@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/client/index.html
CHANGED
|
@@ -22,8 +22,8 @@
|
|
|
22
22
|
}
|
|
23
23
|
})();
|
|
24
24
|
</script>
|
|
25
|
-
<script type="module" crossorigin src="./assets/index-
|
|
26
|
-
<link rel="stylesheet" crossorigin href="./assets/index-
|
|
25
|
+
<script type="module" crossorigin src="./assets/index-CyKf66AB.js"></script>
|
|
26
|
+
<link rel="stylesheet" crossorigin href="./assets/index-CxBvTZU2.css">
|
|
27
27
|
</head>
|
|
28
28
|
<body>
|
|
29
29
|
<div id="root"></div>
|
package/dist/middleware.cjs
CHANGED
|
@@ -141,7 +141,10 @@ function redactEvalItem(item) {
|
|
|
141
141
|
detailsOut[name] = {
|
|
142
142
|
score: detail.score,
|
|
143
143
|
...detail.duration !== void 0 ? { duration: detail.duration } : {},
|
|
144
|
-
...detail.cost !== void 0 ? { cost: detail.cost } : {}
|
|
144
|
+
...detail.cost !== void 0 ? { cost: detail.cost } : {},
|
|
145
|
+
// `skipped` is a structural boolean (the `applies` predicate verdict),
|
|
146
|
+
// not user/LLM content — preserve it so the client's N/A chip renders.
|
|
147
|
+
...detail.skipped !== void 0 ? { skipped: detail.skipped } : {}
|
|
145
148
|
// metadata deliberately omitted — may contain LLM scorer reasoning
|
|
146
149
|
};
|
|
147
150
|
}
|
|
@@ -1835,8 +1838,13 @@ function createEvalRoutes(connMgr, evalLoader) {
|
|
|
1835
1838
|
const first = results[0];
|
|
1836
1839
|
const partial2 = results.length < runs;
|
|
1837
1840
|
const failureMsg = runFailure ? redactErrorMessage(runFailure, redactOn) || String(runFailure) || void 0 : void 0;
|
|
1841
|
+
const aggDegraded = unionDegradedScorers(results);
|
|
1838
1842
|
const result = {
|
|
1839
1843
|
...first,
|
|
1844
|
+
summary: {
|
|
1845
|
+
...first.summary,
|
|
1846
|
+
...aggDegraded.length > 0 ? { degraded: aggDegraded } : {}
|
|
1847
|
+
},
|
|
1840
1848
|
_multiRun: {
|
|
1841
1849
|
aggregate,
|
|
1842
1850
|
allRuns: results,
|
|
@@ -2094,6 +2102,23 @@ function createEvalRoutes(connMgr, evalLoader) {
|
|
|
2094
2102
|
}
|
|
2095
2103
|
return { app: app5, closeActiveRuns };
|
|
2096
2104
|
}
|
|
2105
|
+
function unionDegradedScorers(results) {
|
|
2106
|
+
const byScorer = /* @__PURE__ */ new Map();
|
|
2107
|
+
for (const run of results) {
|
|
2108
|
+
const degraded = run.summary?.degraded;
|
|
2109
|
+
if (!Array.isArray(degraded)) continue;
|
|
2110
|
+
for (const d of degraded) {
|
|
2111
|
+
const existing = byScorer.get(d.scorer);
|
|
2112
|
+
if (!existing) {
|
|
2113
|
+
byScorer.set(d.scorer, { ...d, runsAffected: 1 });
|
|
2114
|
+
} else {
|
|
2115
|
+
const worse = d.rate > existing.rate ? d : existing;
|
|
2116
|
+
byScorer.set(d.scorer, { ...worse, runsAffected: existing.runsAffected + 1 });
|
|
2117
|
+
}
|
|
2118
|
+
}
|
|
2119
|
+
}
|
|
2120
|
+
return [...byScorer.values()];
|
|
2121
|
+
}
|
|
2097
2122
|
|
|
2098
2123
|
// src/server/routes/playground.ts
|
|
2099
2124
|
var import_hono11 = require("hono");
|