@stll/folio-core 0.37.4 → 0.38.0

package/dist/version-comparison.js

@@ -1,7 +1,7 @@
 import { folioAIBlockIdStability } from "./ai-edits/block-identity.js";
 import { FolioDocxReviewer } from "./ai-edits/headless.js";
 import { alignFolioContentBlocks, alignFolioContentStructure, createFolioContentAlignmentWorkSession, exceedsFolioContentLcsBudget } from "./compare/content-alignment.js";
-import { compareAlignedFolioContent, createContentComparisonWorkSession } from "./compare/content.js";
+import { FolioContentInlinePresentationProjectionError, compareAlignedFolioContent, createContentComparisonWorkSession } from "./compare/content.js";
 import { pairFolioDocumentStories } from "./document-stories.js";
 import { FOLIO_DOCUMENT_METADATA_PROPERTIES, FOLIO_DOCUMENT_PRIVACY_TRANSFORMS, PRIVATE_METADATA_PROPERTIES_BY_TRANSFORM, isFolioDocumentPrivacyTransform } from "./docx/metadataPrivacy.js";
 import { TaggedError, panic } from "better-result";
@@ -21,39 +21,35 @@ import { TaggedError, panic } from "better-result";
 * ## Alignment
 *
 * Compatible body and table/container segments are aligned by the neutral
-* comparison core. Within each compatible segment, blocks are paired in
-* three passes, each only considering blocks the previous pass left unpaired:
+* comparison core. Quadratic structural candidate alignment shares bounded
+* cell and token budgets across the comparison. Within each compatible
+* segment, blocks are paired in four confidence-ordered passes:
 *
 * 1. **Stable-id pairing.** Blocks whose ids are equal and whose snapshot
-*    provenance marks those ids stable are paired directly. A source
-*    `w14:paraId` is stable identity independent of text, so an equal-id pair
-*    with different text is a genuine edit (`modified`). Deterministically
-*    synthesized ids and `seq-NNNN` fallbacks are positional instead: the
-*    adapter preserves that provenance so an ordinal shift falls through to
-*    exact-text alignment rather than becoming a false identity match.
-* 2. **Exact-text pairing.** An order-preserving LCS over remaining blocks,
-*    matched by exact text equality. This is what recovers same-text blocks
-*    that pass 1 missed because a fallback id shifted with the ordinal. Its
-*    O(m·n) table is skipped ({@link exceedsLcsBudget}) once the unpaired
-*    counts on both sides would exceed a fixed cell budget, so a document
-*    with few/no stable ids can't force a quadratic-sized allocation; those
-*    blocks fall through to pass 3 instead.
-* 3. **Positional fallback.** Whatever a monotonicity filter leaves
-*    unpaired is split into the gaps between anchored pairs (pass 1 + 2,
-*    time-ordered); within each gap the shorter side is zipped positionally
-*    against the longer one (`modified`), and any excess on either side is
-*    reported as `added` / `deleted`.
-*
-* The combined anchor set from passes 1 and 2 is re-filtered to the longest
-* increasing subsequence by revised-side index before pass 3 runs, so a
-* pathological crossing match (content reordered across versions) can't
-* produce an out-of-order gap — the alignment always walks both documents
-* forward.
+*    provenance marks those ids stable become the highest-confidence
+*    monotone anchor candidates. A source `w14:paraId` is stable identity
+*    independent of text, so an equal-id pair with different text is a genuine
+*    edit (`modified`). Deterministically synthesized ids and `seq-NNNN`
+*    fallbacks are positional instead: the adapter preserves that provenance
+*    so an ordinal shift falls through to exact-text alignment rather than
+*    becoming a false identity match.
+* 2. **Exact-text pairing.** Within each stable-id gap, nonblank text that is
+*    unique on both sides becomes an exact anchor. An O(n log n) increasing-
+*    subsequence selection keeps those anchors ordered; repeated or blank
+*    text is deliberately not treated as identity evidence.
+* 3. **Residual-id continuity.** Within the stable-plus-exact gaps, a unique
+*    id whose provenance changed between positional and stable may pair. This
+*    preserves identity across serialize-and-reopen boundaries without
+*    letting weaker evidence cross an established correspondence.
+* 4. **Positional fallback.** Whatever remains is split into the gaps between
+*    the confidence-ordered anchors; within each gap the shorter side is
+*    zipped positionally against the longer one (`modified`), and any excess
+*    on either side is reported as `added` / `deleted`.
 *
 * ## Move detection
 *
 * Relocated content would otherwise report as an unrelated `deleted` +
-* `added` pair (both order-preserving passes drop crossing matches by
+* `added` pair (the monotone alignment passes drop crossing matches by
 * design). The neutral comparison core re-classifies eligible exact and
 * closely edited pairs as `movedFrom` / `movedTo` entries sharing a
 * `moveGroupId`. Candidate counts and similarity work share bounded budgets
@@ -90,8 +86,10 @@ const INLINE_FORMAT_PROPERTIES = {
 	color: true
 };
 const PARAGRAPH_FORMAT_PROPERTIES = {
+	indentation: true,
 	styleId: true,
 	listLevel: true,
+	listReference: true,
 	alignment: true,
 	spacing: true
 };
@@ -184,7 +182,10 @@ const compareStoryBlocks = ({ baseStory, revisedStory, baseBlocks, revisedBlocks
 		idStability: folioAIBlockIdStability,
 		maxChanges: Number.MAX_SAFE_INTEGER
 	});
-	if (compared.isErr()) return panic("A version comparison exceeded an unreachable internal result limit", { cause: compared.error });
+	if (compared.isErr()) {
+		if (compared.error instanceof FolioContentInlinePresentationProjectionError) return panic("Version comparison requires an alignable inline formatting projection", { cause: compared.error });
+		return panic("A version comparison exceeded an unreachable internal result limit", { cause: compared.error });
+	}
 	const changes = [];
 	const counts = createSummaryCounts();
 	const baseBlockById = /* @__PURE__ */ new Map();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stll/folio-core",
-  "version": "0.37.4",
+  "version": "0.38.0",
   "description": "Headless, framework-neutral core of folio: the OOXML (.docx) parser, document model, ProseMirror integration, and page-layout engine. No React.",
   "keywords": [
     "document-model",
@@ -121,7 +121,7 @@
     "perf": "bun scripts/profile-editor.ts"
   },
   "dependencies": {
-    "@stll/docx-core": "^0.19.4",
+    "@stll/docx-core": "^0.20.0",
     "@stll/docx-utils": "^0.1.0",
     "@stll/template-conditions": ">=0.4.0 <1.0.0",
     "better-result": "3.0.1",