Fix compile error

Author: gspencergoog

specification/0.9/eval/src/basic_schema_matcher.ts

@@ -19,11 +19,19 @@ import { SchemaMatcher, ValidationResult } from "./schema_matcher";
 export class BasicSchemaMatcher extends SchemaMatcher {
   constructor(
     public propertyPath: string,
-    public propertyValue?: any,
+    public propertyValue?: any
   ) {
     super();
   }
 
+  get description(): string {
+    return `Basic match for property '${this.propertyPath}'${
+      this.propertyValue !== undefined
+        ? ` with value '${JSON.stringify(this.propertyValue)}'`
+        : ""
+    }`;
+  }
+
   validate(schema: any): ValidationResult {
     if (!schema) {
       const result: ValidationResult = {
@@ -54,7 +62,7 @@ export class BasicSchemaMatcher extends SchemaMatcher {
         const error = `Property '${
           this.propertyPath
         }' has value '${JSON.stringify(
-          actualValue,
+          actualValue
         )}', but expected '${JSON.stringify(this.propertyValue)}'.`;
         return { success: false, error };
       }
@@ -69,6 +77,10 @@ export class AnySchemaMatcher extends SchemaMatcher {
     super();
   }
 
+  get description(): string {
+    return `Any of: [${this.matchers.map((m) => m.description).join(", ")}]`;
+  }
+
   validate(schema: any): ValidationResult {
     const errors: string[] = [];
     for (const matcher of this.matchers) {
@@ -93,6 +105,10 @@ export class AllSchemaMatcher extends SchemaMatcher {
     super();
   }
 
+  get description(): string {
+    return `All of: [${this.matchers.map((m) => m.description).join(", ")}]`;
+  }
+
   validate(schema: any): ValidationResult {
     for (const matcher of this.matchers) {
       const result = matcher.validate(schema);

specification/0.9/eval/src/index.ts

@@ -110,7 +110,8 @@ function generateSummary(
 
     let totalModelFailedRuns = 0;
 
-    for (const promptName in promptsInModel) {
+    const sortedPromptNames = Object.keys(promptsInModel).sort();
+    for (const promptName of sortedPromptNames) {
       const runs = promptsInModel[promptName];
       const totalRuns = runs.length;
       const failedRuns = runs.filter(
@@ -132,7 +133,15 @@ function generateSummary(
     }
 
     const totalRunsForModel = resultsByModel[modelName].length;
-    summary += `\n\n**Total failed runs:** ${totalModelFailedRuns} / ${totalRunsForModel}`;
+    const successPercentage =
+      totalRunsForModel === 0
+        ? "0.0"
+        : (
+            ((totalRunsForModel - totalModelFailedRuns) / totalRunsForModel) *
+            100.0
+          ).toFixed(1);
+
+    summary += `\n\n**Total failed runs:** ${totalModelFailedRuns} / ${totalRunsForModel} (${successPercentage}% success)`;
   }
 
   summary += "\n\n---\n\n## Overall Summary\n";
@@ -150,7 +159,13 @@ function generateSummary(
   ].join(", ");
 
   summary += `\n- **Total tool failures:** ${totalToolErrorRuns} / ${totalRuns}`;
-  summary += `\n- **Number of runs with any failure (tool error or validation):** ${totalRunsWithAnyFailure} / ${totalRuns}`;
+  const successPercentage =
+    totalRuns === 0
+      ? "0.0"
+      : (((totalRuns - totalRunsWithAnyFailure) / totalRuns) * 100.0).toFixed(
+          1
+        );
+  summary += `\n- **Number of runs with any failure (tool error or validation):** ${totalRunsWithAnyFailure} / ${totalRuns} (${successPercentage}% success)`;
   const latencies = results.map((r) => r.latency).sort((a, b) => a - b);
   const totalLatency = latencies.reduce((acc, l) => acc + l, 0);
   const meanLatency = (totalLatency / totalRuns).toFixed(0);

specification/0.9/eval/src/prompts.ts

@@ -723,7 +723,6 @@ Each activity in the inner lists should be a 'Row' containing a 'CheckBox' (to m
     promptText: `Create a job application form. It should have 'TextField's for "Name", "Email", "Phone", "Resume URL". A 'MultipleChoice' for "Years of Experience" (options: "0-1", "2-5", "5+"). A 'Button' "Submit Application". Create these as static components, not data bound.`,
     matchers: [
       new MessageTypeMatcher("updateComponents"),
-      new SurfaceUpdateSchemaMatcher("Text", "text", "Job Application"),
       new SurfaceUpdateSchemaMatcher("TextField", "label", "Name"),
       new SurfaceUpdateSchemaMatcher("TextField", "label", "Email"),
       new SurfaceUpdateSchemaMatcher("TextField", "label", "Resume URL"),
@@ -734,8 +733,8 @@ Each activity in the inner lists should be a 'Row' containing a 'CheckBox' (to m
     name: "courseSyllabus",
     description: "A course syllabus outline.",
     promptText: `First, generate a 'createSurface' message with surfaceId 'main'. Then, generate a 'updateComponents' message for a course syllabus. 'Text' (h1) "Introduction to Computer Science". 'List' of modules.
-    - Module 1: 'Card' with 'Text' "Algorithms" and 'List' ("Sorting", "Searching").
-    - Module 2: 'Card' with 'Text' "Data Structures" and 'List' ("Arrays", "Linked Lists").`,
+- For module 1, a 'Card' with 'Text' "Algorithms" and 'List' ("Sorting", "Searching").
+- For module 2, a 'Card' with 'Text' "Data Structures" and 'List' ("Arrays", "Linked Lists").`,
     matchers: [
       new MessageTypeMatcher("updateComponents"),
       new SurfaceUpdateSchemaMatcher(