@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/cli.cjs
CHANGED
|
@@ -134,7 +134,10 @@ function redactEvalItem(item) {
|
|
|
134
134
|
detailsOut[name] = {
|
|
135
135
|
score: detail.score,
|
|
136
136
|
...detail.duration !== void 0 ? { duration: detail.duration } : {},
|
|
137
|
-
...detail.cost !== void 0 ? { cost: detail.cost } : {}
|
|
137
|
+
...detail.cost !== void 0 ? { cost: detail.cost } : {},
|
|
138
|
+
// `skipped` is a structural boolean (the `applies` predicate verdict),
|
|
139
|
+
// not user/LLM content — preserve it so the client's N/A chip renders.
|
|
140
|
+
...detail.skipped !== void 0 ? { skipped: detail.skipped } : {}
|
|
138
141
|
// metadata deliberately omitted — may contain LLM scorer reasoning
|
|
139
142
|
};
|
|
140
143
|
}
|
|
@@ -1828,8 +1831,13 @@ function createEvalRoutes(connMgr, evalLoader) {
|
|
|
1828
1831
|
const first = results[0];
|
|
1829
1832
|
const partial2 = results.length < runs;
|
|
1830
1833
|
const failureMsg = runFailure ? redactErrorMessage(runFailure, redactOn) || String(runFailure) || void 0 : void 0;
|
|
1834
|
+
const aggDegraded = unionDegradedScorers(results);
|
|
1831
1835
|
const result = {
|
|
1832
1836
|
...first,
|
|
1837
|
+
summary: {
|
|
1838
|
+
...first.summary,
|
|
1839
|
+
...aggDegraded.length > 0 ? { degraded: aggDegraded } : {}
|
|
1840
|
+
},
|
|
1833
1841
|
_multiRun: {
|
|
1834
1842
|
aggregate,
|
|
1835
1843
|
allRuns: results,
|
|
@@ -2087,6 +2095,23 @@ function createEvalRoutes(connMgr, evalLoader) {
|
|
|
2087
2095
|
}
|
|
2088
2096
|
return { app: app5, closeActiveRuns };
|
|
2089
2097
|
}
|
|
2098
|
+
function unionDegradedScorers(results) {
|
|
2099
|
+
const byScorer = /* @__PURE__ */ new Map();
|
|
2100
|
+
for (const run of results) {
|
|
2101
|
+
const degraded = run.summary?.degraded;
|
|
2102
|
+
if (!Array.isArray(degraded)) continue;
|
|
2103
|
+
for (const d of degraded) {
|
|
2104
|
+
const existing = byScorer.get(d.scorer);
|
|
2105
|
+
if (!existing) {
|
|
2106
|
+
byScorer.set(d.scorer, { ...d, runsAffected: 1 });
|
|
2107
|
+
} else {
|
|
2108
|
+
const worse = d.rate > existing.rate ? d : existing;
|
|
2109
|
+
byScorer.set(d.scorer, { ...worse, runsAffected: existing.runsAffected + 1 });
|
|
2110
|
+
}
|
|
2111
|
+
}
|
|
2112
|
+
}
|
|
2113
|
+
return [...byScorer.values()];
|
|
2114
|
+
}
|
|
2090
2115
|
|
|
2091
2116
|
// src/server/routes/playground.ts
|
|
2092
2117
|
var import_hono11 = require("hono");
|