@tangle-network/agent-eval 0.106.3 → 0.107.0
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 +1 -1
- package/dist/adapters/langchain.d.ts +1 -1
- package/dist/adapters/otel.d.ts +1 -1
- package/dist/campaign/index.d.ts +73 -8
- package/dist/campaign/index.js +99 -11
- package/dist/campaign/index.js.map +1 -1
- package/dist/{chunk-2HLM4SJJ.js → chunk-6PF6LXRY.js} +2 -2
- package/dist/{chunk-3E5KUXYZ.js → chunk-G4DLZAV5.js} +21 -1
- package/dist/chunk-G4DLZAV5.js.map +1 -0
- package/dist/{chunk-J22CKVXN.js → chunk-V75NN2ZR.js} +2 -2
- package/dist/contract/index.d.ts +6 -6
- package/dist/contract/index.js +2 -2
- package/dist/{gepa-DQ3ruj18.d.ts → gepa-BmqTrdtg.d.ts} +9 -1
- package/dist/hosted/index.d.ts +1 -1
- package/dist/index.d.ts +4 -4
- package/dist/index.js +2 -2
- package/dist/multishot/index.d.ts +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{pre-registration-BkkTmtQG.d.ts → pre-registration-x3f0VpxT.d.ts} +1 -1
- package/dist/{provenance-Bsyjc67Z.d.ts → provenance-CIjRcI6m.d.ts} +2 -2
- package/dist/rl.d.ts +1 -1
- package/dist/{types-Ctf7XIAL.d.ts → types-CNZ0tHET.d.ts} +10 -0
- package/package.json +1 -1
- package/dist/chunk-3E5KUXYZ.js.map +0 -1
- /package/dist/{chunk-2HLM4SJJ.js.map → chunk-6PF6LXRY.js.map} +0 -0
- /package/dist/{chunk-J22CKVXN.js.map → chunk-V75NN2ZR.js.map} +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import {
|
|
2
2
|
recoverTruncatedJson
|
|
3
|
-
} from "./chunk-
|
|
3
|
+
} from "./chunk-G4DLZAV5.js";
|
|
4
4
|
import {
|
|
5
5
|
clamp01
|
|
6
6
|
} from "./chunk-LOJ2QVCE.js";
|
|
@@ -858,4 +858,4 @@ export {
|
|
|
858
858
|
agentProfileModelId,
|
|
859
859
|
agentProfileHash
|
|
860
860
|
};
|
|
861
|
-
//# sourceMappingURL=chunk-
|
|
861
|
+
//# sourceMappingURL=chunk-6PF6LXRY.js.map
|
|
@@ -1619,6 +1619,24 @@ async function runImprovementLoop(opts) {
|
|
|
1619
1619
|
baselineArtifacts.set(cell.cellId, cell.artifact);
|
|
1620
1620
|
baselineJudgeScores.set(cell.cellId, cell.judgeScores);
|
|
1621
1621
|
}
|
|
1622
|
+
let neutralizedArtifacts;
|
|
1623
|
+
let neutralizedJudgeScores;
|
|
1624
|
+
if (opts.neutralize && !winnerIsBaseline) {
|
|
1625
|
+
const neutralizedSurface = opts.neutralize(optimization.winnerSurface, opts.baselineSurface);
|
|
1626
|
+
const neutralizedOnHoldout = await runCampaign2({
|
|
1627
|
+
...opts,
|
|
1628
|
+
dispatchTimeoutMs,
|
|
1629
|
+
scenarios: opts.holdoutScenarios,
|
|
1630
|
+
dispatch: (scenario, ctx) => opts.dispatchWithSurface(neutralizedSurface, scenario, ctx),
|
|
1631
|
+
runDir: `${opts.runDir}/holdout-neutralized`
|
|
1632
|
+
});
|
|
1633
|
+
neutralizedArtifacts = /* @__PURE__ */ new Map();
|
|
1634
|
+
neutralizedJudgeScores = /* @__PURE__ */ new Map();
|
|
1635
|
+
for (const cell of neutralizedOnHoldout.cells) {
|
|
1636
|
+
neutralizedArtifacts.set(cell.cellId, cell.artifact);
|
|
1637
|
+
neutralizedJudgeScores.set(cell.cellId, cell.judgeScores);
|
|
1638
|
+
}
|
|
1639
|
+
}
|
|
1622
1640
|
const gateResult = winnerIsBaseline ? {
|
|
1623
1641
|
decision: "hold",
|
|
1624
1642
|
reasons: [
|
|
@@ -1633,6 +1651,8 @@ async function runImprovementLoop(opts) {
|
|
|
1633
1651
|
baselineArtifacts,
|
|
1634
1652
|
judgeScores,
|
|
1635
1653
|
baselineJudgeScores,
|
|
1654
|
+
neutralizedArtifacts,
|
|
1655
|
+
neutralizedJudgeScores,
|
|
1636
1656
|
scenarios: opts.holdoutScenarios,
|
|
1637
1657
|
cost: {
|
|
1638
1658
|
candidate: winnerOnHoldout.aggregates.totalCostUsd,
|
|
@@ -2157,4 +2177,4 @@ export {
|
|
|
2157
2177
|
provenanceSpansPath,
|
|
2158
2178
|
emitLoopProvenance
|
|
2159
2179
|
};
|
|
2160
|
-
//# sourceMappingURL=chunk-
|
|
2180
|
+
//# sourceMappingURL=chunk-G4DLZAV5.js.map
|