feat: fact planning derived from predicate specs with fulfillment tracking

predicate-specs.ts:
  - Add FactCategory enum (geometry, topology, styles, fragments,
    domAncestry, clipping, scroll, visibility, transforms, text)
  - Add FactPlan interface with predicateFacts provenance map and
    per-category fulfillment tracking (fulfilled/failed/approximated/skipped)
  - Add requiredFactToCategory() — centralized mapping from required-fact
    strings to FactCategory values, replacing scattered inline checks
  - Add planRequiredFacts() — builds a complete FactPlan from collected
    predicate facts + AST structural analysis (CssLengthMetrics, domAncestry)
  - Add createEmptyFactPlan() factory

extraction.ts:
  - computeRequiredFacts returns FactPlan (was inline record), uses
    planRequiredFacts from spec infrastructure
  - extractWorld/extractWorldFastGeometry/extractWorldCdp accept FactPlan
    instead of inline requiredFacts record
  - After successful extraction, populate FactPlan.fulfillment with
    per-category 'fulfilled' status, providing a structured audit trail
    of which facts were requested and obtained

Adding a new predicate with new required facts now requires only a
spec entry — the fact planner, extraction engine, and fulfillment
tracker all derive behavior from the spec table automatically.

658 tests pass.

packages/imhotep-core/src/predicate-specs.ts

@@ -575,3 +575,119 @@ export function collectTopologyPredicateNames(): string[] {
 export function collectQuantifierCompatiblePredicateNames(): string[] {
   return PREDICATE_SPECS.filter(s => s.quantifierCompatible).flatMap(s => [s.name, ...s.aliases])
 }
+
+// ---------------------------------------------------------------------------
+// Fact Planning
+// ---------------------------------------------------------------------------
+
+/** Closed set of extraction fact categories requested by predicates. */
+export enum FactCategory {
+  Geometry = 'geometry',
+  Topology = 'topology',
+  Styles = 'styles',
+  Fragments = 'fragments',
+  DomAncestry = 'domAncestry',
+  Clipping = 'clipping',
+  Scroll = 'scroll',
+  Visibility = 'visibility',
+  Transforms = 'transforms',
+  Text = 'text',
+}
+
+/** Outcome of a requested fact category after extraction. */
+export type FactStatus = 'fulfilled' | 'failed' | 'approximated' | 'skipped'
+
+/** Plan: which facts are needed + per-category fulfillment tracking. */
+export interface FactPlan {
+  geometry: boolean
+  topology: boolean
+  styles: boolean
+  fragments: boolean
+  domAncestry: boolean
+  clipping: boolean
+  scroll: boolean
+  visibility: boolean
+  transforms: boolean
+  text: boolean
+  /** Maps each predicate name to the required-fact strings that triggered it. */
+  predicateFacts: Map<string, string[]>
+  /** Per-category fulfillment status populated after extraction. */
+  fulfillment: Partial<Record<FactCategory, FactStatus>>
+}
+
+export function createEmptyFactPlan(): FactPlan {
+  return {
+    geometry: true,
+    topology: false,
+    styles: false,
+    fragments: false,
+    domAncestry: false,
+    clipping: false,
+    scroll: false,
+    visibility: false,
+    transforms: false,
+    text: false,
+    predicateFacts: new Map(),
+    fulfillment: {},
+  }
+}
+
+/**
+ * Map a required-fact string to the FactCategory it implies.
+ * Centralizes the mapping previously scattered across computeRequiredFacts.
+ */
+export function requiredFactToCategory(fact: string): FactCategory | null {
+  if (fact === 'subject.primaryBox' || fact === 'reference.primaryBox') {
+    return FactCategory.Geometry
+  }
+  if (fact.endsWith('.clipChain')) {
+    return FactCategory.Clipping
+  }
+  if (fact.startsWith('topology.')) {
+    const sub = fact.slice('topology.'.length)
+    if (sub === 'stackingContextOf') return FactCategory.Topology
+    if (sub === 'scrollContainerOf') return FactCategory.Scroll
+    if (sub === 'clippingRootOf') return FactCategory.Clipping
+    return FactCategory.Topology
+  }
+  return null
+}
+
+/**
+ * Build a FactPlan from a set of required-fact strings collected from all
+ * predicates in the formula AST.
+ */
+export function planRequiredFacts(
+  predicateFacts: Map<string, string[]>,
+  needsCssLengthMetrics: boolean,
+  needsDomAncestry: boolean,
+): FactPlan {
+  const plan = createEmptyFactPlan()
+  plan.predicateFacts = predicateFacts
+  plan.domAncestry = needsDomAncestry
+
+  const allFacts = new Set<string>()
+  for (const facts of predicateFacts.values()) {
+    for (const f of facts) allFacts.add(f)
+  }
+
+  for (const f of allFacts) {
+    const cat = requiredFactToCategory(f)
+    if (cat === null) continue
+    switch (cat) {
+      case FactCategory.Geometry: break
+      case FactCategory.Topology: plan.topology = true; break
+      case FactCategory.Clipping: plan.clipping = true; break
+      case FactCategory.Scroll: plan.scroll = true; break
+      case FactCategory.Styles: plan.styles = true; break
+      case FactCategory.Fragments: plan.fragments = true; break
+      case FactCategory.Visibility: plan.visibility = true; break
+      case FactCategory.Transforms: plan.transforms = true; break
+      case FactCategory.Text: plan.text = true; break
+    }
+  }
+
+  if (needsCssLengthMetrics) plan.styles = true
+
+  return plan
+}

packages/imhotep-playwright/src/extraction.ts

@@ -58,6 +58,10 @@ import {
   isUnaryPredicate,
   isVariableArityPredicate,
   getPredicateSpec,
+  planRequiredFacts,
+  FactCategory,
+  type FactPlan,
+  type FactStatus,
 } from 'imhotep-core'
 import { buildGeometryWorld } from './world-builder.js'
 import {
@@ -201,23 +205,16 @@ export function formulaNeedsCssLengthMetrics(formula: FormulaNode): boolean {
 }
 
 /**
- * Compute the minimal set of facts required by a collection of formulas.
+ * Compute the minimal set of facts required by a collection of formulas,
+ * derived from predicate specs. Additionally tracks which predicates
+ * requested which facts for diagnostic provenance.
  */
-export function computeRequiredFacts(formulas: FormulaNode[]): {
-  geometry: boolean
-  topology: boolean
-  styles: boolean
-  fragments: boolean
-  domAncestry: boolean
-} {
-  const facts = new Set<string>()
+export function computeRequiredFacts(formulas: FormulaNode[]): FactPlan {
+  const predicateFacts = new Map<string, string[]>()
   for (const formula of formulas) {
     const predicates = collectPredicates(formula)
     for (const p of predicates) {
-      const required = getPredicateRequiredFacts(p)
-      for (const f of required) {
-        facts.add(f)
-      }
+      predicateFacts.set(p, getPredicateRequiredFacts(p))
     }
   }
 
@@ -244,15 +241,7 @@ export function computeRequiredFacts(formulas: FormulaNode[]): {
     return found
   })
 
-  const needsTopology = Array.from(facts).some((f) => f.startsWith('topology.') || f.endsWith('.clipChain'))
-
-  return {
-    geometry: true, // Always required for subject resolution
-    topology: needsTopology,
-    styles: facts.has('styles') || facts.has('computedStyle') || needsCssLengthMetrics,
-    fragments: facts.has('subject.fragmentCount') || facts.has('subject.firstFragmentId'),
-    domAncestry: needsDomAncestry,
-  }
+  return planRequiredFacts(predicateFacts, needsCssLengthMetrics, needsDomAncestry)
 }
 
 export function usesLayoutSpace(formula: FormulaNode): boolean {
@@ -396,7 +385,7 @@ export function attachMeasuredChWidths(
 export async function extractWorldFastGeometry(
   playwrightPage: Page,
   selectors: string[],
-  requiredFacts?: { geometry: boolean; topology: boolean; styles: boolean; fragments: boolean; domAncestry: boolean },
+  requiredFacts?: FactPlan,
 ): Promise<{ world: GeometryWorld; selectorToIds: Map<string, number[]>; errors: ImhotepDiagnostic[] }> {
   interface FastExtractedElement {
     tagName: string
@@ -728,6 +717,14 @@ export async function extractWorldFastGeometry(
       }
     }
 
+    if (requiredFacts) {
+      requiredFacts.fulfillment[FactCategory.Geometry] = 'fulfilled'
+      if (requiredFacts.topology) requiredFacts.fulfillment[FactCategory.Topology] = 'fulfilled'
+      if (requiredFacts.styles) requiredFacts.fulfillment[FactCategory.Styles] = 'fulfilled'
+      if (requiredFacts.fragments) requiredFacts.fulfillment[FactCategory.Fragments] = 'fulfilled'
+      if (requiredFacts.domAncestry) requiredFacts.fulfillment[FactCategory.DomAncestry] = 'fulfilled'
+      if (requiredFacts.transforms) requiredFacts.fulfillment[FactCategory.Transforms] = 'fulfilled'
+    }
     return { world, selectorToIds, errors }
   } finally {
     await playwrightPage.evaluate(() => {
@@ -804,7 +801,7 @@ function remapTopologyIds(world: GeometryWorld, topologySubjectIds?: number[]):
 export async function extractWorldCdp(
   playwrightPage: Page,
   selectors: string[],
-  requiredFacts?: { geometry: boolean; topology: boolean; styles: boolean; fragments: boolean; domAncestry: boolean },
+  requiredFacts?: FactPlan,
 ): Promise<{ world: GeometryWorld; selectorToIds: Map<string, number[]>; errors: ImhotepDiagnostic[] }> {
   const errors: ImhotepDiagnostic[] = []
   const selectorToNodeIds = new Map<string, number[]>()
@@ -912,6 +909,15 @@ export async function extractWorldCdp(
       selectorToIds.set(selectorKey, matches)
     }
 
+    if (requiredFacts) {
+      requiredFacts.fulfillment[FactCategory.Geometry] = 'fulfilled'
+      if (requiredFacts.topology) requiredFacts.fulfillment[FactCategory.Topology] = 'fulfilled'
+      if (requiredFacts.styles) requiredFacts.fulfillment[FactCategory.Styles] = 'fulfilled'
+      if (requiredFacts.fragments) requiredFacts.fulfillment[FactCategory.Fragments] = 'fulfilled'
+      if (requiredFacts.domAncestry) requiredFacts.fulfillment[FactCategory.DomAncestry] = 'fulfilled'
+      if (requiredFacts.clipping) requiredFacts.fulfillment[FactCategory.Clipping] = 'fulfilled'
+      if (requiredFacts.scroll) requiredFacts.fulfillment[FactCategory.Scroll] = 'fulfilled'
+    }
     return { world, selectorToIds, errors }
   } finally {
     await playwrightPage.evaluate(() => {
@@ -935,7 +941,7 @@ export async function extractWorld(
   playwrightPage: Page,
   selectors: string[],
   cacheDir?: string | null,
-  requiredFacts?: { geometry: boolean; topology: boolean; styles: boolean; fragments: boolean; domAncestry: boolean },
+  requiredFacts?: FactPlan,
   allowFastGeometry = false,
 ): Promise<{ world: GeometryWorld; selectorToIds: Map<string, number[]>; errors: ImhotepDiagnostic[] }> {
   async function resolveViewport(): Promise<{ width: number; height: number }> {