feat: sync pane workspace overhaul from private repo

- ship the expanded left-nav and pane header workspace updates
- add dimension-driven browse flow, map view upgrades, and delete-node tooling
- expand eval coverage and refresh the public UI docs for the new layout

Generated with Codex

tests/evals/dataset.json

@@ -1,11 +1,13 @@
 {
   "id": "golden-ra-h-core-v2",
   "name": "Golden RA-H Core v2",
-  "description": "Slim 5-scenario eval set for core RA-H usage: focused graph writes, skill-guided writes, indexed node search, chunk-grounded insight creation, and hub traversal.",
+  "description": "Core RA-H eval set covering database, tools, skills, search, and ingestion behavior, including graph writes, indexed node search, retrieval regressions, chunk-grounded insight creation, and hub traversal.",
   "inherits": "golden-v1",
   "version": 2,
   "focus": [
     "tool-call correctness",
+    "category-based coverage",
+    "search regressions",
     "skill usage only when appropriate",
     "core retrieval/write behavior",
     "latency/tokens/cost guards"

tests/evals/runner.ts

@@ -3,7 +3,7 @@ import path from 'path';
 import Database from 'better-sqlite3';
 import { randomUUID } from 'crypto';
 import { scenarios } from './scenarios';
-import { Scenario } from './types';
+import { EvalCategory, Scenario } from './types';
 
 type EvalChatRow = {
   trace_id: string;
@@ -38,7 +38,7 @@ type EvalResult = {
 
 const BASE_URL = process.env.RAH_EVALS_BASE_URL || 'http://localhost:3000';
 const DATASET_ENV = process.env.RAH_EVALS_DATASET_ID;
-const SUITE_ENV = (process.env.RAH_EVALS_SUITE || 'all').toLowerCase();
+const CATEGORY_ENV = (process.env.RAH_EVALS_CATEGORY || process.env.RAH_EVALS_SUITE || 'all').toLowerCase();
 const WAIT_TIMEOUT_MS = Number(process.env.RAH_EVALS_TIMEOUT_MS || 60000);
 const LOG_DB_PATH = path.join(process.cwd(), 'logs', 'evals.sqlite');
 const RAH_DB_PATH = process.env.SQLITE_DB_PATH || path.join(
@@ -54,7 +54,7 @@ function loadDatasetId() {
   return parsed.id || 'default';
 }
 
-type EvalSuite = 'all' | 'tools' | 'skills' | 'traversal' | 'internal' | 'external';
+type EvalCategoryFilter = 'all' | EvalCategory;
 
 type FocusedNodeContext = {
   id: number;
@@ -66,25 +66,25 @@ type FocusedNodeContext = {
   metadata: string | null;
 };
 
-function getDefaultScenarioSuites(scenario: Scenario): string[] {
-  const suites = ['internal'];
-  if (scenario.id.startsWith('skill-trigger-')) {
-    suites.push('skills');
-  } else {
-    suites.push('tools');
+function getDefaultScenarioCategories(scenario: Scenario): EvalCategory[] {
+  if (scenario.id.includes('search') || scenario.id.includes('quote')) {
+    return ['search'];
   }
-  if (scenario.id.includes('traverse')) {
-    suites.push('traversal');
+  if (scenario.id.includes('skill')) {
+    return ['skills'];
   }
-  return suites;
+  if (scenario.id.includes('extract') || scenario.id.includes('ingest')) {
+    return ['ingestion', 'tools'];
+  }
+  return ['database', 'tools'];
 }
 
-function shouldRunScenario(scenario: Scenario, suite: EvalSuite) {
-  if (suite === 'all') return true;
-  const suites = scenario.suites && scenario.suites.length > 0
-    ? scenario.suites
-    : getDefaultScenarioSuites(scenario);
-  return suites.includes(suite);
+function shouldRunScenario(scenario: Scenario, category: EvalCategoryFilter) {
+  if (category === 'all') return true;
+  const categories = scenario.categories && scenario.categories.length > 0
+    ? scenario.categories
+    : getDefaultScenarioCategories(scenario);
+  return categories.includes(category);
 }
 
 function resolveFocusedNodeId(query: Scenario['input']['focusedNodeQuery']): number | null {
@@ -424,14 +424,14 @@ async function runScenario(scenario: Scenario, datasetId: string): Promise<EvalR
 }
 
 async function runAll() {
-  const suite: EvalSuite = ['skills', 'tools', 'traversal', 'internal', 'external'].includes(SUITE_ENV)
-    ? (SUITE_ENV as EvalSuite)
+  const category: EvalCategoryFilter = ['database', 'tools', 'skills', 'search', 'ingestion'].includes(CATEGORY_ENV)
+    ? (CATEGORY_ENV as EvalCategoryFilter)
     : 'all';
   const datasetId = loadDatasetId();
   await ensureServerReady();
   await ensureEvalsEnabled();
-  const runnable = scenarios.filter(s => s.enabled !== false && shouldRunScenario(s, suite));
-  console.log(`Running ${runnable.length} scenarios (dataset: ${datasetId}, suite: ${suite})...\n`);
+  const runnable = scenarios.filter(s => s.enabled !== false && shouldRunScenario(s, category));
+  console.log(`Running ${runnable.length} scenarios (dataset: ${datasetId}, category: ${category})...\n`);
 
   const results: EvalResult[] = [];
   for (const scenario of runnable) {

tests/evals/scenarios/index.ts

@@ -1,6 +1,7 @@
 import { scenario as focusedGraphWrite } from './focused-graph-write';
 import { scenario as skillGuidedWrite } from './skill-guided-write';
 import { scenario as nodeIndexSearch } from './node-index-search';
+import { scenario as searchDocToLora } from './search-doc-to-lora';
 import { scenario as chunkQuoteInsight } from './chunk-quote-insight';
 import { scenario as hubTraversal } from './hub-traversal';
 
@@ -8,6 +9,7 @@ export const scenarios = [
   focusedGraphWrite,
   skillGuidedWrite,
   nodeIndexSearch,
+  searchDocToLora,
   chunkQuoteInsight,
   hubTraversal,
 ];

tests/evals/scenarios/search-doc-to-lora.ts

@@ -0,0 +1,21 @@
+import { Scenario } from '../types';
+
+export const scenario: Scenario = {
+  id: 'search-doc-to-lora',
+  name: 'Search Doc-to-LoRA',
+  description: 'Free-text graph search should find the Doc-to-LoRA node by title without speculative dimension constraints or unnecessary web search.',
+  categories: ['search'],
+  tools: ['queryNodes'],
+  input: {
+    message: 'Find the recent Doc-to-LoRA stuff in my graph. Just return the matching graph node or nodes.',
+  },
+  expect: {
+    toolsCalledSoft: ['queryNodes'],
+    toolsNotCalled: ['webSearch'],
+    responseContains: ['Doc-to-LoRA: Learning to Instantly Internalize Contexts'],
+    maxLatencyMs: 15000,
+    maxTotalTokens: 8000,
+    maxEstimatedCostUsd: 0.08,
+  },
+  notes: 'Regression for the failure where free-text search was over-constrained by guessed dimensions and missed the existing Doc-to-LoRA node.',
+};

tests/evals/types.ts

@@ -1,3 +1,5 @@
+export type EvalCategory = 'database' | 'tools' | 'skills' | 'search' | 'ingestion';
+
 export type ScenarioExpectations = {
   skillsRead?: string[];
   skillsReadSoft?: string[];
@@ -32,6 +34,7 @@ export type Scenario = {
   expect?: ScenarioExpectations;
   description?: string;
   tools?: string[];
+  categories?: EvalCategory[];
   suites?: string[];
   enabled?: boolean;
   notes?: string;