@tangle-network/agent-eval 0.102.1 → 0.103.1
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/dist/adapters/http.d.ts +2 -2
- package/dist/adapters/langchain.d.ts +2 -2
- package/dist/adapters/otel.d.ts +5 -5
- package/dist/analyst/index.d.ts +9 -9
- package/dist/analyst/index.js +3 -3
- package/dist/{analyze-runs-Cd-A_K4l.d.ts → analyze-runs-rF2eGxZ_.d.ts} +3 -3
- package/dist/belief-state/index.d.ts +3 -3
- package/dist/belief-state/index.js +1 -1
- package/dist/benchmarks/index.d.ts +2 -2
- package/dist/builder-eval/index.js +2 -2
- package/dist/campaign/index.d.ts +43 -16
- package/dist/campaign/index.js +9 -9
- package/dist/campaign/index.js.map +1 -1
- package/dist/{chunk-Z3FLN24V.js → chunk-2B4HPFXN.js} +104 -43
- package/dist/chunk-2B4HPFXN.js.map +1 -0
- package/dist/{chunk-7RBJANJD.js → chunk-3722PKPB.js} +2 -2
- package/dist/{chunk-YLKDN7JV.js → chunk-66T25VWE.js} +5 -5
- package/dist/chunk-66T25VWE.js.map +1 -0
- package/dist/{chunk-6FIAJHCU.js → chunk-6YVAN6R3.js} +4 -4
- package/dist/{chunk-ABOIVNXL.js → chunk-AGYDMORK.js} +1 -1
- package/dist/chunk-AGYDMORK.js.map +1 -0
- package/dist/{chunk-B2TMQM62.js → chunk-BHCFJGL4.js} +2 -2
- package/dist/{chunk-IH7LYRHL.js → chunk-BXZVBX3D.js} +3 -3
- package/dist/{chunk-6GT4NI4V.js → chunk-CLS3374R.js} +2 -2
- package/dist/{chunk-HRGUJTER.js → chunk-HYGRFL7C.js} +1 -1
- package/dist/{chunk-HRGUJTER.js.map → chunk-HYGRFL7C.js.map} +1 -1
- package/dist/{chunk-2NSLDY4B.js → chunk-J5MUFXEY.js} +2 -2
- package/dist/{chunk-UFSG7ACU.js → chunk-JZXGWLK5.js} +1 -1
- package/dist/chunk-JZXGWLK5.js.map +1 -0
- package/dist/{chunk-IN3SHQML.js → chunk-LMSQ6EFA.js} +2 -2
- package/dist/{chunk-AIGWQEME.js → chunk-LOZOZYHU.js} +2 -2
- package/dist/{chunk-NF7OZ4J7.js → chunk-QCBB6ZIU.js} +2 -2
- package/dist/chunk-RQNOLV3I.js +855 -0
- package/dist/chunk-RQNOLV3I.js.map +1 -0
- package/dist/{chunk-JU6ZX3CX.js → chunk-SMCACT4Z.js} +2 -2
- package/dist/{chunk-6Q2DYRWV.js → chunk-TCPP4Z5Q.js} +5 -5
- package/dist/{chunk-VDGPPGE3.js → chunk-UMEAR2FI.js} +2 -2
- package/dist/{chunk-VDGPPGE3.js.map → chunk-UMEAR2FI.js.map} +1 -1
- package/dist/{chunk-QIT2XZ4E.js → chunk-VNM52AGA.js} +3 -3
- package/dist/chunk-VNM52AGA.js.map +1 -0
- package/dist/{chunk-CMJSTXUR.js → chunk-YFYIOSNV.js} +107 -20
- package/dist/chunk-YFYIOSNV.js.map +1 -0
- package/dist/{code-agent-session-Ce-9u7YM.d.ts → code-agent-session-Cen-qD2y.d.ts} +1 -1
- package/dist/contract/index.d.ts +18 -18
- package/dist/contract/index.js +10 -10
- package/dist/{control-C8RmK9H4.d.ts → control-KofK3gfG.d.ts} +1 -1
- package/dist/control.d.ts +2 -2
- package/dist/control.js +3 -3
- package/dist/{corpus-CiSzzLa5.d.ts → corpus-CysLCiK4.d.ts} +1 -1
- package/dist/{default-registry-ZhqsTr4K.d.ts → default-registry-Ez4cxuZ4.d.ts} +2 -2
- package/dist/diagnose.d.ts +3 -3
- package/dist/diagnose.js +3 -3
- package/dist/{gepa-DeyPTlvx.d.ts → gepa-COlCAkHN.d.ts} +22 -1
- package/dist/governance/index.d.ts +1 -1
- package/dist/hosted/index.d.ts +5 -5
- package/dist/{index-B-bFgiAF.d.ts → index-unCSYRJJ.d.ts} +1 -1
- package/dist/index.d.ts +36 -28
- package/dist/index.js +32 -18
- package/dist/index.js.map +1 -1
- package/dist/{insight-report-k0sRTzKg.d.ts → insight-report-DumCfEur.d.ts} +2 -2
- package/dist/{judge-calibration-0p2QcWNE.d.ts → judge-calibration-7C-IDmKr.d.ts} +3 -0
- package/dist/{kind-factory-D0nk7AKV.d.ts → kind-factory-BLHxwX71.d.ts} +1 -1
- package/dist/meta-eval/index.d.ts +4 -4
- package/dist/meta-eval/index.js +3 -3
- package/dist/{multi-layer-verifier-DUZXrPDA.d.ts → multi-layer-verifier-CI4jdX-q.d.ts} +3 -0
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/pipelines/index.d.ts +1 -1
- package/dist/pipelines/index.js +3 -3
- package/dist/{policy-edit-BDQzzsBU.d.ts → policy-edit-nUhuFtLF.d.ts} +2 -2
- package/dist/{pre-registration-Dzg61IQA.d.ts → pre-registration-CUOSGAZK.d.ts} +45 -12
- package/dist/product-benchmark/index.d.ts +104 -1
- package/dist/product-benchmark/index.js +15 -1
- package/dist/{provenance-BhJm32vN.d.ts → provenance-B2VsP0jP.d.ts} +20 -4
- package/dist/{query-B7GGjRox.d.ts → query-0aTmbmQe.d.ts} +1 -0
- package/dist/{release-report-BQ1Ziyu-.d.ts → release-report-DkaCZ9k4.d.ts} +5 -2
- package/dist/reporting.d.ts +6 -6
- package/dist/reporting.js +4 -4
- package/dist/{researcher-B_ODTAJs.d.ts → researcher-SDAezfML.d.ts} +2 -2
- package/dist/rl.d.ts +9 -9
- package/dist/rl.js +7 -7
- package/dist/{rubric-predictive-validity-0MdjTt8R.d.ts → rubric-predictive-validity-ZGIlJNce.d.ts} +1 -1
- package/dist/{run-campaign-3NWW5PLF.js → run-campaign-DAHKO5CT.js} +3 -3
- package/dist/{run-record-MRdJ-Kq2.d.ts → run-record-CPfd1ARZ.d.ts} +3 -0
- package/dist/{runtime-trajectory-8w0_jmtR.d.ts → runtime-trajectory-DZ8ei-Jo.d.ts} +1 -1
- package/dist/{semantic-concept-judge-D-IlH5v1.d.ts → semantic-concept-judge-C6mDEBIo.d.ts} +3 -3
- package/dist/{statistics-xP-cWc5k.d.ts → statistics-D88peojY.d.ts} +1 -1
- package/dist/{summary-report-C0nnxOD8.d.ts → summary-report-Fc_YFJat.d.ts} +1 -1
- package/dist/traces.d.ts +2 -2
- package/dist/traces.js +4 -4
- package/dist/{types-DFI_Z-ZL.d.ts → types-Bihq6-a3.d.ts} +1 -1
- package/dist/{types-Dz9cKF0g.d.ts → types-DA9yj-Jd.d.ts} +1 -1
- package/dist/workflow/index.d.ts +7 -7
- package/dist/workflow/index.js +3 -3
- package/package.json +1 -1
- package/dist/chunk-63MBSQTX.js +0 -350
- package/dist/chunk-63MBSQTX.js.map +0 -1
- package/dist/chunk-ABOIVNXL.js.map +0 -1
- package/dist/chunk-CMJSTXUR.js.map +0 -1
- package/dist/chunk-QIT2XZ4E.js.map +0 -1
- package/dist/chunk-UFSG7ACU.js.map +0 -1
- package/dist/chunk-YLKDN7JV.js.map +0 -1
- package/dist/chunk-Z3FLN24V.js.map +0 -1
- /package/dist/{chunk-7RBJANJD.js.map → chunk-3722PKPB.js.map} +0 -0
- /package/dist/{chunk-6FIAJHCU.js.map → chunk-6YVAN6R3.js.map} +0 -0
- /package/dist/{chunk-B2TMQM62.js.map → chunk-BHCFJGL4.js.map} +0 -0
- /package/dist/{chunk-IH7LYRHL.js.map → chunk-BXZVBX3D.js.map} +0 -0
- /package/dist/{chunk-6GT4NI4V.js.map → chunk-CLS3374R.js.map} +0 -0
- /package/dist/{chunk-2NSLDY4B.js.map → chunk-J5MUFXEY.js.map} +0 -0
- /package/dist/{chunk-IN3SHQML.js.map → chunk-LMSQ6EFA.js.map} +0 -0
- /package/dist/{chunk-AIGWQEME.js.map → chunk-LOZOZYHU.js.map} +0 -0
- /package/dist/{chunk-NF7OZ4J7.js.map → chunk-QCBB6ZIU.js.map} +0 -0
- /package/dist/{chunk-JU6ZX3CX.js.map → chunk-SMCACT4Z.js.map} +0 -0
- /package/dist/{chunk-6Q2DYRWV.js.map → chunk-TCPP4Z5Q.js.map} +0 -0
- /package/dist/{run-campaign-3NWW5PLF.js.map → run-campaign-DAHKO5CT.js.map} +0 -0
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import {
|
|
2
2
|
runCampaign,
|
|
3
3
|
summarizeBackendIntegrity
|
|
4
|
-
} from "./chunk-
|
|
4
|
+
} from "./chunk-VNM52AGA.js";
|
|
5
5
|
import {
|
|
6
6
|
callLlm
|
|
7
7
|
} from "./chunk-CWNP4DV4.js";
|
|
@@ -764,6 +764,106 @@ function mean(xs) {
|
|
|
764
764
|
return xs.reduce((s, x) => s + x, 0) / xs.length;
|
|
765
765
|
}
|
|
766
766
|
|
|
767
|
+
// src/json-recovery.ts
|
|
768
|
+
function autoCloseTruncatedJson(raw) {
|
|
769
|
+
const stack = [];
|
|
770
|
+
let inString = false;
|
|
771
|
+
let escaped = false;
|
|
772
|
+
for (const c of raw) {
|
|
773
|
+
if (escaped) {
|
|
774
|
+
escaped = false;
|
|
775
|
+
continue;
|
|
776
|
+
}
|
|
777
|
+
if (inString) {
|
|
778
|
+
if (c === "\\") {
|
|
779
|
+
escaped = true;
|
|
780
|
+
continue;
|
|
781
|
+
}
|
|
782
|
+
if (c === '"') {
|
|
783
|
+
inString = false;
|
|
784
|
+
continue;
|
|
785
|
+
}
|
|
786
|
+
continue;
|
|
787
|
+
}
|
|
788
|
+
if (c === '"') {
|
|
789
|
+
inString = true;
|
|
790
|
+
continue;
|
|
791
|
+
}
|
|
792
|
+
if (c === "{" || c === "[") stack.push(c);
|
|
793
|
+
else if (c === "}") {
|
|
794
|
+
if (stack.pop() !== "{") return null;
|
|
795
|
+
} else if (c === "]") {
|
|
796
|
+
if (stack.pop() !== "[") return null;
|
|
797
|
+
}
|
|
798
|
+
}
|
|
799
|
+
if (stack.length === 0 && !inString) return raw;
|
|
800
|
+
let suffix = "";
|
|
801
|
+
if (escaped) suffix += "\\";
|
|
802
|
+
if (inString) suffix += '"';
|
|
803
|
+
while (stack.length > 0) {
|
|
804
|
+
const opener = stack.pop();
|
|
805
|
+
suffix += opener === "{" ? "}" : "]";
|
|
806
|
+
}
|
|
807
|
+
return raw + suffix;
|
|
808
|
+
}
|
|
809
|
+
var UNPARSEABLE = /* @__PURE__ */ Symbol("unparseable");
|
|
810
|
+
function tryParse(candidate) {
|
|
811
|
+
try {
|
|
812
|
+
return JSON.parse(candidate);
|
|
813
|
+
} catch {
|
|
814
|
+
return UNPARSEABLE;
|
|
815
|
+
}
|
|
816
|
+
}
|
|
817
|
+
function lastCommaOutsideString(s) {
|
|
818
|
+
let inString = false;
|
|
819
|
+
let escaped = false;
|
|
820
|
+
let last = -1;
|
|
821
|
+
for (let i = 0; i < s.length; i++) {
|
|
822
|
+
const c = s[i];
|
|
823
|
+
if (escaped) {
|
|
824
|
+
escaped = false;
|
|
825
|
+
continue;
|
|
826
|
+
}
|
|
827
|
+
if (inString) {
|
|
828
|
+
if (c === "\\") escaped = true;
|
|
829
|
+
else if (c === '"') inString = false;
|
|
830
|
+
continue;
|
|
831
|
+
}
|
|
832
|
+
if (c === '"') inString = true;
|
|
833
|
+
else if (c === ",") last = i;
|
|
834
|
+
}
|
|
835
|
+
return last;
|
|
836
|
+
}
|
|
837
|
+
function recoverTruncatedJson(text) {
|
|
838
|
+
const objStart = text.indexOf("{");
|
|
839
|
+
const arrStart = text.indexOf("[");
|
|
840
|
+
const starts = [objStart, arrStart].filter((i) => i >= 0).sort((a, b) => a - b);
|
|
841
|
+
for (const start of starts) {
|
|
842
|
+
const recovered = recoverFrom(text.slice(start));
|
|
843
|
+
if (recovered !== UNPARSEABLE) return recovered;
|
|
844
|
+
}
|
|
845
|
+
return null;
|
|
846
|
+
}
|
|
847
|
+
function recoverFrom(slice) {
|
|
848
|
+
let candidate = slice;
|
|
849
|
+
const lastClose = Math.max(candidate.lastIndexOf("}"), candidate.lastIndexOf("]"));
|
|
850
|
+
if (lastClose > 0) {
|
|
851
|
+
const balanced = tryParse(candidate.slice(0, lastClose + 1));
|
|
852
|
+
if (balanced !== UNPARSEABLE) return balanced;
|
|
853
|
+
}
|
|
854
|
+
for (let i = 0; i < 64; i++) {
|
|
855
|
+
const closed = autoCloseTruncatedJson(candidate);
|
|
856
|
+
if (closed !== null) {
|
|
857
|
+
const parsed = tryParse(closed);
|
|
858
|
+
if (parsed !== UNPARSEABLE) return parsed;
|
|
859
|
+
}
|
|
860
|
+
const cut = lastCommaOutsideString(candidate);
|
|
861
|
+
if (cut <= 0) return UNPARSEABLE;
|
|
862
|
+
candidate = candidate.slice(0, cut);
|
|
863
|
+
}
|
|
864
|
+
return UNPARSEABLE;
|
|
865
|
+
}
|
|
866
|
+
|
|
767
867
|
// src/reflective-mutation.ts
|
|
768
868
|
var DEFAULT_MUTATION_PRIMITIVES = [
|
|
769
869
|
'Strengthen an imperative ("should" \u2192 "must")',
|
|
@@ -888,46 +988,6 @@ function truncate(s, max) {
|
|
|
888
988
|
function quote(s) {
|
|
889
989
|
return s.replace(/`/g, "\\`");
|
|
890
990
|
}
|
|
891
|
-
function autoCloseTruncatedJson(raw) {
|
|
892
|
-
const stack = [];
|
|
893
|
-
let inString = false;
|
|
894
|
-
let escaped = false;
|
|
895
|
-
for (const c of raw) {
|
|
896
|
-
if (escaped) {
|
|
897
|
-
escaped = false;
|
|
898
|
-
continue;
|
|
899
|
-
}
|
|
900
|
-
if (inString) {
|
|
901
|
-
if (c === "\\") {
|
|
902
|
-
escaped = true;
|
|
903
|
-
continue;
|
|
904
|
-
}
|
|
905
|
-
if (c === '"') {
|
|
906
|
-
inString = false;
|
|
907
|
-
continue;
|
|
908
|
-
}
|
|
909
|
-
continue;
|
|
910
|
-
}
|
|
911
|
-
if (c === '"') {
|
|
912
|
-
inString = true;
|
|
913
|
-
continue;
|
|
914
|
-
}
|
|
915
|
-
if (c === "{" || c === "[") stack.push(c);
|
|
916
|
-
else if (c === "}") {
|
|
917
|
-
if (stack.pop() !== "{") return null;
|
|
918
|
-
} else if (c === "]") {
|
|
919
|
-
if (stack.pop() !== "[") return null;
|
|
920
|
-
}
|
|
921
|
-
}
|
|
922
|
-
if (stack.length === 0 && !inString) return raw;
|
|
923
|
-
let suffix = "";
|
|
924
|
-
if (inString) suffix += '"';
|
|
925
|
-
while (stack.length > 0) {
|
|
926
|
-
const opener = stack.pop();
|
|
927
|
-
suffix += opener === "{" ? "}" : "]";
|
|
928
|
-
}
|
|
929
|
-
return raw + suffix;
|
|
930
|
-
}
|
|
931
991
|
function parseReflectionResponse(raw, maxProposals) {
|
|
932
992
|
let text = raw.trim();
|
|
933
993
|
if (text.startsWith("```")) text = text.replace(/^```(?:json)?\n?/, "").replace(/\n?```$/, "");
|
|
@@ -1400,7 +1460,7 @@ async function runImprovementLoop(opts) {
|
|
|
1400
1460
|
const dispatchTimeoutMs = opts.dispatchTimeoutMs ?? DEFAULT_DISPATCH_TIMEOUT_MS;
|
|
1401
1461
|
const optimization = await runOptimization({ ...opts, dispatchTimeoutMs });
|
|
1402
1462
|
const winnerIsBaseline = optimization.winnerSurfaceHash === surfaceHash(opts.baselineSurface);
|
|
1403
|
-
const { runCampaign: runCampaign2 } = await import("./run-campaign-
|
|
1463
|
+
const { runCampaign: runCampaign2 } = await import("./run-campaign-DAHKO5CT.js");
|
|
1404
1464
|
const baselineOnHoldout = await runCampaign2({
|
|
1405
1465
|
...opts,
|
|
1406
1466
|
dispatchTimeoutMs,
|
|
@@ -1930,6 +1990,7 @@ async function emitLoopProvenance(args) {
|
|
|
1930
1990
|
}
|
|
1931
1991
|
|
|
1932
1992
|
export {
|
|
1993
|
+
recoverTruncatedJson,
|
|
1933
1994
|
dominates,
|
|
1934
1995
|
paretoFrontier,
|
|
1935
1996
|
scalarScore,
|
|
@@ -1968,4 +2029,4 @@ export {
|
|
|
1968
2029
|
provenanceSpansPath,
|
|
1969
2030
|
emitLoopProvenance
|
|
1970
2031
|
};
|
|
1971
|
-
//# sourceMappingURL=chunk-
|
|
2032
|
+
//# sourceMappingURL=chunk-2B4HPFXN.js.map
|