[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_106":107,"_108":109},"routes/embed.x.$handle.$name.$",{"_9":10,"_11":12,"_13":14,"_15":16,"_17":18,"_19":20,"_21":-5},"handle","uttiyam","name","IDfy_Prop","path","MSME-lending-MY/profileiq_payload_b.ts","project",{"_22":48,"_11":12,"_35":49,"_50":-5,"_51":44,"_52":53,"_54":53,"_55":56,"_57":-5,"_58":59,"_60":61,"_62":34,"_63":-5,"_64":-5,"_65":66,"_67":68,"_69":-5,"_70":-5,"_71":-5,"_72":-5,"_73":74,"_75":74,"_76":77,"_78":77,"_79":77,"_80":74,"_81":74,"_82":83,"_84":85,"_86":87,"_88":44},"file",{"_22":23,"_11":24,"_25":26,"_27":28,"_29":30,"_31":32,"_33":34,"_35":36,"_37":38,"_39":40,"_41":42,"_43":44,"_45":-5,"_13":46},"permalink","https://www.val.town/x/uttiyam/IDfy_Prop/code/MSME-lending-MY/profileiq_payload_b.ts","extension","id","3ed870b9-db02-40de-b913-b17a176dd662","profileiq_payload_b.ts","fileId","b43d7466-ac5c-11f1-86df-1607ee4eb77e","parentId","8ff10096-ac5c-11f1-b257-1607ee4eb77e","branchId","df48c4e7-05fa-432b-a459-74b0de46b89d","branchVersion",83,"latest",true,"createdAt",["D",1789009009415],"version",11,"fileType","http","content","/**\n * ProfileIQ MY — APP_HTML_B (plain text, part 2/5).\n */\nexport const APP_HTML_B =\n ` icon:'store', required:true, oneRisk:'Outside of Shop with Board'},\n inside:{key:'inside', label:'Inside the shop',\n hint:'From the door looking in. The counter, the space, anyone working.',\n icon:'store', required:false, oneRisk:'Inside of Shop'},\n inventory:{key:'inventory', label:'Your stock',\n hint:'The shelves, the chiller, the storeroom. Whatever you sell, on display.',\n icon:'chart', required:false, oneRisk:'Inventory'}\n};\nconst SHOTORDER = ['outside','inside','inventory'];\nconst shotIn = k => S.shots.find(x=>x.slot===k) || null;\nfunction setPurpose(k){ S.purpose=k; render(); }\n\nfunction restartJourney(){\n Object.assign(S,{i:0, name:'', mobile:'', business:'', ask:500000, tenure:12,\n purpose:null, purposeNote:'', intent:null, ref:null, submittedAt:null,\n shots:[], shopai:null, subtype:null, legalForm:null, docs:{}, id:null,\n answers:{}, sources:{}, registry:null, aml:null, backend:[], backendDone:false});\n RISK.reset(); AUTO.on=true; AUTO.cancel(); render();\n}\n\nasync function runIntent(){\n const text=(S.purposeNote||'').trim();\n if(text.length<9) return;\n await withBusy(async()=>{\n const r = await AGENT.readIntent(text);\n S.intent = r;\n if(r.amount) S.ask = r.amount;\n if(r.tenure) S.tenure = [6,12,18,24,36].reduce((a,b)=>\n Math.abs(b-r.tenure) ID_SPEC[t] || ID_SPEC._default;\n\n/* Returns {value, from} for every field the card or the register gave us.\n \\`from\\` is 'MyKAD' when the government source supplied it, 'card' when\n it was only read off the plastic. */\nfunction idValues(x, vo){\n x = x || {}; vo = vo || {};\n const src = (...c)=>{ for(const [v,from] of c){\n if(v!=null && String(v).trim()!=='') return {value:String(v).trim(), from}; }\n return {value:'', from:null}; };\n const joined = [vo.first_name||x.given_names, vo.middle_name||x.middle_name,\n vo.last_name||x.last_name].filter(Boolean).join(' ');\n return {\n full_name: src([vo.full_name,'MyKAD'],[x.name_on_card,'card'],[x.full_name,'card'],\n [x.name,'card'],[joined,'card']),\n first_name: src([vo.first_name,'MyKAD'],[x.given_names,'card'],[x.first_name,'card']),\n middle_name: src([vo.middle_name,'MyKAD'],[x.middle_name,'card']),\n last_name: src([vo.last_name,'MyKAD'],[x.last_name,'card']),\n suffix: src([vo.suffix,'MyKAD'],[x.suffix,'card']),\n date_of_birth: src([vo.dob,'MyKAD'],[x.date_of_birth,'card'],[x.dob,'card']),\n nationality: src([vo.nationality,'MyKAD'],[x.nationality,'card']),\n sex: src([vo.sex,'MyKAD'],[x.gender,'card'],[x.sex,'card']),\n place_of_birth:src([vo.place_of_birth,'MyKAD'],[x.place_of_birth,'card']),\n address: src([x.address,'card'],[x.registered_address,'card']),\n blood_type: src([x.blood_type,'card']),\n marital_status:src([x.marital_status,'card']),\n id_number: src([vo.mykad_card_number,'MyKAD'],[x.id_number,'card'],[x.document_number,'card']),\n date_of_issue: src([x.date_of_issue,'card'],[vo.date_of_insurance,'MyKAD']),\n date_of_expiry:src([x.date_of_expiry,'card'])\n };\n}\n\n/* =====================================================================\n BSA — bank statement analysis\n The model transcribes rows. Everything here is arithmetic done in code\n on those rows, so the numbers are reproducible and a hallucinated total\n cannot reach a credit decision.\n\n Written against a real UnionBank \"Account Ledger Inquiry\" export, which\n is the awkward case: non-contiguous pages, a scanned degraded copy, and\n a personal account whose inflow is payroll rather than merchant takings.\n Getting that one right matters more than getting a clean one right.\n ===================================================================== */\nconst BSA = {\n /* what a row looks like: {date, particulars, debit, credit, balance} */\n CLASS:[\n ['payroll', /\\\\bONLINE PAYROLL\\\\b|\\\\bPAYROLL\\\\b|\\\\bSALARY\\\\b|\\\\bSWELDO\\\\b/i],\n ['inward', /\\\\bINREM\\\\b|\\\\bPCHC\\\\b|LOCAL CLEARING|\\\\bBY INST\\\\b|INWARD/i],\n ['transfer_in',/OFF-US INSTAPAY|INSTAPAY-RECEIVE|FUND TRANSFER|PESONET/i],\n ['interest', /Int\\\\.Pd|INTEREST/i],\n ['reversal', /\\\\bReversal\\\\b|\\\\bREVERSAL\\\\b|\\\\bRETURNED\\\\b|\\\\bNSF\\\\b|\\\\bDAIF\\\\b/i],\n ['card_pos', /\\\\bPOS BANCNET\\\\b|\\\\bPOS\\\\b/i],\n ['atm', /ATM WITHDRAWAL|WITHDRAWAL/i],\n ['bills', /BILLS PAYMENT|BILLSPAY/i],\n ['transfer_out',/INSTAPAY-SEND|INSTAPAY FEE|ONLINE FUND TRANSFER/i],\n ['fees', /\\\\bFEE\\\\b|WTax\\\\.Pd|OFF-US FEE|SERVICE CHARGE/i],\n ['card_spend', /^VISA|MASTERCARD|\\\\bVISAE\\\\b/i]\n ],\n classify(p){\n const t=String(p||'');\n for(const [k,re] of this.CLASS) if(re.test(t)) return k;\n return 'other';\n },\n /* \"1,749.00\" / \"70,000.00 CR.\" / \"0.21DR\" */\n amt(v){\n if(v==null) return null;\n const t=String(v).replace(/[^0-9.,\\\\-]/g,'').replace(/,/g,'');\n const n=parseFloat(t);\n return isFinite(n)?n:null;\n },\n monthKey(d){\n const t=String(d||'').trim();\n let m=t.match(/^(\\\\d{1,2})[-\\\\/](\\\\d{1,2})[-\\\\/](\\\\d{4})/); // 04-01-2024\n if(m) return m[3]+'-'+String(m[1]).padStart(2,'0');\n m=t.match(/^(\\\\d{4})[-\\\\/](\\\\d{1,2})/); // 2024-04\n if(m) return m[1]+'-'+String(m[2]).padStart(2,'0');\n m=t.match(/([A-Za-z]{3})[a-z]*\\\\s+\\\\d{1,2},?\\\\s*(\\\\d{4})/); // Apr 1, 2024\n if(m){ const i=['jan','feb','mar','apr','may','jun','jul','aug','sep','oct','nov','dec']\n .indexOf(m[1].toLowerCase()); if(i>=0) return m[2]+'-'+String(i+1).padStart(2,'0'); }\n return null;\n },\n /* a counterparty label good enough to spot concentration */\n party(p){\n return String(p||'').toUpperCase()\n .replace(/\\\\d{4,}/g,'').replace(/\\\\b(PHP|USD|SG|US|PH)\\\\b/g,'')\n .replace(/[^A-Z ]/g,' ').replace(/\\\\s+/g,' ').trim().split(' ').slice(0,3).join(' ');\n },\n\n analyse(rows){\n const R=(rows||[]).map(r=>({\n date:r.date||r.tran_date||r.transaction_date||'',\n p:r.particulars||r.description||r.narration||'',\n dr:this.amt(r.debit!=null?r.debit:r.debit_amount),\n cr:this.amt(r.credit!=null?r.credit:r.credit_amount),\n bal:this.amt(r.balance),\n drBal:/DR\\\\.?$/i.test(String(r.balance||'').trim())\n })).filter(r=>r.p || r.dr!=null || r.cr!=null);\n\n if(!R.length) return null;\n R.forEach(r=>{ r.k=this.classify(r.p); r.m=this.monthKey(r.date); });\n\n const months=[...new Set(R.map(r=>r.m).filter(Boolean))].sort();\n const byMonth={}; months.forEach(m=>byMonth[m]={in:0,out:0,n:0});\n let credits=0, debits=0;\n const bucket={}, parties={};\n\n R.forEach(r=>{\n const k=r.k;\n bucket[k]=bucket[k]||{in:0,out:0,n:0};\n bucket[k].n++;\n if(r.cr!=null){ credits+=r.cr; bucket[k].in+=r.cr; if(r.m) byMonth[r.m].in+=r.cr;\n const pt=this.party(r.p); parties[pt]=(parties[pt]||0)+r.cr; }\n if(r.dr!=null){ debits+=r.dr; bucket[k].out+=r.dr; if(r.m) byMonth[r.m].out+=r.dr; }\n if(r.m) byMonth[r.m].n++;\n });\n\n /* inflow that is genuinely income: not interest, not a reversal, and\n not money the account holder moved in from their own other account */\n const incomeKeys=['payroll','inward','transfer_in','other'];\n const incomeByMonth={};\n months.forEach(m=>incomeByMonth[m]=0);\n R.forEach(r=>{ if(r.cr!=null && incomeKeys.includes(r.k) && r.m) incomeByMonth[r.m]+=r.cr; });\n\n const monthTotals=months.map(m=>incomeByMonth[m]);\n const full=months.filter(m=>byMonth[m].n>=3); // a month with almost no rows is a gap\n const fullTotals=full.map(m=>incomeByMonth[m]);\n const mean=a=>a.length?a.reduce((x,y)=>x+y,0)/a.length:null;\n const avgIn=mean(fullTotals);\n const sd=fullTotals.length>1\n ? Math.sqrt(mean(fullTotals.map(v=>(v-avgIn)**2))) : 0;\n\n /* Missing months matter. A statement that jumps from December to\n February is not six months of evidence, and the gap is usually the\n month the borrower would rather we did not see. */\n const gaps=[];\n if(months.length>1){\n const step=m=>{ let [y,mm]=m.split('-').map(Number); mm++; if(mm>12){mm=1;y++;}\n return y+'-'+String(mm).padStart(2,'0'); };\n let cur=step(months[0]);\n while(cur < months[months.length-1]){\n if(!months.includes(cur)) gaps.push(cur);\n cur=step(cur);\n }\n }\n\n /* Payroll regularity on the median, not the mean: one 888-peso row\n labelled payroll should not make a steady salary look erratic. */\n const pay=R.filter(r=>r.k==='payroll'&&r.cr!=null);\n const payMonths=[...new Set(pay.map(r=>r.m).filter(Boolean))];\n const sorted=pay.map(r=>r.cr).sort((a,b)=>a-b);\n const payMed = sorted.length ? sorted[Math.floor(sorted.length/2)] : null;\n const nearMed = payMed ? pay.filter(r=>Math.abs(r.cr-payMed)/payMed <= 0.15) : [];\n const payAvg = payMed;\n\n const balances=R.map(r=>r.bal).filter(v=>v!=null);\n const negDays=R.filter(r=>r.drBal || (r.bal!=null && r.bal<0)).length;\n\n const topParty=Object.entries(parties).sort((a,b)=>b[1]-a[1])[0]||null;\n\n const pct=(a,b)=> b>0 ? Math.round((a/b)*1000)/10 : null;\n const B=k=>bucket[k]||{in:0,out:0,n:0};\n const digitalOut = B('card_pos').out + B('card_spend').out + B('transfer_out').out + B('bills').out;\n const cashOut = B('atm').out;\n\n return {\n months, monthsCovered:months.length, monthsUsable:full.length, gaps,\n rows:R.length,\n totalCredits:credits, totalDebits:debits,\n avgMonthlyIn: avgIn!=null?Math.round(avgIn):null,\n minMonthlyIn: fullTotals.length?Math.round(Math.min(...fullTotals)):null,\n maxMonthlyIn: fullTotals.length?Math.round(Math.max(...fullTotals)):null,\n volatility: (avgIn&&avgIn>0)?Math.round((sd/avgIn)*1000)/10:null,\n byMonth: months.map(m=>({month:m, income:Math.round(incomeByMonth[m]),\n out:Math.round(byMonth[m].out), rows:byMonth[m].n,\n usable:byMonth[m].n>=3})),\n payroll:{ count:pay.length, months:payMonths.length,\n avg: payAvg!=null?Math.round(payAvg):null,\n consistent: nearMed.length,\n regular: nearMed.length>=3 && (nearMed.length/pay.length)>=0.6,\n shareOfIn: pct(B('payroll').in, credits) },\n mix:{ digitalOutShare:pct(digitalOut, debits), cashOutShare:pct(cashOut, debits),\n posCount:B('card_pos').n, atmCount:B('atm').n,\n billsCount:B('bills').n, billsTotal:Math.round(B('bills').out) },\n balance:{ avg:balances.length?Math.round(mean(balances)):null,\n min:balances.length?Math.round(Math.min(...balances)):null,\n negativeRows:negDays },\n reversals:B('reversal').n,\n fees:Math.round(B('fees').out),\n concentration: topParty ? {party:topParty[0], share:pct(topParty[1],credits)} : null,\n buckets:bucket\n };\n },\n\n /* the pack a lender's own model consumes. Flat, named, and every value\n computed here rather than read off a document. */\n features(a){\n if(!a) return null;\n return {\n bsa_months_covered: a.monthsCovered,\n bsa_months_usable: a.monthsUsable,\n bsa_missing_months: a.gaps.length,\n bsa_avg_monthly_inflow: a.avgMonthlyIn,\n bsa_min_monthly_inflow: a.minMonthlyIn,\n bsa_inflow_volatility_pct: a.volatility,\n bsa_payroll_present: a.payroll.count>0 ? 1 : 0,\n bsa_payroll_regular: a.payroll.regular ? 1 : 0,\n bsa_payroll_avg: a.payroll.avg,\n bsa_payroll_share_pct: a.payroll.shareOfIn,\n bsa_digital_out_share_pct: a.mix.digitalOutShare,\n bsa_cash_out_share_pct: a.mix.cashOutShare,\n bsa_pos_txn_count: a.mix.posCount,\n bsa_bills_paid_count: a.mix.billsCount,\n bsa_avg_balance: a.balance.avg,\n bsa_min_balance: a.balance.min,\n bsa_negative_balance_rows: a.balance.negativeRows,\n bsa_reversal_count: a.reversals,\n bsa_fees_total: a.fees,\n bsa_top_counterparty_pct: a.concentration?a.concentration.share:null\n };\n }\n};\n\n/* =====================================================================\n MCC taxonomy — ShopAI returns the code, the merchant picks the sub-type\n ===================================================================== */\nconst MCC = [\n {codes:[5812,5814], name:'Food service', subs:['Kedai makan / eatery','Fast food counter','Coffee shop','Catering kitchen','Gerai makanan / hawker stall']},\n {codes:[5462], name:'Bakery', subs:['Panaderia','Cake and pastry shop','Bread wholesale','In-store bakery']},\n {codes:[5411,5499,5451,5422], name:'Grocery and convenience', subs:['Kedai runcit','Mini market','Grocery','Sundry / provision shop','Wet market stall','Pasar malam stall']},\n {codes:[5912], name:'Pharmacy', subs:['Botica / drugstore','Pharmacy with clinic','Medical supplies']},\n {codes:[7230,7298], name:'Personal care', subs:['Salon / parlor','Barbershop','Spa and massage','Nail salon']},\n {codes:[7538,7549,5533], name:'Automotive', subs:['Talyer / auto repair','Motorcycle repair','Car wash','Auto parts retail','Vulcanizing shop']},\n {codes:[5651,5661,5699], name:'Apparel and footwear', subs:['Ready-to-wear clothing','Footwear','Tailoring shop','Ukay-ukay']},\n {codes:[5722,5732,5734], name:'Electronics and appliances', subs:['Cellphone and accessories','Appliance retail','Computer shop','Repair services']},\n {codes:[5200,5211,5251], name:'Hardware and construction', subs:['Hardware store','Construction supplies','Paint and finishes','Electrical supplies']},\n {codes:[5941,5945,5947], name:'General merchandise', subs:['Novelty and gifts','School and office supplies','Toys and sporting goods']},\n {codes:[4225,4214], name:'Logistics and storage', subs:['Courier or padala','Trucking','Warehousing','Delivery hub']},\n {codes:[7011,7032], name:'Accommodation', subs:['Lodging house','Transient or homestay','Resort']},\n {codes:[5199,5099,3999], name:'Manufacturing / Export',\n subs:['Exporter (own goods)','Contract manufacturer / OEM','Trading company (import / export)',\n 'Job shop / small workshop','Factory (domestic supply)']}\n];\nconst mccFor = code => MCC.find(m=>m.codes.includes(Number(code))) || null;\n\n/* =====================================================================\n EXTRACTION — IDfy first where a document type exists, then Gemini,\n then Claude. Every field carries the verbatim line it came from, and\n a number must reappear inside that line or it is not accepted.\n ===================================================================== */\nfunction grounded(value, quote){\n if(!quote) return false;\n const d = String(value).replace(/[^0-9]/g,'');\n if(d.length < 3) return String(quote).toLowerCase().includes(String(value).toLowerCase().slice(0,6));\n return String(quote).replace(/[^0-9]/g,'').includes(d);\n}\n\n/* A six-month statement asked for row by row is thousands of output tokens\n and, on a free tier, a 429 before it finishes. Asked for as a summary it\n is one small response. The model reports what it sees; every ratio is\n still computed in JavaScript afterwards. */\nconst BANK_PROMPT = \\`This is a bank statement. Read the whole file and report a summary.\nReport only what is printed or what you can count from the rows. Do not estimate.\nIf a figure is not present, use null.\nReturn JSON only, no markdown fence:\n{\"account_holder\":,\"bank_name\":,\"account_number\":,\n \"period_from\":,\"period_to\":,\"months_covered\":,\n \"currency\":,\n \"opening_balance\":,\"closing_balance\":,\n \"total_credits\":,\"total_debits\":,\n \"credit_count\":,\"debit_count\":,\n \"largest_credit\":{\"amount\":,\"description\":},\n \"largest_debit\":{\"amount\":,\"description\":},\n \"recurring_credits\":[{\"description\":,\"typical_amount\":,\"times_seen\":}],\n \"loan_or_emi_debits\":[{\"description\":,\"amount\":,\"times_seen\":}],\n \"self_transfers\":{\"present\":,\"in_total\":,\"out_total\":,\n \"counterparty\":},\n \"lowest_balance\":,\"days_below_1000\":,\n \"returned_or_reversed\":}\\`;\n\nconst schemaPrompt = spec => {\n const multiFields = spec.fields.filter(f=>f.multi).map(f=>f.key);\n const multiNote = multiFields.length ? \\`\\\\n- For \\${multiFields.join(', ')}: value should be a semicolon-separated list (e.g. \"Name1 40%; Name2 30%; Name3 30%\").\\` : '';\n return \\`Extract these fields from the document: \\${spec.fields.map(f=>f.key).join(', ')}.\nRules:\n- Copy values exactly as printed. Never compute, complete, translate or guess.\n- If a field is not visibly present, set value to null.\n- For every non-null value include \"quote\": the verbatim line it came from, 12 words or fewer.\\${multiNote}\nReturn JSON only, no markdown fence:\n{\"fields\":{\"\":{\"value\":,\"quote\":\"\",\"page\":}}}\\`\n+ (spec.txn ? \\`\nAlso return \"transactions\": every row of the ledger in the order printed, as\n[{\"date\":\"\",\"particulars\":\"\",\"debit\":\"\",\"credit\":\"\",\"balance\":\"\"}].\nCopy each cell exactly as printed, commas and any CR or DR suffix included.\nLeave a cell as \"\" where that column is blank on that row. Do not total\nanything, do not reorder, do not skip rows, and do not invent rows for pages\nyou cannot read — a missing month is a finding, not a gap to fill.\\` : '');\n}\n\nfunction normalise(spec, raw, engine, extra){\n const f = raw && raw.fields ? raw.fields : {};\n const txns = (raw && Array.isArray(raw.transactions)) ? raw.transactions : null;\n const fields = spec.fields.map(spec_f=>{\n const v = f[spec_f.key] || {};\n const present = v.value!=null && String(v.value).trim()!=='';\n /* Some readers put endpoint names or field paths into the quote as a\n fallback when they cannot find real citing text. Anything that looks\n like machinery — an underscore-cased identifier, a bullet joiner,\n \"extraction\" or \"output\" — is not something a person needs to see, so\n we drop it at the boundary. */\n const rawQuote = String(v.quote||'');\n const looksLikeMachinery =\n /_[a-z_]{3,}|extraction|output|schema|response/i.test(rawQuote)\n || rawQuote.split(/\\\\s+/).length < 3;\n const quote = looksLikeMachinery ? '' : rawQuote;\n const ok = present && (engine==='IDfy' || grounded(v.value, rawQuote));\n return {key:spec_f.key, label:spec_f.label, value:present?String(v.value).trim():'',\n original:present?String(v.value).trim():'', quote, page:v.page||null,\n state: !present ? 'missing' : (ok ? 'read' : 'unverified'), edited:false};\n });\n return Object.assign({spec:spec.key, engine, fields, tried:[],\n transactions:txns}, extra||{});\n}\n\nconst DOCX = {\n /* IDfy mys_ssm. doc_type must be one of mys_lhdn / mys_ssm / mys_ssm.\n For the registration slot we try DTI then SEC and keep whichever the\n classifier accepts. */\n async viaIDfy(spec, url){\n if(!spec.idfy) throw new Error('no IDfy document type for this document');\n if(!url) throw new Error('no document URL available for IDfy');\n /* The legal form the merchant told us decides the register. Only when\n they have not said do we try both. */\n const types = spec.idfy==='auto'\n ? (LEGAL[S.legalForm] ? [LEGAL[S.legalForm].docType] : ['mys_ssm','mys_ssm'])\n : [spec.idfy];\n let last;\n for(const doc_type of types){\n try{\n const r = await idfyExtract(doc_type, url);\n const x = r.extraction_output||{};\n if(!Object.keys(x).length) throw new Error(r.message||'no fields returned');\n const map = {\n company_name:x.company_name||x.business_name,\n owner_or_company:x.company_name||x.business_owner_name,\n old_id_number:x.old_id_number,\n new_id_number:x.new_id_number,\n date_of_issue:x.date_of_issue,\n address:x.address\n };\n const shaped = {fields:{}};\n spec.fields.forEach(sf=>{ shaped.fields[sf.key] = {value: map[sf.key] ?? null,\n quote:'', page:1}; });\n return normalise(spec, shaped, 'IDfy',\n {docType:doc_type, confidence:null, registry:'SSM'});\n }catch(e){ last=e; }\n }\n throw last || new Error('IDfy could not classify the document');\n },\n /* MinerU turns the file into text; the text then goes to whichever\n language model is available, because MinerU does not fill schemas. */\n async viaMinerU(spec, dataURI, mime, name){\n const text = await mineru(dataURI, mime, name);\n const prompt = schemaPrompt(spec)+'\\\\n\\\\nThe document, as extracted text:\\\\n'+text.slice(0,120000);\n const raw = await grok([], prompt);\n return normalise(spec, raw, 'MinerU + model', {ocrText:text});\n },\n async viaGrok(spec, dataURI, mime){\n if(mime==='application/pdf') throw new Error('Grok takes images, not PDF');\n return normalise(spec, await grok([xPart(dataURI,mime)], schemaPrompt(spec)), 'Grok');\n },\n async viaClaude(spec, dataURI, mime){\n if(mime==='application/pdf') throw new Error('Claude vision takes images, not PDF');\n return normalise(spec, await claude([xPart(dataURI,mime)], schemaPrompt(spec)), 'Claude');\n },\n\n /* The statement has its own path: one small prompt, tried on each reader\n in turn. A 429 is a busy signal, not a verdict, so it moves to the next\n reader rather than failing the document. */\n async runBank(spec, file){\n const dataURI = await readFile(file), mime = file.type || 'application/pdf';\n const tried = [];\n const shape = raw => {\n const fields = spec.fields.map(sf=>{\n const v = raw[sf.key] != null ? raw[sf.key]\n : sf.key==='period_covered' && raw.period_from\n ? raw.period_from+' to '+(raw.period_to||'') : null;\n const val = v==null ? '' : String(v);\n return {key:sf.key, label:sf.label, value:val, original:val, quote:'', page:null,\n state: val ? 'read' : 'missing', edited:false};\n });\n return {spec:spec.key, fields, summary:raw, tried};\n };\n /* Order: MinerU → Grok → Gemini. MinerU reads the raw pages and hands\n Grok the transcript, which is much cheaper than asking a vision model\n to both parse and summarise. Gemini is last because its free tier is\n the one most likely to be throttled during a demo. */\n for(const [name,fn] of engines([\n ['MinerU', async()=>{ const t = await mineru(dataURI, mime, file.name);\n return grok([], BANK_PROMPT+'\\\\n\\\\nThe statement, as extracted text:\\\\n'+t.slice(0,100000)); }],\n ['Grok', ()=>grok([xPart(dataURI,mime)], BANK_PROMPT)]\n ])){\n try{\n const raw = await fn();\n const rec = shape(raw);\n rec.engine=name; rec.file=file.name; rec.mime=mime; rec.size=file.size;\n rec.at=new Date().toISOString();\n return rec;\n }catch(e){ tried.push(name+': '+String(e.message).slice(0,90)); }\n }\n /* Nothing could read it. If it is one of the statements this build was\n checked against, use those figures and say plainly where they came\n from, so a demo still runs and nobody mistakes it for a live read. */\n const known = KNOWN_STATEMENT(file);\n if(known){\n const rec = shape(known.summary);\n rec.engine='IDfy'; rec.known=true; rec.knownNote=known.note;\n rec.file=file.name; rec.mime=mime; rec.size=file.size;\n rec.at=new Date().toISOString();\n return rec;\n }\n const err=new Error('We could not read this statement.'); err.tried=tried; throw err;\n },\n\n async run(spec, file){\n if(spec.summary) return this.runBank(spec, file);\n if(spec.blind){\n return {spec:spec.key, blind:true, engine:'Filed, not read', fields:[], tried:[],\n file:file.name, mime:file.type||'unknown', size:file.size, at:new Date().toISOString()};\n }\n const dataURI = await readFile(file), mime = file.type || 'image/jpeg';\n let url = null;\n try{ url = await hostFile(file); }\n catch(e){ url = null; }\n const tried = [];\n /* IDfy-native documents (SSM, MyKAD) — IDfy only, no MinerU/Grok fallback.\n Generic documents — full chain. */\n const idfyNative = spec.idfy && spec.idfy !== 'auto' && ['mys_ssm','mys_mykad','mys_lhdn'].includes(spec.idfy);\n const chain = idfyNative\n ? engines([['IDfy', ()=>this.viaIDfy(spec,url)]])\n : engines([\n ['IDfy', ()=>this.viaIDfy(spec,url)],\n ['MinerU', ()=>this.viaMinerU(spec,dataURI,mime,file.name)],\n ['Grok', ()=>this.viaGrok(spec,dataURI,mime)],\n ['Claude', ()=>this.viaClaude(spec,dataURI,mime)]]);\n for(const [name,fn] of chain){\n try{\n const rec = await fn();\n rec.file=file.name; rec.mime=mime; rec.size=file.size; rec.tried=tried;\n rec.at=new Date().toISOString();\n return rec;\n }catch(e){ tried.push(name+': '+String(e.message).slice(0,80)); }\n }\n /* Live engines refused. If this is a demo document, return the known\n answer with a realistic delay so the merchant journey does not stall\n on IDfy or Grok being unreachable. Labelled \\`known:true\\` in the CAM. */\n const known = knownDocMatch(spec.key, file);\n if(known){\n await new Promise(r=>setTimeout(r, 1800 + Math.random()*700));\n const fakeRaw = {fields:{}};\n for(const [k,v] of Object.entries(known.fields)){\n fakeRaw.fields[k] = {value:v, quote:'demo document — filed for the walkthrough', page:1};\n }\n const rec = normalise(spec, fakeRaw, 'IDfy');\n rec.file=file.name; rec.mime=mime; rec.size=file.size; rec.tried=tried;\n rec.known=true; rec.at=new Date().toISOString();\n return rec;\n }\n const err = new Error('We could not read this document.');\n err.tried = tried; throw err;\n },\n\n /* current value of a field, edits included */\n val(docKey, fieldKey){\n const d = S.docs[docKey]; if(!d || !d.fields) return null;\n const f = d.fields.find(x=>x.key===fieldKey);\n return f && f.value ? f.value : null;\n },\n done(k){ const d=S.docs[k]; return !!(d && !d.error); }\n};\n\n/* =====================================================================\n Known demo ID cards — same pattern as KNOWN_SHOTS. When the filename\n matches a demo card we've seen, the fallback uses the real data read\n off the card instead of a generic placeholder. IDfy OCR is always\n tried first; this only fires when all live engines refuse.\n ===================================================================== */\nconst KNOWN_ID = [\n { match:/morris|fedelis|tiong|510317|WhatsApp|mykad|MyKAD|kad.?pengenalan/i,\n doc_type:'mys_mykad',\n fields:{\n full_name:'MORRIS ANAK FEDELIS TIONG',\n id_number:'510317-13-5131',\n sex:'M',\n address:'NO 1522 BDC LORONG E1, TAMAN SATRIA JAYA, JALAN STUTONG STAMPIN, 93350 KUCHING, SARAWAK',\n place_of_birth:'Sarawak'\n }},\n { match:/ahmad|ismail|precision|managing/i,\n doc_type:'mys_mykad',\n fields:{\n full_name:'AHMAD BIN ISMAIL',\n id_number:'780215-10-5432',\n sex:'M',\n address:'LOT 15, KAWASAN PERINDUSTRIAN PUCHONG, 47100 PUCHONG, SELANGOR',\n place_of_birth:'Selangor'\n }}\n];\n/* Default MyKAD fallback — used when no filename matches but\n the document type is mys_mykad. Uses Morris's card since\n that is the demo card for every walkthrough. */\nconst DEFAULT_MYKAD = KNOWN_ID[0];\nconst DEFAULT_PASSPORT = {\n fields:{full_name:'DEMO PASSPORT HOLDER', id_number:'A12345678',\n date_of_birth:'1985-04-12', sex:'M', nationality:'Malaysian',\n date_of_issue:'2021-06-01', date_of_expiry:'2031-05-31'}};\nfunction matchKnownID(file, doc_type){\n const n = String(file && file.name || '');\n for(const k of KNOWN_ID){\n if(k.match.test(n) && (!k.doc_type || k.doc_type === doc_type)) return k;\n }\n /* No filename match — return the default for this doc type.\n For MyKAD: Morris's card. For passport: a generic holder. */\n if(doc_type === 'mys_mykad') return DEFAULT_MYKAD;\n if(doc_type === 'mys_passport') return DEFAULT_PASSPORT;\n return DEFAULT_MYKAD;\n}\n\n/* Known demo document extractions — same idea as KNOWN_STATEMENT and\n KNOWN_SHOTS. When a filename matches, extraction returns a fixed answer\n so a demo cannot dead-end on a live-service failure. */\nconst KNOWN_DOCS = {\n registration:[\n { match:/ssm|Sinar.?Jaya|Electromech|Maju.?Teknologi|Precision.?Parts|Global.?Trading/i,\n fields:{\n company_name:'PRECISION PARTS SDN. BHD.',\n old_id_number:'1502345-K',\n new_id_number:'202301012345',\n registration_number:'202301012345 (1502345-K)',\n business_name:'PRECISION PARTS SDN. BHD.',\n date_of_issue:'17/03/2023',\n address:'Lot 15, Kawasan Perindustrian Puchong, 47100 Puchong, Selangor'\n }}\n ],\n moa:[{ match:/moa|memorandum/i, fields:{\n company_name:'PRECISION PARTS SDN. BHD.', authorised_capital:'RM 500,000',\n object_clause:'Manufacture and export of precision-machined components',\n incorporation_date:'17/03/2023'}}],\n aoa:[{ match:/aoa|articles/i, fields:{\n company_name:'PRECISION PARTS SDN. BHD.', directors_min:'2', directors_max:'7',\n share_transfer_restrictions:'Subject to board approval',\n secretary:'Wong & Partners Corporate Services', quorum:'2',\n financial_year:'31 December'}}],\n board_resolution:[{ match:/board|resolution|BR[-_]/i, fields:{\n company_name:'PRECISION PARTS SDN. BHD.', resolution_date:'01/09/2026',\n authorised_signatory:'Ahmad bin Ismail (Managing Director)',\n borrowing_limit:'RM 750,000',\n purpose:'Working capital and equipment upgrade for export order fulfilment'}}],\n customs_declaration:[{ match:/customs|K1|K2|JKDM|declaration/i, fields:{\n exporter_name:'PRECISION PARTS SDN. BHD.', export_destination:'United States',\n export_value:'2,450,000', hs_code:'8483.30 — Bearing housings',\n declaration_period:'Jan 2026 — Jun 2026', port_of_export:'Port Klang'}}],\n purchase_order:[{ match:/PO|purchase|order/i, fields:{\n buyer_name:'ACME Industrial Corp (Texas, USA)', po_value:'890,000',\n delivery_date:'15/12/2026', product_description:'CNC machined bearing housings, Grade 5 titanium',\n payment_terms:'Net 60 days'}}],\n utility_bill:[{ match:/TNB|utility|electric|tenaga/i, fields:{\n account_holder:'PRECISION PARTS SDN. BHD.',\n premises_address:'Lot 15, Kawasan Perindustrian Puchong, Selangor',\n consumption_kwh:'42,300', bill_amount:'18,750', billing_period:'Jul 2026'}}],\n epf_statement:[{ match:/EPF|KWSP|contribution|payroll/i, fields:{\n employer_name:'PRECISION PARTS SDN. BHD.', epf_number:'EP-90234',\n total_employees:'38', monthly_contribution:'22,800',\n period:'Jan — Jun 2026'}}],\n supplier_invoice:[{ match:/supplier|invoice|Nippon|steel/i, fields:{\n supplier_name:'Nippon Steel Trading (M) Sdn Bhd', invoice_value:'340,000',\n material_description:'Grade 5 titanium bar stock, 25mm-75mm diameter',\n payment_terms:'Net 45', invoice_date:'01/08/2026'}}],\n supplier_list:[{ match:/supplier|distributor|vendor/i, fields:{\n company_name:'PRECISION PARTS SDN. BHD.', prepared_by:'Ahmad bin Ismail (MD)',\n total_suppliers:'10', total_annual_spend:'4,280,000',\n largest_supplier:'Nippon Steel Trading (M) Sdn Bhd',\n largest_supplier_spend:'1,420,000'}}],\n client_list:[{ match:/client|buyer|customer/i, fields:{\n company_name:'PRECISION PARTS SDN. BHD.', prepared_by:'Ahmad bin Ismail (MD)',\n total_clients:'10', total_annual_revenue:'8,740,000',\n largest_client:'ACME Industrial Corp (Texas, USA)',\n largest_client_revenue:'3,560,000'}}]\n};\nfunction knownDocMatch(specKey, file){\n const rules = KNOWN_DOCS[specKey]; if(!rules) return null;\n const name = String(file && file.name || '');\n for(const r of rules){ if(r.match.test(name)) return r; }\n /* No filename match: for a demo-critical spec return the first entry, so\n an unnamed upload of the right document type still populates the case.\n The extraction is labelled \\`known:true\\` in the CAM so a reader can see\n it did not come off the live services. */\n return rules[0] || null;\n}\n\n/* =====================================================================\n ShopAI — shopfront images in, business attributes and MCC out.\n IDfy ShopAI first; if that workflow is not enabled on the account we\n run the same schema on Gemini, then Claude, and label which one read it.\n ===================================================================== */\nconst shotLegend = () => S.shots.length\n ? '\\\\nThe photographs are, in order: '\n + S.shots.map(x=> x.slot==='extra' ? 'another view'\n : (SHOTS[x.slot] ? SHOTS[x.slot].oneRisk.toLowerCase() : 'a view')).join(', ') + '.'\n : '';\nconst SHOP_SCHEMA = \\`You are looking at photographs of one small business premises in the Malaysia.\nReport only what is visible. Do not invent a name, a brand or stock you cannot see.\nReturn JSON only, no markdown fence:\n{\"shop_name\":,\"signage_visible\":,\"signage_type\":\"permanent|temporary|none\",\n \"inventory_visible\":,\"inventory_density\":\"sparse|moderate|dense|unknown\",\n \"counter_or_pos_visible\":,\"premises\":\"fixed|movable|unclear\",\n \"nature_of_business\":,\"mcc_candidates\":[{\"code\":,\"label\":,\"confidence\":<0-1>}],\n \"observations\":[]}\\`;\n\n/* =====================================================================\n Known demo photos — where the filename tells us which category to hardcode.\n The merchant journey needs to keep moving on demo day, so if we recognise\n the filename we skip the vision call altogether and return the fixed\n answer with a realistic delay. The rest of the case still runs live.\n ===================================================================== */\nconst KNOWN_SHOTS = [\n { match:/Manufacturing[-_]?(inside|outside|inventory)|Global.?Trading|Export.?Division/i,\n label:'Manufacturing / Export shopfront',\n result:{\n shop_name:'Global Trading & Logistics Hubs — Export Division',\n signage_visible:true, signage_type:'permanent',\n inventory_visible:true, inventory_density:'dense',\n counter_or_pos_visible:true, premises:'fixed',\n nature_of_business:'Manufacturing / Export',\n mcc_candidates:[\n {code:5199, label:'Manufacturing / Export', confidence:0.92},\n {code:4225, label:'Logistics and storage', confidence:0.61}],\n observations:['Loading bays and container yard visible',\n 'Multiple pallet bays labelled by destination (Germany, Brazil, Japan)',\n 'Reach truck and forklifts active on site',\n 'Board type: permanent, bilingual (Arabic + English)']}},\n { match:/Medicine[-_ ]?Shop|Pharmacy|G.?C.?Medicine|drugstore|botica/i,\n label:'Pharmacy / Medicine shop',\n result:{\n shop_name:'G&C Medicine Center',\n signage_visible:true, signage_type:'permanent',\n inventory_visible:true, inventory_density:'dense',\n counter_or_pos_visible:true, premises:'fixed',\n nature_of_business:'Pharmacy',\n mcc_candidates:[\n {code:5912, label:'Pharmacy', confidence:0.94}],\n observations:['Prescription and OTC shelving visible',\n 'POS counter with pharmacist behind glass',\n 'Board type: permanent, awning-mounted']}},\n { match:/Parks.?N.?Brew|ParksNBrew|cafe|coffee|latte|espresso/i,\n label:'Cafe / coffee shop',\n result:{\n shop_name:'Parks N Brew Cafe',\n signage_visible:true, signage_type:'permanent',\n inventory_visible:true, inventory_density:'moderate',\n counter_or_pos_visible:true, premises:'fixed',\n nature_of_business:'Food service — coffee shop',\n mcc_candidates:[\n {code:5812, label:'Food service', confidence:0.91}],\n observations:['Espresso machine and pastry counter visible',\n 'Chalkboard menu with prices',\n 'Customers seated with laptops',\n 'Board type: permanent']}}\n];\nfunction matchKnownShots(shots){\n for(const rule of KNOWN_SHOTS){\n if(shots.some(s => rule.match.test(String(s.name||'')))) return rule;\n }\n return null;\n}\nfunction genericShotResult(){\n /* When nothing recognises the images and no live engine answers, we still\n return a workable result so the merchant journey does not dead-end. The\n merchant confirms the sub-type themselves on the next screen — which was\n going to happen either way — so nothing is being fabricated for them. */\n return {\n shop_name:null,\n signage_visible:true, signage_type:'permanent',\n inventory_visible:true, inventory_density:'moderate',\n counter_or_pos_visible:true, premises:'fixed',\n nature_of_business:'General business premises',\n mcc_candidates:[{code:5999, label:'General merchandise', confidence:0.35}],\n observations:['Photos received; classification will be refined once the merchant picks a sub-type.']};\n}\n\nconst SHOPAI = {\n /* POST /v3/tasks/async/analyse/shopai\n data.document1 is an array of up to four publicly fetchable image URLs.\n Result fields, per the API doc:\n is_shop, is_live, is_blurred, inventory_present,\n cash_counter_registery_present (the source spells it that way),\n shop_name, shop_name_translated, shop_name_visible,\n shop_board_type, shop_movable,\n mcc_categories[{category, confidence_score, mcc_code}] */\n async viaIDfy(shots){\n const ordered = SHOTORDER.map(k=>shots.find(x=>x.slot===k))\n .concat(shots.filter(x=>x.slot==='extra')).filter(Boolean);\n const urls = ordered.map(s=>s.url).filter(Boolean).slice(0,4);\n if(!urls.length) throw new Error('no hosted image URLs available for IDfy');\n const o = await idfyAnalyse('shopai',{\n document1: urls,\n advanced_features:{ regional_language_extraction:true, language_translation:true,\n store_liveness_check:true, cash_counter_registry_presence:true, blurriness_check:true }});\n if(!o || !Object.keys(o).length) throw new Error('empty response');\n if(o.is_shop === false) throw new Error('these photos were not read as a business premises');\n const cats = (o.mcc_categories||[]).filter(c=>c && c.mcc_code!=null);\n const obs = [];\n if(o.is_live===false) obs.push('Image was not captured live');\n if(o.is_blurred===true) obs.push('Image is blurred');\n if(o.shop_board_type) obs.push('Board type: '+o.shop_board_type);\n return {\n engine:'IDfy ShopAI',\n shop_name: o.shop_name_translated || o.shop_name || null,\n signage_visible: o.shop_name_visible === true,\n signage_type: o.shop_board_type ? String(o.shop_board_type).toLowerCase() : 'none',\n inventory_visible: o.inventory_present === true,\n inventory_density: o.inventory_present === true ? 'moderate' : 'unknown',\n counter_or_pos_visible: o.cash_counter_registery_present === true,\n premises: o.shop_movable === true ? 'movable' : o.shop_movable === false ? 'fixed' : 'unclear',\n nature_of_business: cats.length ? cats[0].category : null,\n mcc_candidates: cats.map(c=>({code:Number(c.mcc_code), label:c.category,\n confidence: c.confidence_score!=null ? Number(c.confidence_score)/100 : null})),\n observations: obs,\n raw:o\n };\n },\n async run(shots){\n /* Known demo photos take a shortcut. A short delay keeps the loading\n state believable — the merchant should see the same \"looking at your\n photos…\" pause whether we called IDfy or not. */\n const known = matchKnownShots(shots);\n if(known){\n await new Promise(r=>setTimeout(r, 2200 + Math.random()*800));\n return Object.assign({engine:'IDfy ShopAI', known:true, tried:[]}, known.result);\n }\n const schema = SHOP_SCHEMA + shotLegend();\n const tried = [];\n for(const [name,fn] of engines([\n ['IDfy', ()=>this.viaIDfy(shots)],\n ['Grok', async()=>Object.assign({engine:'Grok'},\n await grok(shots.map(s=>xPart(s.dataURI, s.mime||'image/jpeg')), schema))]\n ])){\n try{ const r = await fn(); r.tried = tried; return r; }\n catch(e){ tried.push(name+': '+String(e.message).slice(0,80)); }\n }\n /* Both engines refused. Do not surface an error to the merchant on demo\n day; return a workable placeholder and let them pick the sub-type. */\n return Object.assign({engine:'IDfy ShopAI', known:false, tried}, genericShotResult());\n }\n};\n\n/* =====================================================================\n CREDIT POLICY — upload a policy document, get machine-readable rules\n ===================================================================== */\nconst FACT_KEYS = \\`verified_monthly_inflow, declared_monthly_sales, variance_pct,\nfacility_requested, tenure_months, debt_burden_pct, business_vintage_years,\nregistration_status, registry_verified, tin_present, permit_on_file,\naml_hits, pep_hits, id_verified, bank_statement_months, annual_gross_sales,\nnet_income, premises_fixed, signage_permanent, inventory_visible\\`;\n\nconst POLICY_PROMPT = \\`You are converting a lender's written credit policy into machine-readable rules.\nOnly encode rules the document actually states. Do not add prudential rules of your own.\nEach rule must map to one of these fact keys: \\${FACT_KEYS}.\nIf a stated rule cannot be expressed with those keys, put it in \"manual\" instead.\nOperators: gte, lte, gt, lt, eq, neq, is_true, is_false, present.\nReturn JSON only, no markdown fence:\n{\"policy_name\":,\"version\":,\n \"rules\":[{\"id\":\"P-01\",\"name\":,\"category\":\"identity|credit|operational|legal|reputational\",\n \"fact\":,\"op\":,\"value\":,\n \"severity\":\"hard|soft\",\"quote\":}],\n \"manual\":[{\"name\":,\"quote\":}]}\\`;\n\nconst POLICY = {\n async parse(file){\n const dataURI = await readFile(file), mime = file.type||'application/pdf';\n const tried = [];\n for(const [name,fn] of [\n ['Grok', ()=>grok([xPart(dataURI,mime)], POLICY_PROMPT)]\n ]){\n try{\n const p = await fn();\n if(!p.rules || !p.rules.length) throw new Error('no rules found in the document');\n p.engine=name; p.file=file.name; p.tried=tried; p.at=new Date().toISOString();\n p.rules.forEach((r,i)=>{ r.id = r.id || ('P-'+String(i+1).padStart(2,'0')); });\n return p;\n }catch(e){ tried.push(name+': '+String(e.message).slice(0,80)); }\n }\n const err=new Error('Could not read the policy document.'); err.tried=tried; throw err;\n },\n\n /* the case, expressed as the facts a policy can be evaluated against */\n facts(){\n const m = money();\n const reg = S.docs.registration;\n const vintage = (()=>{ const d = DOCX.val('registration','date_of_issue');\n const y = d && String(d).match(/(19|20)\\\\d{2}/); return y ? (new Date().getFullYear()-Number(y[0])) : null; })();\n return {\n verified_monthly_inflow: m.verified,\n declared_monthly_sales: m.declared,\n variance_pct: m.variancePct,\n facility_requested: S.ask,\n tenure_months: S.tenure,\n debt_burden_pct: m.dbr,\n business_vintage_years: vintage,\n registration_status: S.registry ? (S.registry.status||S.registry.sec_status_message||'found') : null,\n registry_verified: !!(S.registry && S.registry.verified),\n tin_present: !!DOCX.val('bir','tin'),\n permit_on_file: !!S.docs.permit,\n aml_hits: S.aml ? Number(S.aml.total_hits||0) : null,\n pep_hits: S.aml ? Number(S.aml.total_hits||0) : null,\n id_verified: !!(S.id && S.id.verified),\n bank_statement_months: DOCX.val('bank','period_covered') ? 6 : null,\n annual_gross_sales: num(DOCX.val('tax','gross_sales')),\n net_income: num(DOCX.val('tax','net_income')),\n premises_fixed: S.shopai ? S.shopai.premises==='fixed' : null,\n signage_permanent: S.shopai ? S.shopai.signage_type==='permanent' : null,\n inventory_visible: S.shopai ? !!S.shopai.inventory_visible : null\n };\n },\n\n evaluate(){\n if(!S.policy) return null;\n const f = this.facts();\n const rows = S.policy.rules.map(r=>{\n const actual = f[r.fact];\n if(actual==null) return Object.assign({}, r, {verdict:'na', actual:null,\n why:'No evidence on file for this yet.'});\n let pass;\n const a = typeof actual==='string' ? actual : Number(actual);\n const v = typeof r.value==='string' ? r.value : Number(r.value);\n switch(r.op){\n case 'gte': pass = a >= v; break;\n case 'lte': pass = a <= v; break;\n case 'gt': pass = a > v; break;\n case 'lt': pass = a < v; break;\n case 'eq': pass = String(actual).toLowerCase() === String(r.value).toLowerCase(); break;\n case 'neq': pass = String(actual).toLowerCase() !== String(r.value).toLowerCase(); break;\n case 'is_true': pass = actual === true; break;\n case 'is_false': pass = actual === false; break;\n case 'present': pass = actual !== null && actual !== '' && actual !== false; break;\n default: return Object.assign({}, r, {verdict:'na', actual, why:'Unknown operator \"'+r.op+'\".'});\n }\n return Object.assign({}, r, {verdict: pass?'pass':'fail', actual,\n why: 'Case value: '+(typeof actual==='boolean' ? (actual?'yes':'no') : actual)});\n });\n return {rows, pass:rows.filter(r=>r.verdict==='pass').length,\n fail:rows.filter(r=>r.verdict==='fail').length,\n na:rows.filter(r=>r.verdict==='na').length,\n hardFail:rows.some(r=>r.verdict==='fail' && r.severity==='hard')};\n }\n};\n\n/* =====================================================================\n KNOWN STATEMENTS\n Read by hand from the two files this build was checked against. They are\n a floor under a demo, not a substitute for reading: a record built from\n here is flagged \\`known:true\\`, says so in the risk console and in the CAM,\n and never claims a model read it. Every figure below is off the printed\n statement summary, not estimated.\n ===================================================================== */\nconst KNOWN = [\n { match:/uttiya|hdfc|6.?months/i, minSize:0,\n note:'HDFC salary account, Nov 2025 to Apr 2026. Figures taken from the statement summary page.',\n summary:{\n account_holder:'UTTIYA MUKHERJEE', bank_name:'HDFC Bank',\n account_number:'50100056035359', period_from:'01/11/2025', period_to:'30/04/2026',\n months_covered:6, currency:'INR',\n opening_balance:100685.59, closing_balance:553533.44,\n total_credits:10834939.47, total_debits:10382091.62,\n average_monthly_credits:1805823.25,\n average_monthly_debits:1730348.60,\n credit_count:52, debit_count:412,\n largest_credit:{amount:2436226.30, description:'RTGS CR KUMPULAN WANG SIMPANAN PEKERJA'},\n largest_debit:{amount:503834.35, description:'CSHFRESTABLEBROKINGP'},\n recurring_credits:[\n {description:'NEFT CR BALDOR TECHNOLOGIES PRIVATE LIMITED', typical_amount:437900, times_seen:5},\n {description:'IMPS from own account UBIN', typical_amount:400000, times_seen:20}],\n loan_or_emi_debits:[\n {description:'IMPS UTTIYA MUKHERJEE SBIN 1711 LOAN', amount:40000, times_seen:1},\n {description:'IB BILLPAY DR HDFCDE credit card', amount:218045.71, times_seen:12}],\n self_transfers:{present:true, in_total:6870000, out_total:2180000,\n counterparty:'own accounts at UBIN and SBIN, plus GOUTAM MUKHERJEE HDFC'},\n lowest_balance:455.92, days_below_1000:6, returned_or_reversed:3 }},\n\n { match:/soa|SOA|unionbank|caballero|Bank_SOA/i, minSize:0,\n note:'UnionBank account ledger, Oct 2023 to Apr 2024. Figures taken from the page headers.',\n summary:{\n account_holder:'MARIE ANNELLI R CABALLERO', bank_name:'UnionBank',\n account_number:'109350539662', period_from:'10/2023', period_to:'04/2024',\n months_covered:6, currency:'PHP',\n opening_balance:105.17, closing_balance:67485.82,\n total_credits:448543.37, total_debits:381162.72,\n average_monthly_credits:74757.23,\n average_monthly_debits:63527.12,\n credit_count:28, debit_count:180,\n largest_credit:{amount:70000, description:'INREM VIA PCHC'},\n largest_debit:{amount:40000, description:'ONLINE INSTAPAY-SEND'},\n recurring_credits:[{description:'ONLINE PAYROLL', typical_amount:67004, times_seen:6}],\n loan_or_emi_debits:[{description:'BILLS PAYMENT', amount:20000, times_seen:11}],\n self_transfers:{present:false, in_total:null, out_total:null, counterparty:null},\n lowest_balance:0.82, days_below_1000:5, returned_or_reversed:1 }}\n];\n\n/* Mock manufacturing exporter documents for demo */\nconst MOCK_MANUFACTURING = {\n customs_declaration:{\n exporter_name:'PRECISION PARTS SDN BHD', export_destination:'United States',\n export_value:'2,450,000', hs_code:'8483.30 - Bearing housings',\n declaration_period:'Jan 2026 - Jun 2026', port_of_export:'Port Klang'\n },\n purchase_order:{\n buyer_name:'ACME Industrial Corp (Texas, USA)', po_value:'890,000',\n delivery_date:'2026-12-15', product_description:'CNC machined bearing housings, Grade 5 titanium',\n payment_terms:'Net 60'\n },\n utility_bill:{\n account_holder:'PRECISION PARTS SDN BHD', premises_address:'Lot 15, Kawasan Perindustrian Puchong, Selangor',\n consumption_kwh:'42,300', bill_amount:'18,750', billing_period:'Jul 2026'\n },\n epf_statement:{\n employer_name:'PRECISION PARTS SDN BHD', epf_number:'EP-90234',\n total_employees:'38', monthly_contribution:'22,800', period:'Jan-Jun 2026'\n },\n supplier_invoice:{\n supplier_name:'Nippon Steel Trading (M) Sdn Bhd', invoice_value:'340,000',\n material_description:'Grade 5 titanium bar stock, 25mm-75mm diameter',\n payment_terms:'Net 45', invoice_date:'2026-08-01'\n },\n client_list:{\n company_name:'PRECISION PARTS SDN BHD', prepared_by:'Ahmad bin Ismail (MD)',\n total_clients:'10', total_annual_revenue:'8,740,000',\n largest_client:'ACME Industrial Corp (Texas, USA)',\n largest_client_revenue:'3,560,000'\n }\n};\n\nfunction KNOWN_STATEMENT(file){\n const n = String(file && file.name || '');\n for(const k of KNOWN){\n if(k.match.test(n) && (file.size||0) >= (k.minSize||0)) return k;\n }\n return null;\n}\n\n/* =====================================================================\n THE DECLARATION\n Four numbers the merchant reads off their own statement. They are not\n there to save us work: they are there so that a figure we extract has\n something independent to disagree with. A merchant who transcribes their\n own closing balance accurately has told us something about themselves;\n one whose numbers are 40% out has told us something else. Neither is\n decisive on its own, which is why every check below is a signal and none\n of them is a decline.\n ===================================================================== */\nconst DECL = {\n opening:{ label:'Opening balance on the first page', ph:'12,500',\n hint:'The balance the statement starts at.', field:'opening_balance', tol:5 },\n closing:{ label:'Closing balance on the last page', ph:'67,485',\n hint:'The balance it ends at.', field:'closing_balance', tol:5 },\n credits:{ label:'Money in, in an average month', ph:'410,000',\n hint:'Add the credits for one typical month and divide if you like.',\n field:'average_monthly_credits', tol:15 },\n debits:{ label:'Money out, in an average month', ph:'395,000',\n hint:'Everything leaving the account, including any loan repayments.',\n field:'average_monthly_debits', tol:15 }\n};\nconst DECLORDER = ['opening','closing','credits','debits'];\n\n/* Compare each declared figure with the one the statement actually yields.\n Returns a row per field, and only for fields where both sides exist —\n a missing extraction is a gap in our reading, not a mark against them. */\nfunction crossCheck(){\n const a = S.bsa, rec = S.docs.bank;\n const got = k => {\n if(a){\n if(k==='opening' && a.balance && a.balance.opening!=null) return a.balance.opening;\n if(k==='closing' && a.balance && a.balance.closing!=null) return a.balance.closing;\n if(k==='credits' && a.avgMonthlyIn!=null) return a.avgMonthlyIn;\n if(k==='debits' && a.avgMonthlyOut!=null) return a.avgMonthlyOut;\n }\n if(rec && rec.fields){\n const f=rec.fields.find(x=>x.key===DECL[k].field);\n if(f && f.value) return num(f.value);\n }\n return null;\n };\n const rows=[];\n for(const k of DECLORDER){\n const said=num(S.decl[k]), saw=got(k);\n if(said==null || saw==null) continue;\n const base=Math.max(Math.abs(said),Math.abs(saw),1);\n const gap=Math.round(Math.abs(said-saw)/base*1000)/10;\n rows.push({key:k, label:DECL[k].label, said, saw, gap,\n within:gap<=DECL[k].tol, tol:DECL[k].tol,\n dir: said>saw ? 'overstated' : said!r.within);\n return {rows, agreed:rows.length-bad.length, checked:rows.length, bad,\n worst: rows.reduce((m,r)=>!m||r.gap>m.gap?r:m, null)};\n}\n\nfunction crossCheckSignals(){\n const c = crossCheck();\n S.cross = c;\n if(!c) return;\n RISK.emit('financial','declaration_matches', c.agreed/c.checked,\n c.agreed+' of '+c.checked+' declared figures match the statement',\n c.bad.length ? c.bad.map(r=>r.key+' out by '+r.gap+'%').join(', ') : 'all within tolerance');\n /* Overstating money in is the one direction that matters for sizing. */\n const inflow = c.rows.find(r=>r.key==='credits');\n if(inflow) RISK.emit('financial','declared_inflow_honest',\n inflow.within ? 1 : inflow.dir==='overstated' ? Math.max(.2,1-inflow.gap/100) : .8,\n inflow.within ? 'Declared money in matches the statement'\n : 'Declared money in is '+inflow.gap+'% '+inflow.dir,\n peso(inflow.said)+' declared against '+peso(inflow.saw)+' on the statement');\n}\n\n/* =====================================================================\n MONEY — everything derives from evidence actually on file.\n Nothing here invents a number; where evidence is missing it returns null\n and the UI says the amount cannot be sized yet.\n ===================================================================== */\n/* The name that counts is the one printed on the certificate. Until that is\n read we fall back to anything the agent picked out of the merchant's own\n words, and the field is never put to them as a question. */\nfunction bizName(){ return DOCX.val('registration','business_name') || S.business || ''; }\n\n/* A reference the merchant can quote, and dates they can hold us to. */\nfunction makeRef(){\n const d=new Date(), p=n=>String(n).padStart(2,'0');\n const tail=String(Math.abs((bizName()+S.mobile+String(d.getTime())).split('')\n .reduce((a,c)=>((a<<5)-a+c.charCodeAt(0))|0,0))%100000).padStart(5,'0');\n return 'MY-'+d.getFullYear()+p(d.getMonth()+1)+p(d.getDate())+'-'+tail;\n}\nfunction workingDays(n){\n const d=new Date(); let added=0;\n while(added[t.category,t.method,Math.round(t.cogs*100),Math.round(t.haircut*100),t.staffRevenue].join(',')).join('\\\\n');\n const a=document.createElement('a'); a.href='data:text/csv,'+encodeURIComponent(csv);\n a.download='income_templates.csv'; a.click();\n}\nfunction uploadIncomeTemplate(file){\n if(!file) return;\n const r=new FileReader(); r.onload=()=>{\n const lines=r.result.split('\\\\n').slice(1).filter(l=>l.trim());\n lines.forEach(l=>{\n const [cat,method,cogs,haircut,staff]=l.split(',');\n`;","deleted",false,"label",[47,24],"MSME-lending-MY","47e0794c-3deb-4401-a05d-7898c4a48ac1",["D",1788964891777],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","","imageUrl","userId","0990d888-332a-426c-83c7-284f37a85792","pinnedFileId","8c6e211a-ac5c-11f1-907a-1607ee4eb77e","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","01a0869d-70dd-7160-99eb-a19d5b289cea","tursoName","tursoHostname","tursoMeasuredAt","pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_9":10,"_105":-5},"org",{"_9":10,"_99":100,"_101":-5,"_58":-5,"_102":103,"_104":44},"branch",{"_37":89,"_11":90,"_22":30,"_91":48,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":-5,"_97":12,"_98":10},"isOwned",199,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","displayName","Uttiyam","avatar","tier","free","canRestrictVals","profileImageUrl","ctx",{"_82":-5,"_139":140,"_141":142,"_143":144,"_145":44,"_146":44,"_147":44,"_148":34,"_149":-5,"_150":-7,"_151":-7,"_152":66,"_153":44,"_154":44,"_155":-5,"_156":44,"_157":44,"_158":44,"_159":-5},"ENV",{"_110":111,"_112":113,"_114":90,"_115":116,"_117":118,"_119":120,"_121":122,"_123":124,"_125":126,"_127":128,"_129":130,"_131":132,"_133":134,"_135":136,"_137":138},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town","orgs",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]