@sanity/ailf-studio 2.0.0 → 2.0.2

package/dist/index.js

@@ -9,11 +9,11 @@ import { useClient } from "sanity";
 
 // src/lib/constants.ts
 var API_VERSION = "2026-03-11";
-var isProduction = process.env.NODE_ENV === "production";
-var ARTIFACT_API_BASE_URL = isProduction ? "https://ailf-api.sanity.build/v1" : "http://localhost:3000/v1";
 var ENV = globalThis.process?.env ?? {};
+var isProduction = process.env.NODE_ENV === "production";
+var ARTIFACT_API_BASE_URL = ENV.SANITY_STUDIO_AILF_API_BASE_URL ?? (isProduction ? "https://ailf-api.sanity.build/v1" : "http://localhost:3000/v1");
 var REFERENCE_DATASET = ENV.SANITY_STUDIO_AILF_REF_DATASET ?? "next";
-var REFERENCE_WORKSPACE = ENV.SANITY_STUDIO_AILF_REF_WORKSPACE ?? "editorial";
+var EDITORIAL_STUDIO_URL = ENV.SANITY_STUDIO_ORIGIN ?? "https://admin.sanity.io";
 var AILF_DATASET = ENV.SANITY_STUDIO_AILF_DATASET ?? "ailf-prod-private";
 
 // src/actions/ArchiveTaskAction.tsx
@@ -681,7 +681,7 @@ Click into any report for the full breakdown: per-area scores, diagnostics, and
   {
     "id": "scoring-model",
     "title": "Understanding Scores",
-    "body": "## The three dimensions\n\nEvery evaluation task is scored on three dimensions, each graded 0\u2013100:\n\n- **Task Completion (50% weight)** \u2014 Can the AI implement the requested feature?\n  Does the output actually do what was asked?\n- **Code Correctness (25% weight)** \u2014 Is the generated code idiomatic, correct,\n  and following best practices?\n- **Doc Coverage (25% weight)** \u2014 Did the documentation provide the information\n  needed to implement the feature?\n\n## How the overall score is calculated\n\nThe three dimensions combine into a single **AI Literacy Score** per task using\nnamed scoring profiles from `config/rubrics.yaml`:\n\n```\nGold (with docs):    Total = Task \xD7 0.50 + Code \xD7 0.25 + Docs \xD7 0.25\nBaseline (no docs):  Total = Task \xD7 0.60 + Code \xD7 0.40\n```\n\nThe gold profile includes all three dimensions. The baseline profile excludes\nDoc Coverage because it is undefined when no documentation is provided. This\nensures Doc Lift (ceiling \u2212 floor) is a clean structural measurement of\ndocumentation value.\n\nThe weighted composite produces a score from 0\u2013100. Scores are then averaged\nacross all tasks in a feature area to produce a **per-area score**, and across\nall areas to produce the **overall score**.\n\n## What the numbers mean\n\n| Score range  | Interpretation                                                    |\n| ------------ | ----------------------------------------------------------------- |\n| **80\u2013100**   | Docs are working well \u2014 AI agents produce correct implementations |\n| **70\u201379**    | Needs attention \u2014 there may be gaps in specific dimensions        |\n| **Below 70** | Weak \u2014 AI agents consistently struggle with this area             |\n\n## Ceiling decomposition (baseline mode)\n\nWhen running in baseline mode, each task is evaluated twice \u2014 with and without\ndocumentation. This produces:\n\n- **Floor score** \u2014 Score without docs (what the model knows from training data\n  alone)\n- **Ceiling score** \u2014 Score with gold-standard docs injected directly into the\n  prompt\n- **Doc Lift** \u2014 Ceiling minus floor. Positive means docs help; negative means\n  docs hurt.\n- **Doc Quality Gap** \u2014 100 minus ceiling. Room for documentation improvement.\n\n## Three-layer decomposition (full mode)\n\nFull mode adds a third measurement \u2014 what happens when AI agents find docs on\ntheir own:\n\n- **Floor** \u2014 No docs (parametric knowledge only)\n- **Ceiling** \u2014 Gold-standard docs injected (best the docs can do)\n- **Actual** \u2014 Agent-retrieved docs (real-world performance)\n- **Retrieval Gap** \u2014 Ceiling minus actual (quality lost to findability)\n- **Infrastructure Efficiency** \u2014 Actual \xF7 ceiling (what fraction of doc quality\n  reaches agents)\n\n## Cost tracking\n\nEach evaluation also tracks token costs:\n\n- **Provider cost** \u2014 Token usage for generating implementations\n- **Grader cost** \u2014 Token usage for the grading model's assessments\n- **Total cost** \u2014 Both combined, reported in the score summary",
+    "body": "## The three dimensions\n\nEvery evaluation task is scored on three dimensions, each graded 0\u2013100:\n\n- **Task Completion (50% weight)** \u2014 Can the AI implement the requested feature?\n  Does the output actually do what was asked?\n- **Code Correctness (25% weight)** \u2014 Is the generated code idiomatic, correct,\n  and following best practices?\n- **Doc Coverage (25% weight)** \u2014 Did the documentation provide the information\n  needed to implement the feature?\n\n## How the overall score is calculated\n\nThe three dimensions combine into a single **AI Literacy Score** per task using\nnamed scoring profiles from `packages/eval/config/rubrics.ts`:\n\n```\nGold (with docs):    Total = Task \xD7 0.50 + Code \xD7 0.25 + Docs \xD7 0.25\nBaseline (no docs):  Total = Task \xD7 0.60 + Code \xD7 0.40\n```\n\nThe gold profile includes all three dimensions. The baseline profile excludes\nDoc Coverage because it is undefined when no documentation is provided. This\nensures Doc Lift (ceiling \u2212 floor) is a clean structural measurement of\ndocumentation value.\n\nThe weighted composite produces a score from 0\u2013100. Scores are then averaged\nacross all tasks in a feature area to produce a **per-area score**, and across\nall areas to produce the **overall score**.\n\n## What the numbers mean\n\n| Score range  | Interpretation                                                    |\n| ------------ | ----------------------------------------------------------------- |\n| **80\u2013100**   | Docs are working well \u2014 AI agents produce correct implementations |\n| **70\u201379**    | Needs attention \u2014 there may be gaps in specific dimensions        |\n| **Below 70** | Weak \u2014 AI agents consistently struggle with this area             |\n\n## Ceiling decomposition (baseline mode)\n\nWhen running in baseline mode, each task is evaluated twice \u2014 with and without\ndocumentation. This produces:\n\n- **Floor score** \u2014 Score without docs (what the model knows from training data\n  alone)\n- **Ceiling score** \u2014 Score with gold-standard docs injected directly into the\n  prompt\n- **Doc Lift** \u2014 Ceiling minus floor. Positive means docs help; negative means\n  docs hurt.\n- **Doc Quality Gap** \u2014 100 minus ceiling. Room for documentation improvement.\n\n## Three-layer decomposition (full mode)\n\nFull mode adds a third measurement \u2014 what happens when AI agents find docs on\ntheir own:\n\n- **Floor** \u2014 No docs (parametric knowledge only)\n- **Ceiling** \u2014 Gold-standard docs injected (best the docs can do)\n- **Actual** \u2014 Agent-retrieved docs (real-world performance)\n- **Retrieval Gap** \u2014 Ceiling minus actual (quality lost to findability)\n- **Infrastructure Efficiency** \u2014 Actual \xF7 ceiling (what fraction of doc quality\n  reaches agents)\n\n## Cost tracking\n\nEach evaluation also tracks token costs:\n\n- **Provider cost** \u2014 Token usage for generating implementations\n- **Grader cost** \u2014 Token usage for the grading model's assessments\n- **Total cost** \u2014 Both combined, reported in the score summary",
     "source": "docs/help/scoring-model.md",
     "related": [
       "three-layer",
@@ -3135,9 +3135,8 @@ function ReleasePicker(props) {
 
 // src/schema/task.ts
 function articleStudioUrl(document2) {
-  const origin = typeof window !== "undefined" && window.location?.origin ? window.location.origin : "";
   const typePart = document2.type ? `;type=${document2.type}` : "";
-  return `${origin}/${REFERENCE_WORKSPACE}/intent/edit/id=${document2.id}${typePart}/`;
+  return `${EDITORIAL_STUDIO_URL}/intent/edit/id=${document2.id}${typePart}/`;
 }
 var ASSERTION_TYPES = [
   { title: "LLM Rubric", value: "llm-rubric" },
@@ -13569,7 +13568,9 @@ function DocBadge({
       children: isLinked ? /* @__PURE__ */ jsx69(
         "a",
         {
-          href: `/intent/edit/id=${doc.documentId}`,
+          href: `${EDITORIAL_STUDIO_URL}/intent/edit/id=${doc.documentId}/`,
+          target: "_blank",
+          rel: "noopener noreferrer",
           onMouseEnter: () => setHovered(true),
           onMouseLeave: () => setHovered(false),
           style: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sanity/ailf-studio",
-  "version": "2.0.0",
+  "version": "2.0.2",
   "description": "AI Literacy Framework — Sanity Studio dashboard plugin",
   "type": "module",
   "license": "MIT",