@claritylabs/cl-sdk 1.1.4 → 1.3.0

package/dist/index.mjs

@@ -28462,6 +28462,63 @@ var require_dist4 = __commonJS({
   }
 });
 
+// src/types/enums.ts
+var POLICY_TYPES = [
+  "general_liability",
+  "commercial_property",
+  "commercial_auto",
+  "non_owned_auto",
+  "workers_comp",
+  "umbrella",
+  "excess_liability",
+  "professional_liability",
+  "cyber",
+  "epli",
+  "directors_officers",
+  "fiduciary_liability",
+  "crime_fidelity",
+  "inland_marine",
+  "builders_risk",
+  "environmental",
+  "ocean_marine",
+  "surety",
+  "product_liability",
+  "bop",
+  "management_liability_package",
+  "property",
+  "other"
+];
+
+// src/types/context-keys.ts
+var CONTEXT_KEY_MAP = [
+  { extractedField: "insuredName", category: "company_info", contextKey: "company_name", description: "Primary named insured" },
+  { extractedField: "insuredDba", category: "company_info", contextKey: "dba_name", description: "Doing-business-as name" },
+  { extractedField: "insuredAddress", category: "company_info", contextKey: "company_address", description: "Primary insured mailing address" },
+  { extractedField: "insuredEntityType", category: "company_info", contextKey: "entity_type", description: "Legal entity type" },
+  { extractedField: "insuredFein", category: "company_info", contextKey: "fein", description: "Federal Employer ID Number" },
+  { extractedField: "insuredSicCode", category: "company_info", contextKey: "sic_code", description: "SIC classification code" },
+  { extractedField: "insuredNaicsCode", category: "company_info", contextKey: "naics_code", description: "NAICS classification code" },
+  { extractedField: "classifications[].description", category: "operations", contextKey: "description_of_operations", description: "Description of business operations" },
+  { extractedField: "classifications[].basisAmount(payroll)", category: "operations", contextKey: "annual_payroll", description: "Annual payroll from classification schedule" },
+  { extractedField: "classifications[].basisAmount(revenue)", category: "operations", contextKey: "annual_revenue", description: "Annual revenue from classification schedule" },
+  { extractedField: "totalPremium", category: "financial", contextKey: "current_premium", description: "Total policy premium" },
+  { extractedField: "locations[].buildingValue", category: "financial", contextKey: "total_property_values", description: "Sum of building values" },
+  { extractedField: "locations[].contentsValue", category: "financial", contextKey: "total_contents_values", description: "Sum of contents values" },
+  { extractedField: "policyTypes", category: "coverage", contextKey: "coverage_types", description: "Lines of business covered" },
+  { extractedField: "coverages[].limit", category: "coverage", contextKey: "current_limits", description: "Current coverage limits" },
+  { extractedField: "coverages[].deductible", category: "coverage", contextKey: "current_deductibles", description: "Current deductibles" },
+  { extractedField: "experienceMod.factor", category: "loss_history", contextKey: "experience_mod", description: "Workers comp experience modification factor" },
+  { extractedField: "lossSummary.totalClaims", category: "loss_history", contextKey: "total_claims", description: "Total claim count from loss runs" },
+  { extractedField: "locations[]", category: "premises", contextKey: "premises_addresses", description: "All insured location addresses" },
+  { extractedField: "locations[].constructionType", category: "premises", contextKey: "construction_type", description: "Building construction type" },
+  { extractedField: "locations[].yearBuilt", category: "premises", contextKey: "year_built", description: "Year built for primary location" },
+  { extractedField: "locations[].sprinklered", category: "premises", contextKey: "sprinkler_system", description: "Sprinkler system presence" },
+  { extractedField: "vehicles[]", category: "vehicles", contextKey: "vehicle_schedule", description: "Complete vehicle schedule" },
+  { extractedField: "vehicles[].length", category: "vehicles", contextKey: "vehicle_count", description: "Number of insured vehicles" },
+  { extractedField: "classifications[](WC)", category: "employees", contextKey: "employee_count_by_class", description: "Employee count by WC classification" },
+  { extractedField: "classifications[].basisAmount(payroll,byState)", category: "employees", contextKey: "annual_payroll_by_state", description: "Annual payroll by state" }
+];
+
 // src/types/platform.ts
 var PLATFORM_CONFIGS = {
   email: {
@@ -28537,7 +28594,7 @@ Respond with JSON only. The JSON must follow this exact structure:
     "broker": "insurance broker name if identifiable, or null",
     "policyNumber": "policy or quote reference number",
     "documentType": "policy" or "quote",
-    "policyTypes": ["general_liability", "workers_comp", "commercial_auto", "non_owned_auto", "property", "umbrella", "professional_liability", "cyber", "epli", "directors_officers", "other"],
+    "policyTypes": ["general_liability", "commercial_property", "commercial_auto", "non_owned_auto", "workers_comp", "umbrella", "excess_liability", "professional_liability", "cyber", "epli", "directors_officers", "fiduciary_liability", "crime_fidelity", "inland_marine", "builders_risk", "environmental", "ocean_marine", "surety", "product_liability", "bop", "management_liability_package", "property", "other"],
     "policyYear": number,
     "effectiveDate": "MM/DD/YYYY",
     "expirationDate": "MM/DD/YYYY",
@@ -28623,26 +28680,42 @@ CLASSIFICATION SIGNALS:
 - QUOTE signals: "quote", "proposal", "indication" wording, subjectivities, "subject to" conditions, quote expiration date, "proposed premium", "terms and conditions may vary"
 
 If uncertain, lean toward "policy" for documents with declarations pages and binding language, "quote" for everything else.`;
-var METADATA_PROMPT = `You are an expert insurance document analyst. Extract ONLY the high-level metadata from this insurance document. Do NOT extract full section content \u2014 that will be done in a separate pass.
+var METADATA_PROMPT = `You are an expert insurance document analyst. Extract the high-level metadata AND structured declarations data from this insurance document. Do NOT extract full section content \u2014 that will be done in a separate pass.
 
 Respond with JSON only:
 
 {
   "metadata": {
     "carrier": "primary insurance company name",
+    "carrierLegalName": "legal entity name of insurer, or null",
+    "carrierNaicNumber": "NAIC company code, or null",
+    "carrierAmBestRating": "AM Best rating (e.g. 'A+ XV'), or null",
+    "carrierAdmittedStatus": "admitted" or "non_admitted" or "surplus_lines" or null,
     "security": "insurer or underwriter entity providing coverage, or null",
     "underwriter": "named individual underwriter, or null",
     "mga": "MGA or Program Administrator, or null",
-    "broker": "insurance broker, or null",
+    "broker": "insurance broker agency name, or null",
+    "brokerContactName": "individual producer name, or null",
+    "brokerLicenseNumber": "producer license number, or null",
     "policyNumber": "policy number",
+    "priorPolicyNumber": "previous policy number if renewal, or null",
     "documentType": "policy" or "quote",
-    "policyTypes": ["general_liability", ...],
+    "policyTypes": ["general_liability", "commercial_property", "commercial_auto", "non_owned_auto", "workers_comp", "umbrella", "excess_liability", "professional_liability", "cyber", "epli", "directors_officers", "fiduciary_liability", "crime_fidelity", "inland_marine", "builders_risk", "environmental", "ocean_marine", "surety", "product_liability", "bop", "management_liability_package", "property", "other"],
+    "coverageForm": "occurrence" or "claims_made" or "accident" or null,
     "policyYear": number,
     "effectiveDate": "MM/DD/YYYY",
     "expirationDate": "MM/DD/YYYY",
+    "effectiveTime": "e.g. 12:01 AM, or null",
+    "retroactiveDate": "MM/DD/YYYY for claims-made policies, or null",
     "isRenewal": boolean,
+    "isPackage": boolean,
+    "programName": "named program, or null",
     "premium": "$X,XXX",
-    "insuredName": "name of insured party",
+    "insuredName": "name of primary named insured",
+    "insuredDba": "doing-business-as name, or null",
+    "insuredAddress": { "street1": "", "city": "", "state": "", "zip": "" } or null,
+    "insuredEntityType": "corporation" or "llc" or "partnership" or "sole_proprietor" or "joint_venture" or "trust" or "nonprofit" or "municipality" or "other" or null,
+    "insuredFein": "FEIN, or null",
     "summary": "1-2 sentence summary"
   },
   "metadataSource": {
@@ -28651,34 +28724,94 @@ Respond with JSON only:
     "premiumPage": number or null,
     "effectiveDatePage": number or null
   },
+  "additionalNamedInsureds": [
+    { "name": "insured name", "relationship": "subsidiary, affiliate, etc., or null" }
+  ],
   "coverages": [
     { "name": "coverage name", "limit": "$X,XXX,XXX", "deductible": "$X,XXX or null", "pageNumber": number, "sectionRef": "section ref or null" }
   ],
+  "limits": {
+    "perOccurrence": "$X,XXX,XXX or null",
+    "generalAggregate": "$X,XXX,XXX or null",
+    "productsCompletedOpsAggregate": "or null",
+    "personalAdvertisingInjury": "or null",
+    "fireDamage": "or null",
+    "medicalExpense": "or null",
+    "combinedSingleLimit": "or null",
+    "bodilyInjuryPerPerson": "or null",
+    "bodilyInjuryPerAccident": "or null",
+    "propertyDamage": "or null",
+    "eachOccurrenceUmbrella": "or null",
+    "umbrellaAggregate": "or null",
+    "umbrellaRetention": "or null",
+    "statutory": boolean or null,
+    "employersLiability": { "eachAccident": "", "diseasePolicyLimit": "", "diseaseEachEmployee": "" } or null,
+    "defenseCostTreatment": "inside_limits" or "outside_limits" or "supplementary" or null
+  },
+  "deductibles": {
+    "perClaim": "or null",
+    "perOccurrence": "or null",
+    "selfInsuredRetention": "or null",
+    "waitingPeriod": "or null"
+  },
+  "locations": [
+    { "number": 1, "address": { "street1": "", "city": "", "state": "", "zip": "" }, "description": "or null", "buildingValue": "or null", "contentsValue": "or null" }
+  ],
+  "vehicles": [
+    { "number": 1, "year": 2024, "make": "", "model": "", "vin": "", "vehicleType": "or null" }
+  ],
+  "classifications": [
+    { "code": "12345", "description": "class description", "premiumBasis": "payroll or revenue or area", "basisAmount": "or null", "rate": "or null", "premium": "or null" }
+  ],
+  "formInventory": [
+    { "formNumber": "CG 00 01", "editionDate": "04 13", "title": "or null", "formType": "coverage or endorsement or declarations or application or notice or other" }
+  ],
+  "taxesAndFees": [
+    { "name": "fee name", "amount": "$X,XXX", "type": "tax or fee or surcharge or assessment or null" }
+  ],
   "totalPages": number,
   "tableOfContents": [
     { "title": "section title", "pageStart": number, "pageEnd": number }
   ]
-}`;
-var QUOTE_METADATA_PROMPT = `You are an expert insurance document analyst. Extract ONLY the high-level metadata from this insurance QUOTE or PROPOSAL document. Do NOT extract full section content \u2014 that will be done in a separate pass.
+}
+
+IMPORTANT:
+- policyTypes should include ALL coverage types found in the document
+- coverageForm is the primary trigger type: "occurrence" for occurrence-based, "claims_made" for claims-made, "accident" for auto/WC
+- isPackage is true if this is a Commercial Package Policy (CPP) with multiple coverage parts
+- Extract locations ONLY if a location/premises schedule is visible on the declarations
+- Extract vehicles ONLY if a vehicle schedule is visible
+- Extract classifications ONLY if a classification/rating schedule is visible
+- formInventory: list ALL form numbers found in any forms schedule or endorsement schedule
+- For limits, extract the standard limit fields that appear on the declarations page
+- For deductibles, extract from the declarations or deductible schedule`;
+var QUOTE_METADATA_PROMPT = `You are an expert insurance document analyst. Extract the high-level metadata AND structured data from this insurance QUOTE or PROPOSAL. Do NOT extract full section content \u2014 that will be done in a separate pass.
 
 Respond with JSON only:
 
 {
   "metadata": {
     "carrier": "primary insurance company name",
-    "security": "insurer or underwriter entity providing coverage, or null",
+    "carrierLegalName": "legal entity name, or null",
+    "carrierNaicNumber": "NAIC code, or null",
+    "carrierAdmittedStatus": "admitted or non_admitted or surplus_lines, or null",
+    "security": "insurer or underwriter entity, or null",
     "underwriter": "named individual underwriter, or null",
     "mga": "MGA or Program Administrator, or null",
     "broker": "insurance broker, or null",
+    "brokerContactName": "individual producer, or null",
     "quoteNumber": "quote or proposal reference number",
-    "policyTypes": ["general_liability", ...],
+    "policyTypes": ["general_liability", "commercial_property", "commercial_auto", "non_owned_auto", "workers_comp", "umbrella", "excess_liability", "professional_liability", "cyber", "epli", "directors_officers", "fiduciary_liability", "crime_fidelity", "inland_marine", "builders_risk", "environmental", "ocean_marine", "surety", "product_liability", "bop", "management_liability_package", "property", "other"],
+    "coverageForm": "occurrence or claims_made or accident, or null",
     "quoteYear": number,
     "proposedEffectiveDate": "MM/DD/YYYY or null",
     "proposedExpirationDate": "MM/DD/YYYY or null",
     "quoteExpirationDate": "MM/DD/YYYY \u2014 when this quote offer expires, or null",
+    "retroactiveDate": "MM/DD/YYYY for claims-made, or null",
     "isRenewal": boolean,
     "premium": "$X,XXX \u2014 total proposed premium",
     "insuredName": "name of insured party",
+    "insuredAddress": { "street1": "", "city": "", "state": "", "zip": "" } or null,
     "summary": "1-2 sentence summary of the quote"
   },
   "metadataSource": {
@@ -28688,16 +28821,31 @@ Respond with JSON only:
     "effectiveDatePage": number or null
   },
   "coverages": [
-    { "name": "coverage name", "proposedLimit": "$X,XXX,XXX", "proposedDeductible": "$X,XXX or null", "pageNumber": number, "sectionRef": "section ref or null" }
+    { "name": "coverage name", "proposedLimit": "$X,XXX,XXX", "proposedDeductible": "$X,XXX or null", "pageNumber": number, "sectionRef": "or null" }
   ],
+  "limits": {
+    "perOccurrence": "or null",
+    "generalAggregate": "or null",
+    "defenseCostTreatment": "inside_limits or outside_limits or supplementary, or null"
+  },
+  "deductibles": {
+    "perClaim": "or null",
+    "perOccurrence": "or null",
+    "selfInsuredRetention": "or null",
+    "waitingPeriod": "or null"
+  },
   "premiumBreakdown": [
     { "line": "coverage line name", "amount": "$X,XXX" }
   ],
   "subjectivities": [
-    { "description": "subjectivity description", "category": "pre_binding" or "post_binding" or "information" or null, "pageNumber": number or null }
+    { "description": "subjectivity description", "category": "pre_binding or post_binding or information, or null", "dueDate": "or null", "pageNumber": number or null }
   ],
   "underwritingConditions": [
-    { "description": "condition description", "pageNumber": number or null }
+    { "description": "condition description", "category": "or null", "pageNumber": number or null }
+  ],
+  "warrantyRequirements": ["warranty text"],
+  "taxesAndFees": [
+    { "name": "fee name", "amount": "$X,XXX", "type": "tax or fee or surcharge, or null" }
   ],
   "totalPages": number,
   "tableOfContents": [
@@ -28707,8 +28855,10 @@ Respond with JSON only:
 
 IMPORTANT:
 - quoteExpirationDate is when the quote offer itself expires (not the proposed policy period)
-- subjectivities are conditions that must be met before or after binding (look for "subject to", "subjectivities", "conditions precedent")
-- premiumBreakdown should list each coverage line's individual premium if available`;
+- subjectivities are conditions that must be met before or after binding
+- premiumBreakdown should list each coverage line's individual premium if available
+- warrantyRequirements: extract any warranty provisions required for coverage
+- For limits and deductibles, extract the proposed structure from the quote`;
 function buildSectionsPrompt(pageStart, pageEnd) {
   return `You are an expert insurance document analyst. Extract ALL sections, clauses, endorsements, and schedules found on pages ${pageStart} through ${pageEnd} of this document. Preserve the original language verbatim.
 
@@ -28729,6 +28879,35 @@ Respond with JSON only:
       ]
     }
   ],
+  "endorsements": [
+    {
+      "formNumber": "e.g. CG 21 47",
+      "editionDate": "e.g. 12 07, or null",
+      "title": "endorsement title",
+      "coverageType": "policyTypes value if coverage-specific, or null",
+      "pageStart": number,
+      "effectType": "broadening or restrictive or informational or null",
+      "additionalPremium": "$X,XXX or null",
+      "content": "full verbatim text of the endorsement"
+    }
+  ],
+  "exclusions": [
+    {
+      "title": "exclusion title or short description",
+      "formNumber": "form number if part of a named endorsement, or null",
+      "coverageType": "policyTypes value if coverage-specific, or null",
+      "pageNumber": number,
+      "content": "full verbatim exclusion text"
+    }
+  ],
+  "conditions": [
+    {
+      "title": "condition title",
+      "coverageType": "policyTypes value if coverage-specific, or null",
+      "pageNumber": number,
+      "content": "full verbatim condition text"
+    }
+  ],
   "regulatoryContext": { "content": "verbatim text", "pageNumber": number } or null,
   "complaintContact": { "content": "verbatim text", "pageNumber": number } or null,
   "costsAndFees": { "content": "verbatim text", "pageNumber": number } or null,
@@ -28743,6 +28922,19 @@ SECTION TYPE GUIDANCE:
 - "insuring_agreement" \u2014 the insuring agreement clause (only if standalone, not inside a policy_form)
 - Other types for standalone sections only
 
+ENDORSEMENT GUIDANCE:
+- List every endorsement found in the page range in the "endorsements" array
+- effectType: "broadening" adds or expands coverage; "restrictive" limits or excludes coverage; "informational" changes administrative terms only
+- additionalPremium: extract if a premium charge or credit is shown on the endorsement
+
+EXCLUSION GUIDANCE:
+- List named exclusions from exclusion schedules or endorsements in the "exclusions" array
+- Also capture exclusions embedded within insuring agreements or conditions as separate entries if clearly labeled
+- Preserve the full verbatim exclusion text
+
+CONDITION GUIDANCE:
+- List policy conditions (duties after loss, cooperation clause, cancellation, etc.) in the "conditions" array
+
 IMPORTANT: Only extract content from pages ${pageStart}-${pageEnd}. Preserve original language exactly.`;
 }
 var buildPolicySectionsPrompt = buildSectionsPrompt;
@@ -28766,11 +28958,19 @@ Respond with JSON only:
       ]
     }
   ],
+  "exclusions": [
+    {
+      "title": "exclusion title or short description",
+      "coverageType": "policyTypes value if coverage-specific, or null",
+      "pageNumber": number,
+      "content": "full verbatim exclusion text"
+    }
+  ],
   "subjectivities": [
-    { "description": "subjectivity text", "category": "pre_binding" or "post_binding" or "information" or null, "pageNumber": number or null }
+    { "description": "subjectivity text", "category": "pre_binding or post_binding or information, or null", "dueDate": "or null", "pageNumber": number or null }
   ],
   "underwritingConditions": [
-    { "description": "condition text", "pageNumber": number or null }
+    { "description": "condition text", "category": "or null", "pageNumber": number or null }
   ]
 }
 
@@ -28783,6 +28983,11 @@ SECTION TYPE GUIDANCE:
 - "exclusion" \u2014 excluded coverages, limitations
 - "other" \u2014 anything else
 
+EXCLUSION GUIDANCE:
+- List named exclusions from any exclusion schedule, endorsement, or coverage summary in the "exclusions" array
+- Preserve the full verbatim exclusion text
+- Set coverageType if the exclusion applies to a specific coverage line
+
 IMPORTANT: Only extract content from pages ${pageStart}-${pageEnd}. Preserve original language exactly.`;
 }
 function buildSupplementaryEnrichmentPrompt(fields) {
@@ -29779,8 +29984,134 @@ var AGENT_TOOLS = [
 
 // src/extraction/pipeline.ts
 import { generateText } from "ai";
-var SONNET_MODEL = "claude-sonnet-4-6";
-var HAIKU_MODEL = "claude-haiku-4-5-20251001";
+
+// src/extraction/pdf.ts
+import {
+  PDFDocument,
+  PDFTextField,
+  PDFCheckBox,
+  PDFDropdown,
+  PDFRadioGroup,
+  StandardFonts,
+  rgb
+} from "pdf-lib";
+async function extractPageRange(pdfBase64, startPage, endPage) {
+  const srcBytes = typeof Buffer !== "undefined" ? Buffer.from(pdfBase64, "base64") : Uint8Array.from(atob(pdfBase64), (c) => c.charCodeAt(0));
+  const srcDoc = await PDFDocument.load(srcBytes, { ignoreEncryption: true });
+  const totalPages = srcDoc.getPageCount();
+  const start = Math.max(startPage - 1, 0);
+  const end = Math.min(endPage, totalPages) - 1;
+  if (start === 0 && end >= totalPages - 1) {
+    return pdfBase64;
+  }
+  const newDoc = await PDFDocument.create();
+  const indices = Array.from({ length: end - start + 1 }, (_, i) => start + i);
+  const pages = await newDoc.copyPages(srcDoc, indices);
+  pages.forEach((page) => newDoc.addPage(page));
+  const bytes = await newDoc.save();
+  if (typeof Buffer !== "undefined") {
+    return Buffer.from(bytes).toString("base64");
+  }
+  let binary = "";
+  const uint8 = new Uint8Array(bytes);
+  for (let i = 0; i < uint8.length; i++) {
+    binary += String.fromCharCode(uint8[i]);
+  }
+  return btoa(binary);
+}
+async function getPdfPageCount(pdfBase64) {
+  const srcBytes = typeof Buffer !== "undefined" ? Buffer.from(pdfBase64, "base64") : Uint8Array.from(atob(pdfBase64), (c) => c.charCodeAt(0));
+  const doc = await PDFDocument.load(srcBytes, { ignoreEncryption: true });
+  return doc.getPageCount();
+}
+function getAcroFormFields(pdfDoc) {
+  const form = pdfDoc.getForm();
+  const fields = form.getFields();
+  if (fields.length === 0) return [];
+  return fields.map((field) => {
+    const name = field.getName();
+    if (field instanceof PDFTextField) {
+      return { name, type: "text" };
+    }
+    if (field instanceof PDFCheckBox) {
+      return { name, type: "checkbox" };
+    }
+    if (field instanceof PDFDropdown) {
+      return { name, type: "dropdown", options: field.getOptions() };
+    }
+    if (field instanceof PDFRadioGroup) {
+      return { name, type: "radio", options: field.getOptions() };
+    }
+    return { name, type: "text" };
+  });
+}
+async function fillAcroForm(pdfBytes, mappings) {
+  const pdfDoc = await PDFDocument.load(pdfBytes, { ignoreEncryption: true });
+  const form = pdfDoc.getForm();
+  for (const { acroFormName, value } of mappings) {
+    try {
+      const field = form.getField(acroFormName);
+      if (field instanceof PDFTextField) {
+        field.setText(value);
+      } else if (field instanceof PDFCheckBox) {
+        const lower = value.toLowerCase();
+        if (["yes", "true", "x", "checked", "on"].includes(lower)) {
+          field.check();
+        } else {
+          field.uncheck();
+        }
+      } else if (field instanceof PDFDropdown) {
+        try {
+          field.select(value);
+        } catch {
+        }
+      } else if (field instanceof PDFRadioGroup) {
+        try {
+          field.select(value);
+        } catch {
+        }
+      }
+    } catch {
+    }
+  }
+  form.flatten();
+  return await pdfDoc.save();
+}
+async function overlayTextOnPdf(pdfBytes, overlays) {
+  const pdfDoc = await PDFDocument.load(pdfBytes, { ignoreEncryption: true });
+  const font = await pdfDoc.embedFont(StandardFonts.Helvetica);
+  const pageCount = pdfDoc.getPageCount();
+  for (const overlay of overlays) {
+    if (overlay.page < 0 || overlay.page >= pageCount) continue;
+    const page = pdfDoc.getPage(overlay.page);
+    const { width, height } = page.getSize();
+    const fontSize = overlay.fontSize ?? 10;
+    const x = overlay.x / 100 * width;
+    const y = height - overlay.y / 100 * height - fontSize;
+    if (overlay.isCheckmark) {
+      page.drawText("X", {
+        x,
+        y,
+        size: fontSize,
+        font,
+        color: rgb(0, 0, 0)
+      });
+    } else {
+      page.drawText(overlay.text, {
+        x,
+        y,
+        size: fontSize,
+        font,
+        color: rgb(0, 0, 0)
+      });
+    }
+  }
+  return await pdfDoc.save();
+}
+
+// src/extraction/pipeline.ts
+var SONNET_MODEL = "claude-sonnet-4.6";
+var HAIKU_MODEL = "claude-haiku-4.5.20251001";
 var DEFAULT_METADATA_PROVIDER_OPTIONS = {
   anthropic: { thinking: { type: "enabled", budgetTokens: 4096 } }
 };
@@ -29855,7 +30186,7 @@ function sanitizeNulls(obj) {
 function applyExtracted(extracted) {
   const meta = extracted.metadata ?? extracted;
   const policyTypes = Array.isArray(meta.policyTypes) ? meta.policyTypes : meta.policyType ? [meta.policyType] : ["other"];
-  return {
+  const fields = {
     carrier: meta.carrier || meta.security || "Unknown",
     security: meta.security ?? void 0,
     underwriter: meta.underwriter ?? void 0,
@@ -29877,6 +30208,35 @@ function applyExtracted(extracted) {
     extractionStatus: "complete",
     extractionError: ""
   };
+  if (extracted.metadata?.carrierLegalName) fields.carrierLegalName = extracted.metadata.carrierLegalName;
+  if (extracted.metadata?.carrierNaicNumber) fields.carrierNaicNumber = extracted.metadata.carrierNaicNumber;
+  if (extracted.metadata?.carrierAmBestRating) fields.carrierAmBestRating = extracted.metadata.carrierAmBestRating;
+  if (extracted.metadata?.carrierAdmittedStatus) fields.carrierAdmittedStatus = extracted.metadata.carrierAdmittedStatus;
+  if (extracted.metadata?.mga) fields.mga = extracted.metadata.mga;
+  if (extracted.metadata?.underwriter) fields.underwriter = extracted.metadata.underwriter;
+  if (extracted.metadata?.brokerAgency ?? extracted.metadata?.broker) fields.brokerAgency = extracted.metadata.brokerAgency ?? extracted.metadata.broker;
+  if (extracted.metadata?.brokerContactName) fields.brokerContactName = extracted.metadata.brokerContactName;
+  if (extracted.metadata?.brokerLicenseNumber) fields.brokerLicenseNumber = extracted.metadata.brokerLicenseNumber;
+  if (extracted.metadata?.priorPolicyNumber) fields.priorPolicyNumber = extracted.metadata.priorPolicyNumber;
+  if (extracted.metadata?.programName) fields.programName = extracted.metadata.programName;
+  if (extracted.metadata?.isRenewal != null) fields.isRenewal = extracted.metadata.isRenewal;
+  if (extracted.metadata?.isPackage != null) fields.isPackage = extracted.metadata.isPackage;
+  if (extracted.metadata?.coverageForm) fields.coverageForm = extracted.metadata.coverageForm;
+  if (extracted.metadata?.retroactiveDate) fields.retroactiveDate = extracted.metadata.retroactiveDate;
+  if (extracted.metadata?.effectiveTime) fields.effectiveTime = extracted.metadata.effectiveTime;
+  if (extracted.metadata?.insuredDba) fields.insuredDba = extracted.metadata.insuredDba;
+  if (extracted.metadata?.insuredAddress) fields.insuredAddress = extracted.metadata.insuredAddress;
+  if (extracted.metadata?.insuredEntityType) fields.insuredEntityType = extracted.metadata.insuredEntityType;
+  if (extracted.metadata?.insuredFein) fields.insuredFein = extracted.metadata.insuredFein;
+  if (extracted.additionalNamedInsureds?.length) fields.additionalNamedInsureds = extracted.additionalNamedInsureds;
+  if (extracted.limits) fields.limits = extracted.limits;
+  if (extracted.deductibles) fields.deductibles = extracted.deductibles;
+  if (extracted.locations?.length) fields.locations = extracted.locations;
+  if (extracted.vehicles?.length) fields.vehicles = extracted.vehicles;
+  if (extracted.classifications?.length) fields.classifications = extracted.classifications;
+  if (extracted.formInventory?.length) fields.formInventory = extracted.formInventory;
+  if (extracted.taxesAndFees?.length) fields.taxesAndFees = extracted.taxesAndFees;
+  return fields;
 }
 function mergeChunkedSections(metadataResult, sectionChunks) {
   const allSections = [];
@@ -29884,6 +30244,9 @@ function mergeChunkedSections(metadataResult, sectionChunks) {
   let complaintContact = null;
   let costsAndFees = null;
   let claimsContact = null;
+  const allEndorsements = [];
+  const allExclusions = [];
+  const allPolicyConditions = [];
   for (const chunk of sectionChunks) {
     if (chunk.sections) {
       allSections.push(...chunk.sections);
@@ -29892,8 +30255,11 @@ function mergeChunkedSections(metadataResult, sectionChunks) {
     if (chunk.complaintContact) complaintContact = chunk.complaintContact;
     if (chunk.costsAndFees) costsAndFees = chunk.costsAndFees;
     if (chunk.claimsContact) claimsContact = chunk.claimsContact;
+    if (chunk.endorsements?.length) allEndorsements.push(...chunk.endorsements);
+    if (chunk.exclusions?.length) allExclusions.push(...chunk.exclusions);
+    if (chunk.conditions?.length) allPolicyConditions.push(...chunk.conditions);
   }
-  return {
+  const result = {
     metadata: metadataResult.metadata,
     metadataSource: metadataResult.metadataSource,
     coverages: metadataResult.coverages,
@@ -29906,6 +30272,10 @@ function mergeChunkedSections(metadataResult, sectionChunks) {
     },
     totalPages: metadataResult.totalPages
   };
+  if (allEndorsements.length) result.document.endorsements = allEndorsements;
+  if (allExclusions.length) result.document.exclusions = allExclusions;
+  if (allPolicyConditions.length) result.document.conditions = allPolicyConditions;
+  return result;
 }
 function getPageChunks(totalPages, chunkSize = 30) {
   const chunks = [];
@@ -29915,8 +30285,10 @@ function getPageChunks(totalPages, chunkSize = 30) {
   }
   return chunks;
 }
-async function callModel(model, pdfBase64, prompt, maxTokens, providerOptions, log, onTokenUsage) {
-  await log?.(`Calling model (max ${maxTokens} tokens)...`);
+async function callModel(model, pdfBase64, prompt, maxTokens, providerOptions, log, onTokenUsage, pageRange) {
+  const pdfToSend = pageRange ? await extractPageRange(pdfBase64, pageRange[0], pageRange[1]) : pdfBase64;
+  const rangeLabel = pageRange ? ` [pages ${pageRange[0]}\u2013${pageRange[1]}]` : "";
+  await log?.(`Calling model (max ${maxTokens} tokens)${rangeLabel}...`);
   const start = Date.now();
   const { text, usage } = await withRetry(
     () => generateText({
@@ -29925,7 +30297,7 @@ async function callModel(model, pdfBase64, prompt, maxTokens, providerOptions, l
       messages: [{
         role: "user",
         content: [
-          { type: "file", data: pdfBase64, mediaType: "application/pdf" },
+          { type: "file", data: pdfToSend, mediaType: "application/pdf" },
           { type: "text", text: prompt }
         ]
       }],
@@ -30031,7 +30403,9 @@ async function classifyDocumentType(pdfBase64, options) {
     MODEL_TOKEN_LIMITS.classification,
     void 0,
     log,
-    onTokenUsage
+    onTokenUsage,
+    [1, 3]
+    // Only need first 3 pages for classification
   );
   try {
     const parsed = JSON.parse(stripFences(raw));
@@ -30048,7 +30422,7 @@ async function classifyDocumentType(pdfBase64, options) {
 function applyExtractedQuote(extracted) {
   const meta = extracted.metadata ?? extracted;
   const policyTypes = Array.isArray(meta.policyTypes) ? meta.policyTypes : ["other"];
-  return {
+  const fields = {
     carrier: meta.carrier || meta.security || "Unknown",
     security: meta.security ?? void 0,
     underwriter: meta.underwriter ?? void 0,
@@ -30081,11 +30455,38 @@ function applyExtractedQuote(extracted) {
     extractionStatus: "complete",
     extractionError: ""
   };
+  if (meta.carrierLegalName) fields.carrierLegalName = meta.carrierLegalName;
+  if (meta.carrierNaicNumber) fields.carrierNaicNumber = meta.carrierNaicNumber;
+  if (meta.carrierAdmittedStatus) fields.carrierAdmittedStatus = meta.carrierAdmittedStatus;
+  if (meta.coverageForm) fields.coverageForm = meta.coverageForm;
+  if (meta.retroactiveDate) fields.retroactiveDate = meta.retroactiveDate;
+  if (meta.insuredAddress) fields.insuredAddress = meta.insuredAddress;
+  if (extracted.limits) fields.limits = extracted.limits;
+  if (extracted.deductibles) fields.deductibles = extracted.deductibles;
+  if (extracted.warrantyRequirements?.length) fields.warrantyRequirements = extracted.warrantyRequirements;
+  if (extracted.taxesAndFees?.length) fields.taxesAndFees = extracted.taxesAndFees;
+  if (extracted.subjectivities?.length) {
+    fields.enrichedSubjectivities = extracted.subjectivities.map((s) => ({
+      description: s.description,
+      category: s.category ?? void 0,
+      dueDate: s.dueDate ?? void 0,
+      pageNumber: s.pageNumber ?? void 0
+    }));
+  }
+  if (extracted.underwritingConditions?.length) {
+    fields.enrichedUnderwritingConditions = extracted.underwritingConditions.map((c) => ({
+      description: c.description,
+      category: c.category ?? void 0,
+      pageNumber: c.pageNumber ?? void 0
+    }));
+  }
+  return fields;
 }
 function mergeChunkedQuoteSections(metadataResult, sectionChunks) {
   const allSections = [];
   const allSubjectivities = metadataResult.subjectivities || [];
   const allConditions = metadataResult.underwritingConditions || [];
+  const allExclusions = [];
   for (const chunk of sectionChunks) {
     if (chunk.sections) {
       allSections.push(...chunk.sections);
@@ -30096,8 +30497,11 @@ function mergeChunkedQuoteSections(metadataResult, sectionChunks) {
     if (chunk.underwritingConditions) {
       allConditions.push(...chunk.underwritingConditions);
     }
+    if (chunk.exclusions?.length) {
+      allExclusions.push(...chunk.exclusions);
+    }
   }
-  return {
+  const result = {
     metadata: metadataResult.metadata,
     metadataSource: metadataResult.metadataSource,
     coverages: metadataResult.coverages,
@@ -30109,6 +30513,8 @@ function mergeChunkedQuoteSections(metadataResult, sectionChunks) {
     },
     totalPages: metadataResult.totalPages
   };
+  if (allExclusions.length) result.document.exclusions = allExclusions;
+  return result;
 }
 var CHUNK_SIZES = [15, 10, 5];
 async function extractChunkWithRetry(models, pdfBase64, start, end, sizeIndex, promptBuilder, fallbackProviderOptions, log, onTokenUsage, concurrency = 2) {
@@ -30120,7 +30526,9 @@ async function extractChunkWithRetry(models, pdfBase64, start, end, sizeIndex, p
     MODEL_TOKEN_LIMITS.sections,
     void 0,
     log,
-    onTokenUsage
+    onTokenUsage,
+    [start, end]
+    // Only send this chunk's pages
   );
   try {
     return [JSON.parse(stripFences(chunkRaw))];
@@ -30162,7 +30570,9 @@ async function extractChunkWithRetry(models, pdfBase64, start, end, sizeIndex, p
       MODEL_TOKEN_LIMITS.sectionsFallback,
       fallbackProviderOptions,
       log,
-      onTokenUsage
+      onTokenUsage,
+      [start, end]
+      // Only send this chunk's pages
     );
     try {
       return [JSON.parse(stripFences(fallbackRaw))];
@@ -30205,7 +30615,9 @@ async function extractFromPdf(pdfBase64, options) {
     onTokenUsage
   } = options ?? {};
   const resolved = resolveModels(models);
+  const actualPageCount = await getPdfPageCount(pdfBase64);
   await log?.("Pass 1: Extracting metadata...");
+  const metadataPageRange = [1, Math.min(10, actualPageCount)];
   const metadataRaw = await callModel(
     resolved.metadata,
     pdfBase64,
@@ -30213,7 +30625,8 @@ async function extractFromPdf(pdfBase64, options) {
     MODEL_TOKEN_LIMITS.metadata,
     metadataProviderOptions,
     log,
-    onTokenUsage
+    onTokenUsage,
+    metadataPageRange
   );
   let metadataResult;
   try {
@@ -30224,7 +30637,7 @@ async function extractFromPdf(pdfBase64, options) {
     throw new Error(`Metadata JSON parse failed: ${e.message}`);
   }
   await onMetadata?.(metadataRaw);
-  const pageCount = metadataResult.totalPages || 1;
+  const pageCount = actualPageCount;
   await log?.(`Document: ${pageCount} page(s)`);
   const sectionChunks = await extractSectionChunks(
     resolved,
@@ -30292,7 +30705,9 @@ async function extractQuoteFromPdf(pdfBase64, options) {
     onTokenUsage
   } = options ?? {};
   const resolved = resolveModels(models);
+  const actualPageCount = await getPdfPageCount(pdfBase64);
   await log?.("Pass 1: Extracting quote metadata...");
+  const metadataPageRange = [1, Math.min(10, actualPageCount)];
   const metadataRaw = await callModel(
     resolved.metadata,
     pdfBase64,
@@ -30300,7 +30715,8 @@ async function extractQuoteFromPdf(pdfBase64, options) {
     MODEL_TOKEN_LIMITS.metadata,
     metadataProviderOptions,
     log,
-    onTokenUsage
+    onTokenUsage,
+    metadataPageRange
   );
   let metadataResult;
   try {
@@ -30311,7 +30727,7 @@ async function extractQuoteFromPdf(pdfBase64, options) {
     throw new Error(`Quote metadata JSON parse failed: ${e.message}`);
   }
   await onMetadata?.(metadataRaw);
-  const pageCount = metadataResult.totalPages || 1;
+  const pageCount = actualPageCount;
   await log?.(`Quote document: ${pageCount} page(s)`);
   const sectionChunks = await extractSectionChunks(
     resolved,
@@ -30328,107 +30744,13 @@ async function extractQuoteFromPdf(pdfBase64, options) {
   const mergedRaw = JSON.stringify(merged);
   return { rawText: mergedRaw, extracted: merged };
 }
-
-// src/extraction/pdf.ts
-import {
-  PDFDocument,
-  PDFTextField,
-  PDFCheckBox,
-  PDFDropdown,
-  PDFRadioGroup,
-  StandardFonts,
-  rgb
-} from "pdf-lib";
-function getAcroFormFields(pdfDoc) {
-  const form = pdfDoc.getForm();
-  const fields = form.getFields();
-  if (fields.length === 0) return [];
-  return fields.map((field) => {
-    const name = field.getName();
-    if (field instanceof PDFTextField) {
-      return { name, type: "text" };
-    }
-    if (field instanceof PDFCheckBox) {
-      return { name, type: "checkbox" };
-    }
-    if (field instanceof PDFDropdown) {
-      return { name, type: "dropdown", options: field.getOptions() };
-    }
-    if (field instanceof PDFRadioGroup) {
-      return { name, type: "radio", options: field.getOptions() };
-    }
-    return { name, type: "text" };
-  });
-}
-async function fillAcroForm(pdfBytes, mappings) {
-  const pdfDoc = await PDFDocument.load(pdfBytes, { ignoreEncryption: true });
-  const form = pdfDoc.getForm();
-  for (const { acroFormName, value } of mappings) {
-    try {
-      const field = form.getField(acroFormName);
-      if (field instanceof PDFTextField) {
-        field.setText(value);
-      } else if (field instanceof PDFCheckBox) {
-        const lower = value.toLowerCase();
-        if (["yes", "true", "x", "checked", "on"].includes(lower)) {
-          field.check();
-        } else {
-          field.uncheck();
-        }
-      } else if (field instanceof PDFDropdown) {
-        try {
-          field.select(value);
-        } catch {
-        }
-      } else if (field instanceof PDFRadioGroup) {
-        try {
-          field.select(value);
-        } catch {
-        }
-      }
-    } catch {
-    }
-  }
-  form.flatten();
-  return await pdfDoc.save();
-}
-async function overlayTextOnPdf(pdfBytes, overlays) {
-  const pdfDoc = await PDFDocument.load(pdfBytes, { ignoreEncryption: true });
-  const font = await pdfDoc.embedFont(StandardFonts.Helvetica);
-  const pageCount = pdfDoc.getPageCount();
-  for (const overlay of overlays) {
-    if (overlay.page < 0 || overlay.page >= pageCount) continue;
-    const page = pdfDoc.getPage(overlay.page);
-    const { width, height } = page.getSize();
-    const fontSize = overlay.fontSize ?? 10;
-    const x = overlay.x / 100 * width;
-    const y = height - overlay.y / 100 * height - fontSize;
-    if (overlay.isCheckmark) {
-      page.drawText("X", {
-        x,
-        y,
-        size: fontSize,
-        font,
-        color: rgb(0, 0, 0)
-      });
-    } else {
-      page.drawText(overlay.text, {
-        x,
-        y,
-        size: fontSize,
-        font,
-        color: rgb(0, 0, 0)
-      });
-    }
-  }
-  return await pdfDoc.save();
-}
 export {
   AGENT_TOOLS,
   APPLICATION_CLASSIFY_PROMPT,
   CLASSIFY_DOCUMENT_PROMPT,
   CLASSIFY_EMAIL_PROMPT,
   COI_GENERATION_TOOL,
+  CONTEXT_KEY_MAP,
   COVERAGE_COMPARISON_TOOL,
   DOCUMENT_LOOKUP_TOOL,
   EXTRACTION_PROMPT,
@@ -30436,6 +30758,7 @@ export {
   METADATA_PROMPT,
   MODEL_TOKEN_LIMITS,
   PLATFORM_CONFIGS,
+  POLICY_TYPES,
   QUOTE_METADATA_PROMPT,
   SONNET_MODEL,
   applyExtracted,
@@ -30474,11 +30797,13 @@ export {
   createUniformModelConfig,
   enrichSupplementaryFields,
   extractFromPdf,
+  extractPageRange,
   extractQuoteFromPdf,
   extractSectionsOnly,
   fillAcroForm,
   getAcroFormFields,
   getPageChunks,
+  getPdfPageCount,
   mergeChunkedQuoteSections,
   mergeChunkedSections,
   overlayTextOnPdf,