@claritylabs/cl-sdk 0.10.2 → 0.10.3

package/dist/index.mjs

@@ -540,6 +540,7 @@ var CoverageValueTypeSchema = z4.enum([
 var CoverageSchema = z4.object({
   name: z4.string(),
   limit: z4.string(),
+  limitType: LimitTypeSchema.optional(),
   limitValueType: CoverageValueTypeSchema.optional(),
   deductible: z4.string().optional(),
   deductibleValueType: CoverageValueTypeSchema.optional(),
@@ -1999,6 +2000,7 @@ function mergeCoverageLimits(existing, incoming) {
   const incomingCoverages = Array.isArray(incoming.coverages) ? incoming.coverages : [];
   const coverageKey = (coverage) => [
     String(coverage.name ?? "").toLowerCase(),
+    String(coverage.limitType ?? "").toLowerCase(),
     String(coverage.limit ?? "").toLowerCase(),
     String(coverage.deductible ?? "").toLowerCase(),
     String(coverage.formNumber ?? "").toLowerCase()
@@ -2964,6 +2966,7 @@ Available extractors:
 - supplementary
 
 Rules:
+- Identify the broad section or form context first, then assign focused extractors within that context.
 - Use specific extractors for declarations, schedules, endorsements, exclusions, conditions, premium pages, and loss runs.
 - Use "sections" for pages that contain substantive policy text or mixed content that should still be preserved as raw sections.
 - Avoid assigning broad ranges mentally; decide page by page.
@@ -2974,6 +2977,8 @@ Rules:
 - Headings like "Limits of Insurance", "Deductible", "Coinsurance", "Loss Conditions", or "Definitions" inside a policy form usually indicate form language, not declarations or schedules.
 - Continuation pages near the end of a form should stay mapped to "sections" plus "conditions"/"exclusions" when applicable, even if they mention limits or deductibles.
 - When a form inventory entry identifies a page range as a specific form type (e.g., endorsement, coverage, application), use that classification to guide your extractor choice. Do not assign "coverage_limits" to pages the inventory identifies as endorsement or condition/exclusion forms unless the page contains actual schedule values.
+- Do not tag a page with "exclusions" or "conditions" if it only contains a table of contents, page-number reference, running header/footer, or a heading that points to another page without substantive wording.
+- If a page appears to be part of a larger exclusion, conditions, or endorsement section within the same form, keep the assignment consistent across nearby pages in that section rather than isolating a single page fragment.
 - Return every page in the supplied chunk exactly once.
 
 Return JSON:
@@ -3158,7 +3163,7 @@ Extract only insured-specific declaration, schedule, or endorsement entries that
 
 Focus on:
 - Every coverage listed on the declarations page or coverage schedule
-- Per-occurrence, aggregate, and sub-limits for each coverage
+- Per-occurrence, individual/occurrence, aggregate, and sub-limits for each coverage
 - Deductible or self-insured retention for each coverage
 - Coverage form type: occurrence-based, claims-made, or accident
 - Retroactive date for claims-made policies
@@ -3170,6 +3175,7 @@ For EACH coverage, also extract:
 - pageNumber: the original page number where the coverage row/value appears
 - sectionRef: the declarations/schedule/endorsement section heading where it appears
 - originalContent: the verbatim row or short source snippet used for this coverage
+- limitType: when applicable, classify the limit as per_occurrence, per_claim, aggregate, per_person, per_accident, statutory, blanket, or scheduled
 - limitValueType: classify the limit as numeric, included, not_included, as_stated, waiting_period, referential, or other
 - deductibleValueType: classify the deductible/value term similarly when deductible is present
 
@@ -3181,6 +3187,7 @@ Critical rules:
 - If a waiting period or hour deductible is shown as part of a specific declarations/schedule row, it may be captured in deductible. Otherwise omit it.
 - Use limitValueType or deductibleValueType to preserve non-numeric terms precisely instead of forcing them into numeric semantics.
 - Preserve one row per real coverage entry. Do not merge adjacent schedule rows into malformed names.
+- Keep individual/per-occurrence limits separate from aggregate limits even when they have the same coverage name, limit amount, deductible, and form number. Use limitType to distinguish them.
 
 Return JSON only.`;
 }
@@ -4058,7 +4065,7 @@ function createExtractor(config) {
     ranges.push({ startPage: start, endPage: previous });
     return ranges;
   }
-  function buildPlanFromPageAssignments(pageAssignments, pageCount) {
+  function buildPlanFromPageAssignments(pageAssignments, pageCount, formInventory) {
     const extractorPages = /* @__PURE__ */ new Map();
     for (const assignment of pageAssignments) {
       const extractors = assignment.extractorNames.length > 0 ? assignment.extractorNames : ["sections"];
@@ -4075,8 +4082,30 @@ function createExtractor(config) {
         extractorPages.set("sections", [...extractorPages.get("sections") ?? [], page]);
       }
     }
+    const contextualExtractors = /* @__PURE__ */ new Set(["conditions", "exclusions", "endorsements"]);
+    const contextualForms = (formInventory?.forms ?? []).filter(
+      (form) => form.pageStart != null && (form.pageEnd ?? form.pageStart) != null
+    );
+    const expandPagesToFormRanges = (extractorName, pages) => {
+      if (!contextualExtractors.has(extractorName)) return pages;
+      const expanded = new Set(pages);
+      for (const page of pages) {
+        for (const form of contextualForms) {
+          const pageStart = form.pageStart;
+          const pageEnd = form.pageEnd ?? form.pageStart;
+          const formType = form.formType;
+          const supportsContextualExpansion = extractorName === "endorsements" ? formType === "endorsement" : formType === "coverage" || formType === "endorsement";
+          if (!supportsContextualExpansion) continue;
+          if (page < pageStart || page > pageEnd) continue;
+          for (let current = pageStart; current <= pageEnd; current += 1) {
+            expanded.add(current);
+          }
+        }
+      }
+      return [...expanded].sort((a, b) => a - b);
+    };
     const tasks = [...extractorPages.entries()].flatMap(
-      ([extractorName, pages]) => groupContiguousPages(pages).map(({ startPage, endPage }) => ({
+      ([extractorName, pages]) => groupContiguousPages(expandPagesToFormRanges(extractorName, pages)).map(({ startPage, endPage }) => ({
         extractorName,
         startPage,
         endPage,
@@ -4244,7 +4273,7 @@ function createExtractor(config) {
       onProgress?.("Resuming from checkpoint (plan complete)...");
     } else {
       onProgress?.(`Building extraction plan from page map for ${primaryType} ${documentType}...`);
-      plan = buildPlanFromPageAssignments(pageAssignments, pageCount);
+      plan = buildPlanFromPageAssignments(pageAssignments, pageCount, formInventory);
       await pipelineCtx.save("plan", {
         id,
         pageCount,