feat(scoring): export the deep-eval and category-score functions from the package root

README.md

@@ -97,6 +97,8 @@ Use the programmatic API when you want to integrate AXIS into an existing test r
 
 `run({ profile })` and `loadConfig(path, { profile })` select a named profile from the config, the same overlay `axis run --profile` applies. `loadConfig` returns the resolved `config` alongside the pre-merge `baseConfig`, for callers that need the suite layout as a whole rather than the active suite.
 
+For per-category scoring without the full `scoreRunResult` pipeline, import `runDeepEval`, `computeCategoryScore` and `aggregateDimension` from `@netlify/axis`. Select your judge adapter, registered with `registerAdapter`, through `runDeepEval`’s `options.judging`. It makes one judge call per category with interactions; setting that category’s `options.weights` value to `0` skips the call. Skipped categories use AXIS’s defaults. Use `computeCategoryScore` for a category score or `aggregateDimension` for a single audit dimension. The package root also exports `computeHeuristicSpeed`, `parseCategoryEvalResponse`, `parseJsonFromText`, and the types `DeepEvalOptions` and `CalibrationParams`. These exports add no scoring behavior changes; existing deep imports keep working.
+
 ## Roadmap
 
 Delivered: scenario runner, four-dimension scoring pipeline, baselines with regression detection, repeated runs with representative-run selection and noise-aware regression bands, MCP/skills wiring, custom adapter API, config profiles for running one repo's scenarios under several agent matrices, built-in adapters for Claude Code, Codex, and Gemini.

src/index.ts

@@ -22,6 +22,11 @@ export type { AcpAdapterSpec } from "./adapters/base/acp-adapter.js";
 export { scoreResults, scoreRunResult, buildScoredOutput } from "./scoring/index.js";
 export { ScoringError } from "./scoring/errors.js";
 export { buildSparseIndex } from "./scoring/sparse-index.js";
+export { runDeepEval, computeHeuristicSpeed, parseCategoryEvalResponse } from "./scoring/deep-eval.js";
+export type { DeepEvalOptions } from "./scoring/deep-eval.js";
+export { computeCategoryScore, aggregateDimension } from "./scoring/category-score.js";
+export type { CalibrationParams } from "./scoring/category-score.js";
+export { parseJsonFromText } from "./scoring/parse-json.js";
 export { getPromptTemplates, interpolate } from "./scoring/prompt-templates.js";
 export type { PromptTemplate, PromptVariable } from "./scoring/prompt-templates.js";
 export { categorizeInteraction } from "./transcript/categorize.js";

test/unit/scoring/root-exports.test.ts

@@ -0,0 +1,21 @@
+import { describe, expect, it } from "vitest";
+import * as axis from "../../../src/index.js";
+import { runDeepEval, computeHeuristicSpeed, parseCategoryEvalResponse } from "../../../src/scoring/deep-eval.js";
+import { computeCategoryScore, aggregateDimension } from "../../../src/scoring/category-score.js";
+import { parseJsonFromText } from "../../../src/scoring/parse-json.js";
+
+describe("package root exports", () => {
+  it("re-exports the scoring internals that programmatic consumers import", () => {
+    const expected = {
+      runDeepEval,
+      computeHeuristicSpeed,
+      parseCategoryEvalResponse,
+      computeCategoryScore,
+      aggregateDimension,
+      parseJsonFromText,
+    };
+    for (const [name, fn] of Object.entries(expected)) {
+      expect(axis[name as keyof typeof axis], name).toBe(fn);
+    }
+  });
+});