vieval 0.0.10 → 0.0.12

package/README.md

@@ -31,6 +31,7 @@ import { defineConfig, loadEnv, requiredEnvFrom } from 'vieval'
 import { chatModelFrom, ChatModels } from 'vieval/plugins/chat-models'
 
 export default defineConfig({
+  env: loadEnv('test', cwd(), ''),
   plugins: [
     ChatModels({
       models: [
@@ -46,23 +47,22 @@ export default defineConfig({
       ],
     }),
   ],
-  env: loadEnv('test', cwd(), ''),
   projects: [
     {
+      evalMatrix: {
+        extend: {
+          rubric: ['default'],
+        },
+      },
+      include: ['evals/*.eval.ts'],
       name: 'default',
       root: '.',
-      include: ['evals/*.eval.ts'],
       runMatrix: {
         extend: {
           model: ['agent-mini'],
           scenario: ['baseline'],
         },
       },
-      evalMatrix: {
-        extend: {
-          rubric: ['default'],
-        },
-      },
     },
   ],
 })
@@ -129,8 +129,8 @@ Use builder style when loading a batch of inputs:
 import { describeTask, expect } from 'vieval'
 
 const arithmeticCases = [
-  { name: 'addition-small', input: { a: 1, b: 2, expected: 3 } },
-  { name: 'addition-large', input: { a: 20, b: 22, expected: 42 } },
+  { input: { a: 1, b: 2, expected: 3 }, name: 'addition-small' },
+  { input: { a: 20, b: 22, expected: 42 }, name: 'addition-large' },
 ]
 
 describeTask('arithmetic-quality', ({ casesFromInputs }) => {
@@ -191,6 +191,7 @@ import { defineConfig, loadEnv, requiredEnvFrom } from 'vieval'
 import { chatModelFrom, ChatModels } from 'vieval/plugins/chat-models'
 
 export default defineConfig({
+  env: loadEnv('test', cwd(), ''),
   plugins: [
     ChatModels({
       models: [
@@ -224,12 +225,17 @@ export default defineConfig({
       ],
     }),
   ],
-  env: loadEnv('test', cwd(), ''),
   projects: [
     {
+      evalMatrix: {
+        extend: {
+          rubric: ['strict', 'lenient'],
+          rubricModel: ['judge-mini', 'judge-large'],
+        },
+      },
+      include: ['evals/*.eval.ts'],
       name: 'chat-evals',
       root: '.',
-      include: ['evals/*.eval.ts'],
       runMatrix: {
         extend: {
           model: ['agent-mini', 'agent-large'],
@@ -237,12 +243,6 @@ export default defineConfig({
           scenario: ['baseline', 'stress'],
         },
       },
-      evalMatrix: {
-        extend: {
-          rubric: ['strict', 'lenient'],
-          rubricModel: ['judge-mini', 'judge-large'],
-        },
-      },
     },
   ],
 })
@@ -258,19 +258,6 @@ import { defineConfig } from 'vieval'
 export default defineConfig({
   projects: [
     {
-      name: 'motion-evals',
-      root: '.',
-      include: ['evals/*.eval.ts'],
-      inferenceExecutors: [{ id: 'motion-engine' }],
-      models: [
-        {
-          id: 'motion-engine:v2',
-          aliases: ['motion-default'],
-          inferenceExecutor: 'motion-engine',
-          inferenceExecutorId: 'motion-engine',
-          model: 'v2',
-        },
-      ],
       async executor(task, context) {
         const model = context.models.find(model =>
           model.id === 'motion-default'
@@ -285,13 +272,26 @@ export default defineConfig({
         const success = model.model === 'v2' && task.matrix.run.scenario === 'baseline'
 
         return {
-          id: task.id,
           entryId: task.entry.id,
+          id: task.id,
           inferenceExecutorId: task.inferenceExecutor.id,
           matrix: task.matrix,
           scores: [{ kind: 'exact', score: success ? 1 : 0 }],
         }
       },
+      include: ['evals/*.eval.ts'],
+      inferenceExecutors: [{ id: 'motion-engine' }],
+      models: [
+        {
+          aliases: ['motion-default'],
+          id: 'motion-engine:v2',
+          inferenceExecutor: 'motion-engine',
+          inferenceExecutorId: 'motion-engine',
+          model: 'v2',
+        },
+      ],
+      name: 'motion-evals',
+      root: '.',
     },
   ],
 })

package/dist/bin/vieval.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { n as runTopLevelCli } from "../cli-DTDgaqeI.mjs";
+import { n as runTopLevelCli } from "../cli-uzS81IPd.mjs";
 import process from "node:process";
 import { errorMessageFrom } from "@moeru/std";
 //#region src/bin/vieval.ts

package/dist/bin/vieval.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"vieval.mjs","names":[],"sources":["../../src/bin/vieval.ts"],"sourcesContent":["#!/usr/bin/env node\n\nimport process from 'node:process'\n\nimport { errorMessageFrom } from '@moeru/std'\n\nimport { runTopLevelCli } from '../cli/index'\n\n/**\n * Bootstraps the published `vieval` executable.\n *\n * Call stack:\n *\n * package manager shim / direct node execution\n *   -> {@link runTopLevelCli} (`../cli`)\n *     -> subcommand orchestration modules\n *\n * Use when:\n * - the installed `vieval` binary starts from the command line\n * - process-bound startup must stay outside import-safe CLI modules\n *\n * Expects:\n * - `process.argv` contains the raw CLI arguments after the node executable path\n *\n * Returns:\n * - resolves after the selected subcommand finishes and updates `process.exitCode`\n */\nrunTopLevelCli(process.argv.slice(2)).catch((error) => {\n  const errorMessage = errorMessageFrom(error) ?? 'Unknown CLI failure.'\n  process.stderr.write(`[vieval] ${errorMessage}\\n`)\n  process.exitCode = 1\n})\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AA2BA,eAAe,QAAQ,KAAK,MAAM,EAAE,CAAC,CAAC,OAAO,UAAU;CACrD,MAAM,eAAe,iBAAiB,MAAM,IAAI;AAChD,SAAQ,OAAO,MAAM,YAAY,aAAa,IAAI;AAClD,SAAQ,WAAW;EACnB"}
+{"version":3,"file":"vieval.mjs","names":[],"sources":["../../src/bin/vieval.ts"],"sourcesContent":["#!/usr/bin/env node\n\nimport process from 'node:process'\n\nimport { errorMessageFrom } from '@moeru/std'\n\nimport { runTopLevelCli } from '../cli/index'\n\n/**\n * Bootstraps the published `vieval` executable.\n *\n * Call stack:\n *\n * package manager shim / direct node execution\n *   -> {@link runTopLevelCli} (`../cli`)\n *     -> subcommand orchestration modules\n *\n * Use when:\n * - the installed `vieval` binary starts from the command line\n * - process-bound startup must stay outside import-safe CLI modules\n *\n * Expects:\n * - `process.argv` contains the raw CLI arguments after the node executable path\n *\n * Returns:\n * - resolves after the selected subcommand finishes and updates `process.exitCode`\n */\nrunTopLevelCli(process.argv.slice(2)).catch((error) => {\n  const errorMessage = errorMessageFrom(error) ?? 'Unknown CLI failure.'\n  process.stderr.write(`[vieval] ${errorMessage}\\n`)\n  process.exitCode = 1\n})\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AA2BA,eAAe,QAAQ,KAAK,MAAM,CAAC,CAAC,CAAC,CAAC,OAAO,UAAU;CACrD,MAAM,eAAe,iBAAiB,KAAK,KAAK;CAChD,QAAQ,OAAO,MAAM,YAAY,aAAa,GAAG;CACjD,QAAQ,WAAW;AACrB,CAAC"}

package/dist/cli/index.d.mts

@@ -1,7 +1,7 @@
 //#region src/cli/index.d.ts
 type Command = 'compare' | 'report' | 'run';
 interface ParsedTopLevelCliArguments {
-  command: Command | 'help';
+  command: 'help' | Command;
   commandArgv: string[];
 }
 /**

package/dist/cli/index.mjs

@@ -1,2 +1,2 @@
-import { n as runTopLevelCli, t as parseTopLevelCliArguments } from "../cli-DTDgaqeI.mjs";
+import { n as runTopLevelCli, t as parseTopLevelCliArguments } from "../cli-uzS81IPd.mjs";
 export { parseTopLevelCliArguments, runTopLevelCli };