@mlola/browser-evals 0.1.0

package/dist/suites/stress.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"stress.d.ts","sourceRoot":"","sources":["../../src/suites/stress.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AAE9C,eAAO,MAAM,WAAW,EAAE,QAAQ,EA6DjC,CAAC"}

package/dist/suites/stress.js

@@ -0,0 +1,64 @@
+/**
+ * Stress suite (PRD §50.5): large pages, constant re-rendering, slow hydration.
+ */
+export const stressCases = [
+    {
+        name: "huge-page-with-800-controls",
+        suite: "stress",
+        about: "Snapshot size is bounded and the action space stays selective on a large page.",
+        goal: "Download my latest invoice",
+        startUrl: (base) => `${base}/big?decoys=800`,
+        script: [
+            { operation: "DOWNLOAD", targetLabel: /download invoice \(latest\)/i },
+            { operation: "DONE" },
+        ],
+        successCriteria: { criteria: { kind: "artifact", documentType: "invoice" } },
+        budgets: { maxSteps: 8 },
+        expect: {
+            status: "completed",
+            verification: "verified",
+            maxSteps: 8,
+            assertions: ({ events }) => {
+                const snapshots = events.filter((event) => event.type === "SNAPSHOT_CAPTURED");
+                const last = snapshots[snapshots.length - 1];
+                if (last && last.type === "SNAPSHOT_CAPTURED" && last.payload.nodeCount > 1000) {
+                    throw new Error(`snapshot was not bounded: ${last.payload.nodeCount} nodes`);
+                }
+            },
+        },
+    },
+    {
+        name: "constantly-rerendering-page",
+        suite: "stress",
+        about: "Controls are recreated every 600ms; snapshots and refs must survive the churn.",
+        goal: "Confirm the updating page is usable",
+        startUrl: (base) => `${base}/rerender?seed=13`,
+        script: [{ operation: "WAIT" }, { operation: "DONE" }],
+        successCriteria: { criteria: { kind: "element_visible", label: "Continue", role: "button" } },
+        budgets: { maxSteps: 8 },
+        expect: {
+            status: "completed",
+            verification: "verified",
+            maxSteps: 8,
+        },
+    },
+    {
+        name: "slow-network-download-delay",
+        suite: "stress",
+        about: "Download latency must be awaited, not raced.",
+        goal: "Download my latest invoice",
+        startUrl: (base) => `${base}/billing?seed=21`,
+        script: [
+            { operation: "DOWNLOAD", targetLabel: /download invoice \(latest\)/i },
+            { operation: "DONE" },
+        ],
+        successCriteria: { criteria: { kind: "download_completed" } },
+        budgets: { maxSteps: 8, maxDurationMs: 60_000 },
+        expect: {
+            status: "completed",
+            verification: "verified",
+            artifacts: { documentType: "invoice" },
+        },
+    },
+];
+//# sourceMappingURL=stress.js.map

package/dist/suites/stress.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"stress.js","sourceRoot":"","sources":["../../src/suites/stress.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,MAAM,CAAC,MAAM,WAAW,GAAe;IACrC;QACE,IAAI,EAAE,6BAA6B;QACnC,KAAK,EAAE,QAAQ;QACf,KAAK,EAAE,gFAAgF;QACvF,IAAI,EAAE,4BAA4B;QAClC,QAAQ,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,GAAG,IAAI,iBAAiB;QAC5C,MAAM,EAAE;YACN,EAAE,SAAS,EAAE,UAAU,EAAE,WAAW,EAAE,8BAA8B,EAAE;YACtE,EAAE,SAAS,EAAE,MAAM,EAAE;SACtB;QACD,eAAe,EAAE,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,UAAU,EAAE,YAAY,EAAE,SAAS,EAAE,EAAE;QAC5E,OAAO,EAAE,EAAE,QAAQ,EAAE,CAAC,EAAE;QACxB,MAAM,EAAE;YACN,MAAM,EAAE,WAAW;YACnB,YAAY,EAAE,UAAU;YACxB,QAAQ,EAAE,CAAC;YACX,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE;gBACzB,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,mBAAmB,CAAC,CAAC;gBAC/E,MAAM,IAAI,GAAG,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;gBAC7C,IAAI,IAAI,IAAI,IAAI,CAAC,IAAI,KAAK,mBAAmB,IAAI,IAAI,CAAC,OAAO,CAAC,SAAS,GAAG,IAAI,EAAE,CAAC;oBAC/E,MAAM,IAAI,KAAK,CAAC,6BAA6B,IAAI,CAAC,OAAO,CAAC,SAAS,QAAQ,CAAC,CAAC;gBAC/E,CAAC;YACH,CAAC;SACF;KACF;IAED;QACE,IAAI,EAAE,6BAA6B;QACnC,KAAK,EAAE,QAAQ;QACf,KAAK,EAAE,gFAAgF;QACvF,IAAI,EAAE,qCAAqC;QAC3C,QAAQ,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,GAAG,IAAI,mBAAmB;QAC9C,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,MAAM,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC;QACtD,eAAe,EAAE,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,iBAAiB,EAAE,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE;QAC7F,OAAO,EAAE,EAAE,QAAQ,EAAE,CAAC,EAAE;QACxB,MAAM,EAAE;YACN,MAAM,EAAE,WAAW;YACnB,YAAY,EAAE,UAAU;YACxB,QAAQ,EAAE,CAAC;SACZ;KACF;IAED;QACE,IAAI,EAAE,6BAA6B;QACnC,KAAK,EAAE,QAAQ;QACf,KAAK,EAAE,8CAA8C;QACrD,IAAI,EAAE,4BAA4B;QAClC,QAAQ,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,GAAG,IAAI,kBAAkB;QAC7C,MAAM,EAAE;YACN,EAAE,SAAS,EAAE,UAAU,EAAE,WAAW,EAAE,8BAA8B,EAAE;YACtE,EAAE,SAAS,EAAE,MAAM,EAAE;SACtB;QACD,eAAe,EAAE,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,oBAAoB,EAAE,EAAE;QAC7D,OAAO,EAAE,EAAE,QAAQ,EAAE,CAAC,EAAE,aAAa,EAAE,MAAM,EAAE;QAC/C,MAAM,EAAE;YACN,MAAM,EAAE,WAAW;YACnB,YAAY,EAAE,UAAU;YACxB,SAAS,EAAE,EAAE,YAAY,EAAE,SAAS,EAAE;SACvC;KACF;CACF,CAAC"}

package/package.json

@@ -0,0 +1,55 @@
+{
+  "name": "@mlola/browser-evals",
+  "version": "0.1.0",
+  "description": "Deterministic eval suites, fixture site and reporters.",
+  "type": "module",
+  "license": "MIT",
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "development": "./src/index.ts",
+      "default": "./dist/index.js"
+    }
+  },
+  "files": [
+    "dist",
+    "LICENSE",
+    "README.md"
+  ],
+  "sideEffects": false,
+  "engines": {
+    "node": ">=22.18"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/mlolahq/browser-runtime.git"
+  },
+  "homepage": "https://mlola.com/browser-runtime",
+  "bugs": {
+    "url": "https://github.com/mlolahq/browser-runtime.git/issues"
+  },
+  "keywords": [
+    "browser-automation",
+    "browser-agent",
+    "runtime",
+    "playwright",
+    "agent",
+    "jev",
+    "llm",
+    "automation"
+  ],
+  "scripts": {
+    "build": "tsc -b"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "dependencies": {
+    "@mlola/browser-protocol": "0.1.0",
+    "@mlola/browser-core": "0.1.0",
+    "@mlola/browser-runtime": "0.1.0",
+    "@mlola/decision-jev": "0.1.0"
+  }
+}