@axlsdk/studio 0.17.9 → 0.18.1
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/README.md +51 -385
- package/dist/{chunk-WUCCIBQ6.js → chunk-RLG3DDQZ.js} +27 -2
- package/dist/chunk-RLG3DDQZ.js.map +1 -0
- package/dist/cli.cjs +26 -1
- package/dist/cli.cjs.map +1 -1
- package/dist/cli.js +1 -1
- package/dist/client/assets/{index-BgDbwM46.css → index-CxBvTZU2.css} +1 -1
- package/dist/client/assets/index-CyKf66AB.js +313 -0
- package/dist/client/index.html +2 -2
- package/dist/middleware.cjs +26 -1
- package/dist/middleware.cjs.map +1 -1
- package/dist/middleware.js +1 -1
- package/dist/server/index.cjs +26 -1
- package/dist/server/index.cjs.map +1 -1
- package/dist/server/index.js +1 -1
- package/package.json +4 -4
- package/dist/chunk-WUCCIBQ6.js.map +0 -1
- package/dist/client/assets/index-Cskx93hn.js +0 -313
package/dist/middleware.js
CHANGED
package/dist/server/index.cjs
CHANGED
|
@@ -137,7 +137,10 @@ function redactEvalItem(item) {
|
|
|
137
137
|
detailsOut[name] = {
|
|
138
138
|
score: detail.score,
|
|
139
139
|
...detail.duration !== void 0 ? { duration: detail.duration } : {},
|
|
140
|
-
...detail.cost !== void 0 ? { cost: detail.cost } : {}
|
|
140
|
+
...detail.cost !== void 0 ? { cost: detail.cost } : {},
|
|
141
|
+
// `skipped` is a structural boolean (the `applies` predicate verdict),
|
|
142
|
+
// not user/LLM content — preserve it so the client's N/A chip renders.
|
|
143
|
+
...detail.skipped !== void 0 ? { skipped: detail.skipped } : {}
|
|
141
144
|
// metadata deliberately omitted — may contain LLM scorer reasoning
|
|
142
145
|
};
|
|
143
146
|
}
|
|
@@ -1831,8 +1834,13 @@ function createEvalRoutes(connMgr, evalLoader) {
|
|
|
1831
1834
|
const first = results[0];
|
|
1832
1835
|
const partial2 = results.length < runs;
|
|
1833
1836
|
const failureMsg = runFailure ? redactErrorMessage(runFailure, redactOn) || String(runFailure) || void 0 : void 0;
|
|
1837
|
+
const aggDegraded = unionDegradedScorers(results);
|
|
1834
1838
|
const result = {
|
|
1835
1839
|
...first,
|
|
1840
|
+
summary: {
|
|
1841
|
+
...first.summary,
|
|
1842
|
+
...aggDegraded.length > 0 ? { degraded: aggDegraded } : {}
|
|
1843
|
+
},
|
|
1836
1844
|
_multiRun: {
|
|
1837
1845
|
aggregate,
|
|
1838
1846
|
allRuns: results,
|
|
@@ -2090,6 +2098,23 @@ function createEvalRoutes(connMgr, evalLoader) {
|
|
|
2090
2098
|
}
|
|
2091
2099
|
return { app: app5, closeActiveRuns };
|
|
2092
2100
|
}
|
|
2101
|
+
function unionDegradedScorers(results) {
|
|
2102
|
+
const byScorer = /* @__PURE__ */ new Map();
|
|
2103
|
+
for (const run of results) {
|
|
2104
|
+
const degraded = run.summary?.degraded;
|
|
2105
|
+
if (!Array.isArray(degraded)) continue;
|
|
2106
|
+
for (const d of degraded) {
|
|
2107
|
+
const existing = byScorer.get(d.scorer);
|
|
2108
|
+
if (!existing) {
|
|
2109
|
+
byScorer.set(d.scorer, { ...d, runsAffected: 1 });
|
|
2110
|
+
} else {
|
|
2111
|
+
const worse = d.rate > existing.rate ? d : existing;
|
|
2112
|
+
byScorer.set(d.scorer, { ...worse, runsAffected: existing.runsAffected + 1 });
|
|
2113
|
+
}
|
|
2114
|
+
}
|
|
2115
|
+
}
|
|
2116
|
+
return [...byScorer.values()];
|
|
2117
|
+
}
|
|
2093
2118
|
|
|
2094
2119
|
// src/server/routes/playground.ts
|
|
2095
2120
|
var import_hono11 = require("hono");
|