.model is required when cell ${cell.cellId} has no paid-call model receipt`);
|
|
441
443
|
if (!modelHasSnapshot(model)) throw new ValidationError(`selfImprove model "${model}" lacks a snapshot version for cell ${cell.cellId}`);
|
|
442
444
|
return campaignCellToRunRecord(cell, {
|