[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_154":155,"_156":157},"routes/_app.x.$handle.$name",{"_95":94,"_32":45,"_103":85,"_27":104,"_13":14,"_105":-5,"_11":106,"_107":108,"_109":110,"_111":112,"_113":114,"_20":-5,"_115":-5,"_116":-5,"_117":25,"_118":119,"_120":25,"_121":-5,"_122":-5,"_123":124},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_43":44,"_32":45,"_38":46,"_47":-5,"_48":25,"_49":50,"_51":50,"_52":53,"_54":-5,"_18":55,"_56":-5,"_57":25,"_58":-5,"_59":-5,"_60":61,"_62":63,"_64":-5,"_65":-5,"_66":-5,"_67":-5,"_68":69,"_70":69,"_71":72,"_73":72,"_74":72,"_75":69,"_76":69,"_77":78,"_79":80,"_81":82,"_83":25},"version",31,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":23,"_24":25,"_26":25,"_27":28,"_29":28},"file",{"_30":31,"_32":28,"_33":34,"_35":37,"_38":39,"_40":14,"_41":42,"_24":25},"previous",{"_30":31,"_32":28,"_33":34,"_35":36},"deleted",false,"deletedHere","path","main.ts","latestPath","fileId","98187ff6-a7e3-11f1-ace8-1607ee4eb77e","name","fileType","interval","content","// process-b2b-orders.ts\n// Val Town interval val — polls orders@sodafountain.uk for genuine B2B/wholesale\n// POs, matches customer + products in Breww, creates the order at book price,\n// and accepts it automatically when confident. Falls back to a Draft + email\n// flag when the customer or product match is ambiguous. See SKILL.md\n// (lifeblood-order-processing) for the full business logic this implements.\n\nimport { blob } from \"https://esm.town/v/std/blob\";\n\n// ---------- Secrets (set these as Val Town secrets on this val) ----------\nconst TENANT_ID = \"2ff21b32-fd4b-4dc0-b43b-8d66bbe9d9fa\";\nconst CLIENT_ID = \"865646e4-96c0-4b95-aa0d-4520883e005f\";\nconst GRAPH_CLIENT_SECRET = Deno.env.get(\"GRAPH_CLIENT_SECRET\")!;\nconst BREWW_API_KEY = Deno.env.get(\"BREWW_API_KEY\")!;\nconst MAILBOX = \"orders@sodafountain.uk\";\nconst NOTIFY_TO = \"chris@sodafountain.uk\";\nconst NOTIFY_FROM = \"lucy@drinklifeblood.com\";\n\nconst BREWW_BASE = \"https://breww.com/api/\";\nconst BREWW_HEADERS = {\n Authorization: `Bearer ${BREWW_API_KEY}`,\n \"User-Agent\":\n \"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 Chrome/124.0.0.0 Safari/537.36\",\n Accept: \"application/json\",\n \"Content-Type\": \"application/json\",\n};\n\n// ---------- Dynamic customer matching ----------\n// No hardcoded account table. Every run pulls the live Breww customer list\n// and matches purely on whether a customer's own name/address text appears\n// in the incoming PO — since a customer's PO will always contain their own\n// business name somewhere (letterhead, delivery address, stock location).\n// This means brand-new customers work automatically the moment they exist\n// as a Breww record; nothing needs to be added here by hand.\n//\n// A learned cache (sender email domain -> customer ID) is kept in a Val\n// Town blob so repeat orders from the same sender skip the fuzzy scan\n// entirely after the first successful match — faster, and immune to PO\n// wording changing between orders.\nconst MIN_MATCH_LENGTH = 6; // ignore customer names/fragments shorter than this to avoid false positives\n\ninterface BrewwCustomer {\n id: number;\n name: string;\n price_book_in_use?: { id: number; name: string } | null;\n billing_address?: { first_line?: string };\n delivery_address?: { first_line?: string };\n entity_type_display?: string;\n primary_email?: string | null;\n contacts?: { primary_email?: string | null }[];\n}\n\n// Price book names that mean \"this is a DTC/Shopify consumer, not a B2B\n// trade account\" — excluded from the matching pool entirely. This is what\n// cuts ~900 total customer records down to the ~100 that are actually\n// wholesale/trade accounts, both for speed and to stop a DTC customer's\n// personal name ever being mistaken for a trade account.\nconst NON_B2B_PRICE_BOOKS = [\"D2C\", \"tester\"];\n\n// Breww customer records that must never be auto-matched, even though they\n// pass every other filter. \"Soda Fountain Ltd\" (Chris's own company) exists\n// as a real B2B-type customer record in Breww, and since Chris's own email\n// signature says \"Soda Fountain Ltd\" on every email he sends — including\n// test orders — the matcher would otherwise pick up his own company name\n// from his own signature and misroute the order to himself.\nconst EXCLUDED_CUSTOMER_IDS = [1958785]; // Soda Fountain Ltd\n\nasync function getAllBrewwCustomers(): Promise {\n let url = BREWW_BASE + \"customers-suppliers/?page_size=200\";\n const all: BrewwCustomer[] = [];\n while (url) {\n const res = await fetch(url, { headers: BREWW_HEADERS });\n const data = await res.json();\n all.push(...(data.results ?? []));\n url = data.next ?? null;\n }\n // Only match against real, B2B trade customers — not leads, suppliers,\n // DTC/Shopify consumers, or explicitly excluded internal records.\n return all.filter(\n (c) =>\n c.entity_type_display === \"Customer\" &&\n c.price_book_in_use?.name &&\n !NON_B2B_PRICE_BOOKS.includes(c.price_book_in_use.name) &&\n !EXCLUDED_CUSTOMER_IDS.includes(c.id),\n );\n}\n\nfunction matchableStrings(c: BrewwCustomer): string[] {\n const strings = [\n c.name,\n c.billing_address?.first_line,\n c.delivery_address?.first_line,\n ]\n .filter((s): s is string => !!s && s.length >= MIN_MATCH_LENGTH);\n // Also match on just the first two words of the customer's name (e.g.\n // \"Forest Farm\" for \"Forest Farm The Organic Dairy\") — real customers\n // very often refer to themselves by a shortened version of their full\n // registered name. If this shorter fragment happens to collide with\n // another customer's, findCustomerInText's ambiguity check catches it\n // and falls back to manual review rather than picking one at random.\n if (c.name) {\n const firstTwoWords = c.name.split(/\\s+/).slice(0, 2).join(\" \");\n if (firstTwoWords.length >= MIN_MATCH_LENGTH && firstTwoWords !== c.name) {\n strings.push(firstTwoWords);\n }\n }\n return strings;\n}\n\n// Matches by the sender's actual email address against a customer's\n// primary_email or any of their logged contacts' emails — far more\n// reliable than scanning free text, since it can't be thrown off by a\n// signature block or casual wording. Checked before the text-based scan.\nfunction findCustomerByEmail(\n customers: BrewwCustomer[],\n senderEmail: string,\n): BrewwCustomer | null {\n const email = senderEmail.toLowerCase();\n if (!email) return null;\n const matches = customers.filter((c) => {\n if (c.primary_email?.toLowerCase() === email) return true;\n return (c.contacts ?? []).some((ct) =>\n ct.primary_email?.toLowerCase() === email\n );\n });\n return matches.length === 1 ? matches[0] : null; // ambiguous or none — fall through\n}\n\n// Finds the customer whose name/address text appears in the PO, preferring\n// the longest (most specific) match. Returns null if zero or multiple\n// distinct customers match, so ambiguous cases fall to manual review\n// instead of guessing.\nfunction findCustomerInText(customers: BrewwCustomer[], text: string) {\n const lowerText = text.toLowerCase();\n let best: { customer: BrewwCustomer; matchLength: number } | null = null;\n const distinctMatches = new Set();\n\n for (const c of customers) {\n for (const s of matchableStrings(c)) {\n if (lowerText.includes(s.toLowerCase())) {\n distinctMatches.add(c.id);\n if (!best || s.length > best.matchLength) {\n best = { customer: c, matchLength: s.length };\n }\n }\n }\n }\n\n if (distinctMatches.size !== 1 || !best) return null; // none, or ambiguous\n return best.customer;\n}\n\n// Flavour keyword -> product name fragment (case-insensitive substring match\n// against Breww's live product list, so this survives product ID changes).\nconst FLAVOUR_KEYWORDS: { keyword: RegExp; nameContains: string }[] = [\n { keyword: /original|\\bgreens?\\b|\\bteals?\\b/i, nameContains: \"The Original\" },\n { keyword: /hopical|tropical|\\bblues?\\b/i, nameContains: \"Hopical Punch\" },\n { keyword: /mango(?:saurus)?|\\boranges?\\b/i, nameContains: \"Mangosaurus\" },\n { keyword: /lemon(?:drop)?|\\byellows?\\b/i, nameContains: \"Lemondrop\" },\n { keyword: /ginger/i, nameContains: \"Ginger Beer\" },\n];\n\n// Ignore these — DTC Shopify order emails, never touch.\nconst IGNORE_SENDER_DOMAINS = [\"drinklifeblood.com\"];\nconst IGNORE_SUBJECT_PATTERNS = [/^\\[Raw Culture\\] Order/i];\n\n// Hard cutoff: never process an email older than this, regardless of read\n// status. This is what should have stopped months-old NorthLink order\n// emails (already handled manually, long since fulfilled and invoiced)\n// from being picked up and treated as new — the switch away from relying\n// on \"unread\" as a filter needed this age boundary as its replacement\n// safety rail, and didn't have one.\nconst MAX_EMAIL_AGE_DAYS = 3;\n\ninterface ReviewItem {\n subject: string;\n reason: string;\n messageId: string;\n}\ninterface ProcessedItem {\n subject: string;\n orderNumber: string;\n orderId: number;\n total: number;\n breww_url: string;\n}\n\nasync function getGraphToken(): Promise {\n const res = await fetch(\n `https://login.microsoftonline.com/${TENANT_ID}/oauth2/v2.0/token`,\n {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({\n grant_type: \"client_credentials\",\n client_id: CLIENT_ID,\n client_secret: GRAPH_CLIENT_SECRET,\n scope: \"https://graph.microsoft.com/.default\",\n }),\n },\n );\n const data = await res.json();\n return data.access_token;\n}\n\nasync function getRecentOrderEmails(token: string) {\n // Deliberately does NOT filter on isRead — relying on read status as the\n // dedup signal is fragile, since simply opening/previewing an email in\n // Outlook marks it read and would hide it from the val before it ever\n // got a chance to process it. Instead we pull recent mail regardless of\n // read state, and rely entirely on our own processedIds/pending blobs\n // (tracked by message ID) to know what's already been handled.\n const url =\n `https://graph.microsoft.com/v1.0/users/${MAILBOX}/mailFolders/inbox/messages` +\n `?$top=50&$orderby=receivedDateTime desc` +\n `&$select=id,subject,from,receivedDateTime,hasAttachments,body`;\n const res = await fetch(url, {\n headers: { Authorization: `Bearer ${token}` },\n });\n const data = await res.json();\n return data.value ?? [];\n}\n\nfunction stripHtml(html: string): string {\n return html\n .replace(//gi, \"\")\n .replace(/<[^>]+>/g, \" \")\n .replace(/ /gi, \" \")\n .replace(/\\s+/g, \" \")\n .trim();\n}\n\n// Returns the best available text to parse an order from: a PDF attachment\n// if there is one, otherwise the plain email body itself. Real B2B\n// customers sometimes just type an order straight into the email rather\n// than attaching a formal PO, so both are treated as valid order sources.\nasync function getOrderText(token: string, email: any): Promise {\n if (email.hasAttachments) {\n const url =\n `https://graph.microsoft.com/v1.0/users/${MAILBOX}/messages/${email.id}/attachments`;\n const res = await fetch(url, {\n headers: { Authorization: `Bearer ${token}` },\n });\n const data = await res.json();\n const pdfAttachment = (data.value ?? []).find((a: any) =>\n a.name?.toLowerCase().endsWith(\".pdf\")\n );\n if (pdfAttachment?.contentBytes) {\n // unpdf (not pdf-parse) — pdf-parse depends on pdf.js's canvas\n // renderer, which needs browser-only globals (DOMMatrix) that don't\n // exist in Val Town's Deno runtime and crashes the whole function on\n // import. unpdf is built for exactly this kind of edge/serverless\n // environment and does text-only extraction with no canvas/native\n // bindings at all.\n const { extractText, getDocumentProxy } = await import(\"npm:unpdf\");\n const buffer = Uint8Array.from(\n atob(pdfAttachment.contentBytes),\n (c) => c.charCodeAt(0),\n );\n const pdf = await getDocumentProxy(buffer);\n const { text } = await extractText(pdf, { mergePages: true });\n if (text?.trim()) return text;\n // fall through to body text if the PDF had no extractable text\n }\n }\n const body = email.body?.content;\n if (!body) return null;\n return email.body.contentType === \"html\" ? stripHtml(body) : body;\n}\n\n// Purely cosmetic now — marks the email read in Outlook so it's visually\n// obvious which orders have been handled when Chris looks at the inbox.\n// No longer used for dedup logic (see getRecentOrderEmails for why).\nasync function markRead(token: string, messageId: string) {\n await fetch(\n `https://graph.microsoft.com/v1.0/users/${MAILBOX}/messages/${messageId}`,\n {\n method: \"PATCH\",\n headers: {\n Authorization: `Bearer ${token}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({ isRead: true }),\n },\n );\n}\n\nasync function getBrewwProducts() {\n const res = await fetch(BREWW_BASE + \"products/?page_size=200\", {\n headers: BREWW_HEADERS,\n });\n const data = await res.json();\n return data.results ?? data;\n}\n\n// Line-item extraction handles two styles of wording:\n// - Structured PO tables (e.g. \"2.00 Case 12 21.50 ... Original\")\n// - Plain English typed straight into an email (\"1 case of the original,\n// 34 cases of hopical, 2 cases of lemon\")\n// Deliberately conservative — if it can't parse a line confidently, it\n// should surface a review item rather than guess.\nfunction extractLines(text: string) {\n const lines: { flavour: string; quantity: number }[] = [];\n for (const { keyword, nameContains } of FLAVOUR_KEYWORDS) {\n const patterns = [\n // Structured PO table: flavour name appears FIRST, then quantity —\n // e.g. \"Raw Original 330ml Raw Original 330ml 2.00 Case 12 21.50\".\n // Requires a case-size digit right after \"Case\" (real tables always\n // have one; casual English never does), so this can't collide with\n // the plain-English pattern below.\n new RegExp(\n `(?:${keyword.source})[\\s\\S]{0,60}?(\\d+(?:\\.\\d+)?)\\s*Case\\s+\\d+`,\n \"i\",\n ),\n // Plain English typed into an email — increasingly casual, e.g.\n // \"1 case of the original\", \"34 cases of hopical\", or even just\n // \"2 mango\", \"1 lemon\" with no \"case\" word at all. \"case(s) of the\"\n // are all optional here; only the number and keyword are required.\n new RegExp(\n `(\\d+(?:\\.\\d+)?)\\s*(?:cases?\\s+)?(?:of\\s+)?(?:the\\s+)?(?:${keyword.source})`,\n \"i\",\n ),\n ];\n for (const pattern of patterns) {\n const lineMatch = text.match(pattern);\n if (lineMatch) {\n lines.push({\n flavour: nameContains,\n quantity: Math.round(parseFloat(lineMatch[1])),\n });\n break; // don't double-count the same flavour against both patterns\n }\n }\n }\n return lines;\n}\n\nasync function createOrder(\n customerId: number,\n priceBook: number,\n poNumber: string,\n orderLines: { product: number; quantity: number }[],\n) {\n // Duplicate guard: if an order already exists for this customer with this\n // exact PO number (e.g. it was already created manually, or by an\n // earlier run whose email dedup we can no longer fully trust), don't\n // create a second one.\n const existingRes = await fetch(\n `${BREWW_BASE}orders/?customer=${customerId}&po_number=${\n encodeURIComponent(poNumber)\n }`,\n { headers: BREWW_HEADERS },\n );\n const existingData = await existingRes.json();\n const existing = (existingData.results ?? existingData)?.[0];\n if (existing?.id) {\n return { ...existing, _duplicate: true };\n }\n\n const payload = {\n customer: customerId,\n price_book: priceBook,\n issue_date: new Date().toISOString().slice(0, 10),\n po_number: poNumber,\n order_status: 1, // draft first, promote after sanity check\n fulfillment_required: true,\n // Critical: invoices must only go out at the point of delivery, per\n // Chris's normal process — never at order creation/acceptance. Without\n // this explicit false, Breww defaults to true and emails the customer\n // an invoice the moment the order is created, regardless of whether\n // anything has actually been fulfilled yet.\n allow_automatic_emailing_to_customer: false,\n order_lines: orderLines,\n };\n const res = await fetch(BREWW_BASE + \"orders/\", {\n method: \"POST\",\n headers: BREWW_HEADERS,\n body: JSON.stringify(payload),\n });\n return res.json();\n}\n\nasync function sendCustomerConfirmation(\n token: string,\n email: any,\n orderNumber: string,\n poNumber: string | null,\n) {\n const poLine = poNumber ? ` Your reference: ${poNumber}.` : \"\";\n const comment =\n `Thanks for your order.${poLine} This has been logged as order ${orderNumber} ` +\n `and will be added to our next processing run.\\n\\n` +\n `If anything above looks off, just reply to this email and we'll sort it out.`;\n\n await fetch(\n `https://graph.microsoft.com/v1.0/users/${MAILBOX}/messages/${email.id}/reply`,\n {\n method: \"POST\",\n headers: {\n Authorization: `Bearer ${token}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({ comment }),\n },\n );\n}\n\nasync function acceptOrder(orderId: number) {\n await fetch(`${BREWW_BASE}orders/${orderId}/`, {\n method: \"PATCH\",\n headers: BREWW_HEADERS,\n body: JSON.stringify({ order_status: 3 }),\n });\n}\n\nasync function sendSummaryEmail(\n token: string,\n processed: ProcessedItem[],\n review: ReviewItem[],\n) {\n if (processed.length === 0 && review.length === 0) return; // nothing to report\n\n const rows = (arr: string[]) => arr.map((r) => `

${r}

`).join(\"\");\n const processedRows = rows(\n processed.map(\n (p) =>\n `${p.subject} — Order ${p.orderNumber}, £${\n p.total.toFixed(2)\n } — ` +\n `view in Breww`,\n ),\n );\n const reviewRows = rows(\n review.map((r) => `${r.subject} — needs manual review: ${r.reason}`),\n );\n\n const html = `\n

B2B order processing run — ${new Date().toISOString()}

\n ${\n processed.length\n ? `

Auto-accepted (${processed.length})

${processedRows}`\n : \"\"\n }\n ${\n review.length\n ? `

Needs manual review (${review.length})

${reviewRows}`\n : \"\"\n }\n `;\n\n await fetch(\n `https://graph.microsoft.com/v1.0/users/${NOTIFY_FROM}/sendMail`,\n {\n method: \"POST\",\n headers: {\n Authorization: `Bearer ${token}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n message: {\n subject:\n `B2B orders processed: ${processed.length} accepted, ${review.length} need review`,\n body: { contentType: \"HTML\", content: html },\n toRecipients: [{ emailAddress: { address: NOTIFY_TO } }],\n },\n saveToSentItems: true,\n }),\n },\n );\n}\n\ninterface PendingItem {\n subject: string;\n reason: string;\n firstSeenAt: string; // ISO date — when we first couldn't resolve this\n lastNotifiedAt: string; // ISO date — throttles repeat notifications\n}\n\nconst RETRY_WINDOW_HOURS = 24; // how long to keep silently retrying before giving up\nconst RENOTIFY_HOURS = 4; // how often to re-include a still-unresolved item in the summary email\n\nfunction hoursSince(iso: string): number {\n return (Date.now() - new Date(iso).getTime()) / (1000 * 60 * 60);\n}\n\nasync function processOrders() {\n const token = await getGraphToken();\n // processedIds = emails that resulted in an actual Breww order (accepted\n // or deliberately left as draft) or were confirmed noise — these are\n // truly done and marked read, never revisited.\n const processedIds: string[] =\n (await blob.getJSON(\"lifeblood_orders_processed\")) ?? [];\n // pending = emails that look like an order but couldn't be resolved yet\n // (unmatched customer, missing price book, unparseable lines...). Left\n // UNREAD deliberately, so every run retries them automatically — if you\n // fix the underlying Breww record, the very next run picks it up without\n // you needing to do anything else. Only stops retrying after\n // RETRY_WINDOW_HOURS, so a genuinely bad email doesn't retry forever.\n const pending: Record =\n (await blob.getJSON(\"lifeblood_orders_pending\")) ?? {};\n // Learned map of sender email domain -> Breww customer ID, built up\n // automatically as orders are matched. Makes repeat senders instant and\n // resilient to their PO wording changing between orders.\n const domainCache: Record =\n (await blob.getJSON(\"lifeblood_customer_domain_cache\")) ?? {};\n const emails = await getRecentOrderEmails(token);\n const allCustomers = await getAllBrewwCustomers();\n const customersById = new Map(allCustomers.map((c) => [c.id, c]));\n const products = await getBrewwProducts();\n\n const processed: ProcessedItem[] = [];\n const review: ReviewItem[] = []; // only items to actually include in this run's email\n\n // Records an unresolved attempt: on first sighting, or once the retry\n // window has expired, or on cadence via RENOTIFY_HOURS, it's included in\n // this run's email; otherwise it retries silently in the background.\n async function markUnresolved(email: any, reason: string) {\n const existing = pending[email.id];\n const now = new Date().toISOString();\n\n if (!existing) {\n pending[email.id] = {\n subject: email.subject,\n reason,\n firstSeenAt: now,\n lastNotifiedAt: now,\n };\n review.push({\n subject: email.subject,\n reason: `${reason} (will keep retrying automatically)`,\n messageId: email.id,\n });\n return;\n }\n\n existing.reason = reason; // keep the latest reason, in case it changed\n if (hoursSince(existing.firstSeenAt) >= RETRY_WINDOW_HOURS) {\n // Given up — stop retrying, mark done, flag for a human definitively.\n review.push({\n subject: email.subject,\n reason:\n `${reason} — unresolved after ${RETRY_WINDOW_HOURS}h of retrying, needs manual handling in Breww`,\n messageId: email.id,\n });\n delete pending[email.id];\n await markRead(token, email.id);\n processedIds.push(email.id);\n return;\n }\n\n if (hoursSince(existing.lastNotifiedAt) >= RENOTIFY_HOURS) {\n existing.lastNotifiedAt = now;\n review.push({\n subject: email.subject,\n reason: `${reason} (still retrying)`,\n messageId: email.id,\n });\n }\n // else: still within the retry window and recently notified — retry\n // silently, no email noise this run.\n }\n\n for (const email of emails) {\n if (processedIds.includes(email.id)) continue;\n\n try {\n const receivedAgeDays =\n (Date.now() - new Date(email.receivedDateTime).getTime()) /\n (1000 * 60 * 60 * 24);\n if (receivedAgeDays > MAX_EMAIL_AGE_DAYS) {\n // Too old to safely auto-process — almost certainly already\n // handled manually before this automation existed. Mark done so\n // it's never reconsidered, but never act on it.\n delete pending[email.id];\n await markRead(token, email.id);\n processedIds.push(email.id);\n continue;\n }\n\n const senderDomain =\n email.from?.emailAddress?.address?.split(\"@\")[1]?.toLowerCase() ?? \"\";\n if (IGNORE_SENDER_DOMAINS.includes(senderDomain)) continue;\n if (IGNORE_SUBJECT_PATTERNS.some((p) => p.test(email.subject))) continue;\n\n const orderText = await getOrderText(token, email);\n if (!orderText) continue; // nothing to read at all — skip silently, not worth flagging\n\n // Fast path: we've matched this sender domain to a customer before.\n // Guards against a previously-poisoned cache entry (e.g. an excluded\n // ID like Soda Fountain Ltd got cached before that exclusion\n // existed) by ignoring any cached match that's since been excluded.\n let account: BrewwCustomer | null = null;\n const senderEmail = email.from?.emailAddress?.address ?? \"\";\n\n // Most reliable signal first: does this exact sender email match a\n // customer's own recorded email address in Breww? This can't be\n // thrown off by casual wording or a signature block the way scanning\n // free text can.\n account = findCustomerByEmail(allCustomers, senderEmail);\n\n if (\n !account && domainCache[senderDomain] &&\n !EXCLUDED_CUSTOMER_IDS.includes(domainCache[senderDomain])\n ) {\n account = customersById.get(domainCache[senderDomain]) ?? null;\n }\n // Otherwise, fuzzy-match against the live customer list. This is what\n // makes brand-new customers work with zero setup — no table to update.\n if (!account) {\n account = findCustomerInText(allCustomers, orderText);\n if (account && senderDomain) {\n domainCache[senderDomain] = account.id; // learn it for next time\n }\n }\n\n const lines = extractLines(orderText);\n\n // No customer AND no order lines found at all — this is almost\n // certainly not an order (a newsletter, a marketing email, general\n // correspondence). Skip silently and permanently, rather than\n // retrying inbox noise forever.\n if (!account && lines.length === 0) {\n delete pending[email.id];\n await markRead(token, email.id);\n processedIds.push(email.id);\n continue;\n }\n\n if (!account) {\n await markUnresolved(\n email,\n \"Found what looks like order quantities, but could not confidently match a customer — \" +\n \"either they're not in Breww yet, or their name matched more than one record.\",\n );\n continue;\n }\n if (!account.price_book_in_use?.id) {\n await markUnresolved(\n email,\n `Matched customer \"${account.name}\" but they have no price book set in Breww`,\n );\n continue;\n }\n if (lines.length === 0) {\n const preview = orderText.replace(/\\s+/g, \" \").trim().slice(0, 150);\n await markUnresolved(\n email,\n `Matched customer \"${account.name}\" but could not parse any order lines. ` +\n `Text seen: \"${preview}\"`,\n );\n continue;\n }\n\n const orderLines: { product: number; quantity: number }[] = [];\n let allMatched = true;\n for (const line of lines) {\n const product = products.find((p: any) =>\n p.name?.toLowerCase().includes(line.flavour.toLowerCase()) &&\n p.name?.toLowerCase().includes(\"12 x 330ml\")\n );\n if (!product) {\n allMatched = false;\n break;\n }\n orderLines.push({ product: product.id, quantity: line.quantity });\n }\n\n if (!allMatched) {\n await markUnresolved(\n email,\n \"One or more product lines could not be matched confidently\",\n );\n continue;\n }\n\n const poNumberMatch = orderText.match(/Order number\\s*\\n?\\s*([\\w-]+)/i);\n const realPoNumber = poNumberMatch?.[1] ?? null; // only set when genuinely found in the order text\n // Breww still needs *some* po_number value for the duplicate-guard to\n // key on, even for informal orders with no real PO reference — falls\n // back to a stable per-email identifier, never shown to the customer.\n const poNumberForBreww = realPoNumber ?? `email-${email.id}`;\n\n const order = await createOrder(\n account.id,\n account.price_book_in_use!.id,\n poNumberForBreww,\n orderLines,\n );\n\n if (!order?.id) {\n await markUnresolved(email, \"Breww order creation failed\");\n continue;\n }\n\n if (order._duplicate) {\n // Already exists in Breww (likely created manually before this val\n // existed, or before this dedup safeguard was added) — don't touch\n // it, just mark this email resolved so we stop re-checking it.\n delete pending[email.id];\n await markRead(token, email.id);\n processedIds.push(email.id);\n continue;\n }\n\n // From here on, an order genuinely exists in Breww — this email is\n // resolved either way (accepted or deliberately left draft), so it's\n // safe to mark done and never retry, whatever happens next.\n delete pending[email.id];\n\n // Sanity checks (apply regardless of test/real):\n // - value must be a real positive number\n // - every line we submitted must actually have been created (guards\n // against a partial order — e.g. only 2 of 4 lines saved — still\n // returning what looks like a valid positive total)\n // Separately: \"test\" in the subject means never auto-accept in Breww\n // (stays Draft), but the confirmation reply still sends either way —\n // that's the point, it's how the pipeline gets verified end to end\n // without a fake order landing as Accepted.\n const isTestOrder = /test/i.test(email.subject);\n const linesCreated = Array.isArray(order.order_lines)\n ? order.order_lines.length\n : orderLines.length;\n\n if (!order.value || order.value <= 0) {\n review.push({\n subject: email.subject,\n reason:\n `Order ${order.number} created but value looks wrong (£${order.value}) — left as Draft`,\n messageId: email.id,\n });\n } else if (linesCreated !== orderLines.length) {\n review.push({\n subject: email.subject,\n reason:\n `Order ${order.number} created but only ${linesCreated} of ${orderLines.length} lines were saved — left as Draft`,\n messageId: email.id,\n });\n } else if (isTestOrder) {\n review.push({\n subject: email.subject,\n reason:\n `Order ${order.number} matched cleanly (£${order.value}) but subject contains \"test\" — left as Draft, not auto-accepted`,\n messageId: email.id,\n });\n await sendCustomerConfirmation(\n token,\n email,\n order.number,\n realPoNumber,\n );\n } else {\n await acceptOrder(order.id);\n await sendCustomerConfirmation(\n token,\n email,\n order.number,\n realPoNumber,\n );\n processed.push({\n subject: email.subject,\n orderNumber: order.number,\n orderId: order.id,\n total: order.value,\n breww_url: `https://app.breww.com/orders/${order.id}/`,\n });\n }\n\n await markRead(token, email.id);\n processedIds.push(email.id);\n } catch (err) {\n // One bad email (malformed PDF, unexpected data, etc.) should never\n // take down the whole run — log it, flag it for review, and move on\n // to the next email rather than crashing the entire function.\n console.error(\n `Error processing email ${email.id} (${email.subject}):`,\n err,\n );\n review.push({\n subject: email.subject,\n reason: `Unexpected error while processing: ${String(err)}`,\n messageId: email.id,\n });\n }\n }\n\n await blob.setJSON(\"lifeblood_orders_processed\", processedIds.slice(-500)); // keep log bounded\n await blob.setJSON(\"lifeblood_orders_pending\", pending);\n await blob.setJSON(\"lifeblood_customer_domain_cache\", domainCache);\n await sendSummaryEmail(token, processed, review);\n\n return {\n processed: processed.length,\n review: review.length,\n stillPending: Object.keys(pending).length,\n };\n}\n\n// Cron val entry point. Val Town cron triggers call this with an Interval\n// object ({ lastRunAt: Date | undefined }) on the schedule set in the val's\n// Cron trigger settings, and ignore whatever is returned.\nexport default async function (interval: { lastRunAt?: Date }) {\n return await processOrders();\n}","// process-b2b-orders.ts\n// Val Town interval val — polls orders@sodafountain.uk for genuine B2B/wholesale\n// POs, matches customer + products in Breww, creates the order at book price,\n// and accepts it automatically when confident. Falls back to a Draft + email\n// flag when the customer or product match is ambiguous. See SKILL.md\n// (lifeblood-order-processing) for the full business logic this implements.\n\nimport { blob } from \"https://esm.town/v/std/blob\";\n\n// ---------- Secrets (set these as Val Town secrets on this val) ----------\nconst TENANT_ID = \"2ff21b32-fd4b-4dc0-b43b-8d66bbe9d9fa\";\nconst CLIENT_ID = \"865646e4-96c0-4b95-aa0d-4520883e005f\";\nconst GRAPH_CLIENT_SECRET = Deno.env.get(\"GRAPH_CLIENT_SECRET\")!;\nconst BREWW_API_KEY = Deno.env.get(\"BREWW_API_KEY\")!;\nconst MAILBOX = \"orders@sodafountain.uk\";\nconst NOTIFY_TO = \"chris@sodafountain.uk\";\nconst NOTIFY_FROM = \"lucy@drinklifeblood.com\";\n\nconst BREWW_BASE = \"https://breww.com/api/\";\nconst BREWW_HEADERS = {\n Authorization: `Bearer ${BREWW_API_KEY}`,\n \"User-Agent\":\n \"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 Chrome/124.0.0.0 Safari/537.36\",\n Accept: \"application/json\",\n \"Content-Type\": \"application/json\",\n};\n\n// ---------- Dynamic customer matching ----------\n// No hardcoded account table. Every run pulls the live Breww customer list\n// and matches purely on whether a customer's own name/address text appears\n// in the incoming PO — since a customer's PO will always contain their own\n// business name somewhere (letterhead, delivery address, stock location).\n// This means brand-new customers work automatically the moment they exist\n// as a Breww record; nothing needs to be added here by hand.\n//\n// A learned cache (sender email domain -> customer ID) is kept in a Val\n// Town blob so repeat orders from the same sender skip the fuzzy scan\n// entirely after the first successful match — faster, and immune to PO\n// wording changing between orders.\nconst MIN_MATCH_LENGTH = 6; // ignore customer names/fragments shorter than this to avoid false positives\n\ninterface BrewwCustomer {\n id: number;\n name: string;\n price_book_in_use?: { id: number; name: string } | null;\n billing_address?: { first_line?: string };\n delivery_address?: { first_line?: string };\n entity_type_display?: string;\n primary_email?: string | null;\n contacts?: { primary_email?: string | null }[];\n}\n\n// Price book names that mean \"this is a DTC/Shopify consumer, not a B2B\n// trade account\" — excluded from the matching pool entirely. This is what\n// cuts ~900 total customer records down to the ~100 that are actually\n// wholesale/trade accounts, both for speed and to stop a DTC customer's\n// personal name ever being mistaken for a trade account.\nconst NON_B2B_PRICE_BOOKS = [\"D2C\", \"tester\"];\n\n// Breww customer records that must never be auto-matched, even though they\n// pass every other filter. \"Soda Fountain Ltd\" (Chris's own company) exists\n// as a real B2B-type customer record in Breww, and since Chris's own email\n// signature says \"Soda Fountain Ltd\" on every email he sends — including\n// test orders — the matcher would otherwise pick up his own company name\n// from his own signature and misroute the order to himself.\nconst EXCLUDED_CUSTOMER_IDS = [1958785]; // Soda Fountain Ltd\n\nasync function getAllBrewwCustomers(): Promise {\n let url = BREWW_BASE + \"customers-suppliers/?page_size=200\";\n const all: BrewwCustomer[] = [];\n while (url) {\n const res = await fetch(url, { headers: BREWW_HEADERS });\n const data = await res.json();\n all.push(...(data.results ?? []));\n url = data.next ?? null;\n }\n // Only match against real, B2B trade customers — not leads, suppliers,\n // DTC/Shopify consumers, or explicitly excluded internal records.\n return all.filter(\n (c) =>\n c.entity_type_display === \"Customer\" &&\n c.price_book_in_use?.name &&\n !NON_B2B_PRICE_BOOKS.includes(c.price_book_in_use.name) &&\n !EXCLUDED_CUSTOMER_IDS.includes(c.id),\n );\n}\n\nfunction matchableStrings(c: BrewwCustomer): string[] {\n const strings = [\n c.name,\n c.billing_address?.first_line,\n c.delivery_address?.first_line,\n ]\n .filter((s): s is string => !!s && s.length >= MIN_MATCH_LENGTH);\n // Also match on just the first two words of the customer's name (e.g.\n // \"Forest Farm\" for \"Forest Farm The Organic Dairy\") — real customers\n // very often refer to themselves by a shortened version of their full\n // registered name. If this shorter fragment happens to collide with\n // another customer's, findCustomerInText's ambiguity check catches it\n // and falls back to manual review rather than picking one at random.\n if (c.name) {\n const firstTwoWords = c.name.split(/\\s+/).slice(0, 2).join(\" \");\n if (firstTwoWords.length >= MIN_MATCH_LENGTH && firstTwoWords !== c.name) {\n strings.push(firstTwoWords);\n }\n }\n return strings;\n}\n\n// Matches by the sender's actual email address against a customer's\n// primary_email or any of their logged contacts' emails — far more\n// reliable than scanning free text, since it can't be thrown off by a\n// signature block or casual wording. Checked before the text-based scan.\nfunction findCustomerByEmail(\n customers: BrewwCustomer[],\n senderEmail: string,\n): BrewwCustomer | null {\n const email = senderEmail.toLowerCase();\n if (!email) return null;\n const matches = customers.filter((c) => {\n if (c.primary_email?.toLowerCase() === email) return true;\n return (c.contacts ?? []).some((ct) =>\n ct.primary_email?.toLowerCase() === email\n );\n });\n return matches.length === 1 ? matches[0] : null; // ambiguous or none — fall through\n}\n\n// Finds the customer whose name/address text appears in the PO, preferring\n// the longest (most specific) match. Returns null if zero or multiple\n// distinct customers match, so ambiguous cases fall to manual review\n// instead of guessing.\nfunction findCustomerInText(customers: BrewwCustomer[], text: string) {\n const lowerText = text.toLowerCase();\n let best: { customer: BrewwCustomer; matchLength: number } | null = null;\n const distinctMatches = new Set();\n\n for (const c of customers) {\n for (const s of matchableStrings(c)) {\n if (lowerText.includes(s.toLowerCase())) {\n distinctMatches.add(c.id);\n if (!best || s.length > best.matchLength) {\n best = { customer: c, matchLength: s.length };\n }\n }\n }\n }\n\n if (distinctMatches.size !== 1 || !best) return null; // none, or ambiguous\n return best.customer;\n}\n\n// Flavour keyword -> product name fragment (case-insensitive substring match\n// against Breww's live product list, so this survives product ID changes).\nconst FLAVOUR_KEYWORDS: { keyword: RegExp; nameContains: string }[] = [\n { keyword: /original|\\bgreens?\\b|\\bteals?\\b/i, nameContains: \"The Original\" },\n { keyword: /hopical|tropical|\\bblues?\\b/i, nameContains: \"Hopical Punch\" },\n { keyword: /mango(?:saurus)?|\\boranges?\\b/i, nameContains: \"Mangosaurus\" },\n { keyword: /lemon(?:drop)?|\\byellows?\\b/i, nameContains: \"Lemondrop\" },\n { keyword: /ginger/i, nameContains: \"Ginger Beer\" },\n];\n\n// Ignore these — DTC Shopify order emails, never touch.\nconst IGNORE_SENDER_DOMAINS = [\"drinklifeblood.com\"];\nconst IGNORE_SUBJECT_PATTERNS = [/^\\[Raw Culture\\] Order/i];\n\n// Hard cutoff: never process an email older than this, regardless of read\n// status. This is what should have stopped months-old NorthLink order\n// emails (already handled manually, long since fulfilled and invoiced)\n// from being picked up and treated as new — the switch away from relying\n// on \"unread\" as a filter needed this age boundary as its replacement\n// safety rail, and didn't have one.\nconst MAX_EMAIL_AGE_DAYS = 3;\n\ninterface ReviewItem {\n subject: string;\n reason: string;\n messageId: string;\n}\ninterface ProcessedItem {\n subject: string;\n orderNumber: string;\n orderId: number;\n total: number;\n breww_url: string;\n}\n\nasync function getGraphToken(): Promise {\n const res = await fetch(\n `https://login.microsoftonline.com/${TENANT_ID}/oauth2/v2.0/token`,\n {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({\n grant_type: \"client_credentials\",\n client_id: CLIENT_ID,\n client_secret: GRAPH_CLIENT_SECRET,\n scope: \"https://graph.microsoft.com/.default\",\n }),\n },\n );\n const data = await res.json();\n return data.access_token;\n}\n\nasync function getRecentOrderEmails(token: string) {\n // Deliberately does NOT filter on isRead — relying on read status as the\n // dedup signal is fragile, since simply opening/previewing an email in\n // Outlook marks it read and would hide it from the val before it ever\n // got a chance to process it. Instead we pull recent mail regardless of\n // read state, and rely entirely on our own processedIds/pending blobs\n // (tracked by message ID) to know what's already been handled.\n const url =\n `https://graph.microsoft.com/v1.0/users/${MAILBOX}/mailFolders/inbox/messages` +\n `?$top=50&$orderby=receivedDateTime desc` +\n `&$select=id,subject,from,receivedDateTime,hasAttachments,body`;\n const res = await fetch(url, {\n headers: { Authorization: `Bearer ${token}` },\n });\n const data = await res.json();\n return data.value ?? [];\n}\n\nfunction stripHtml(html: string): string {\n return html\n .replace(//gi, \"\")\n .replace(/<[^>]+>/g, \" \")\n .replace(/ /gi, \" \")\n .replace(/\\s+/g, \" \")\n .trim();\n}\n\n// Returns the best available text to parse an order from: a PDF attachment\n// if there is one, otherwise the plain email body itself. Real B2B\n// customers sometimes just type an order straight into the email rather\n// than attaching a formal PO, so both are treated as valid order sources.\nasync function getOrderText(token: string, email: any): Promise {\n if (email.hasAttachments) {\n const url =\n `https://graph.microsoft.com/v1.0/users/${MAILBOX}/messages/${email.id}/attachments`;\n const res = await fetch(url, {\n headers: { Authorization: `Bearer ${token}` },\n });\n const data = await res.json();\n const pdfAttachment = (data.value ?? []).find((a: any) =>\n a.name?.toLowerCase().endsWith(\".pdf\")\n );\n if (pdfAttachment?.contentBytes) {\n // unpdf (not pdf-parse) — pdf-parse depends on pdf.js's canvas\n // renderer, which needs browser-only globals (DOMMatrix) that don't\n // exist in Val Town's Deno runtime and crashes the whole function on\n // import. unpdf is built for exactly this kind of edge/serverless\n // environment and does text-only extraction with no canvas/native\n // bindings at all.\n const { extractText, getDocumentProxy } = await import(\"npm:unpdf\");\n const buffer = Uint8Array.from(\n atob(pdfAttachment.contentBytes),\n (c) => c.charCodeAt(0),\n );\n const pdf = await getDocumentProxy(buffer);\n const { text } = await extractText(pdf, { mergePages: true });\n if (text?.trim()) return text;\n // fall through to body text if the PDF had no extractable text\n }\n }\n const body = email.body?.content;\n if (!body) return null;\n return email.body.contentType === \"html\" ? stripHtml(body) : body;\n}\n\n// Purely cosmetic now — marks the email read in Outlook so it's visually\n// obvious which orders have been handled when Chris looks at the inbox.\n// No longer used for dedup logic (see getRecentOrderEmails for why).\nasync function markRead(token: string, messageId: string) {\n await fetch(\n `https://graph.microsoft.com/v1.0/users/${MAILBOX}/messages/${messageId}`,\n {\n method: \"PATCH\",\n headers: {\n Authorization: `Bearer ${token}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({ isRead: true }),\n },\n );\n}\n\nasync function getBrewwProducts() {\n const res = await fetch(BREWW_BASE + \"products/?page_size=200\", {\n headers: BREWW_HEADERS,\n });\n const data = await res.json();\n return data.results ?? data;\n}\n\n// Line-item extraction handles two styles of wording:\n// - Structured PO tables (e.g. \"2.00 Case 12 21.50 ... Original\")\n// - Plain English typed straight into an email (\"1 case of the original,\n// 34 cases of hopical, 2 cases of lemon\")\n// Deliberately conservative — if it can't parse a line confidently, it\n// should surface a review item rather than guess.\nfunction extractLines(text: string) {\n const lines: { flavour: string; quantity: number }[] = [];\n for (const { keyword, nameContains } of FLAVOUR_KEYWORDS) {\n const patterns = [\n // Structured PO table: flavour name appears FIRST, then quantity —\n // e.g. \"Raw Original 330ml Raw Original 330ml 2.00 Case 12 21.50\".\n // Requires a case-size digit right after \"Case\" (real tables always\n // have one; casual English never does), so this can't collide with\n // the plain-English pattern below.\n new RegExp(\n `(?:${keyword.source})[\\s\\S]{0,60}?(\\d+(?:\\.\\d+)?)\\s*Case\\s+\\d+`,\n \"i\",\n ),\n // Plain English typed into an email — increasingly casual, e.g.\n // \"1 case of the original\", \"34 cases of hopical\", or even just\n // \"2 mango\", \"1 lemon\" with no \"case\" word at all. \"case(s) of the\"\n // are all optional here; only the number and keyword are required.\n new RegExp(\n `(\\d+(?:\\.\\d+)?)\\s*(?:cases?\\s+)?(?:x\\s+)?(?:of\\s+)?(?:the\\s+)?(?:${keyword.source})`,\n \"i\",\n ),\n ];\n for (const pattern of patterns) {\n const lineMatch = text.match(pattern);\n if (lineMatch) {\n lines.push({\n flavour: nameContains,\n quantity: Math.round(parseFloat(lineMatch[1])),\n });\n break; // don't double-count the same flavour against both patterns\n }\n }\n }\n return lines;\n}\n\nasync function createOrder(\n customerId: number,\n priceBook: number,\n poNumber: string,\n orderLines: { product: number; quantity: number }[],\n) {\n // Duplicate guard: if an order already exists for this customer with this\n // exact PO number (e.g. it was already created manually, or by an\n // earlier run whose email dedup we can no longer fully trust), don't\n // create a second one.\n const existingRes = await fetch(\n `${BREWW_BASE}orders/?customer=${customerId}&po_number=${\n encodeURIComponent(poNumber)\n }`,\n { headers: BREWW_HEADERS },\n );\n const existingData = await existingRes.json();\n const existing = (existingData.results ?? existingData)?.[0];\n if (existing?.id) {\n return { ...existing, _duplicate: true };\n }\n\n const payload = {\n customer: customerId,\n price_book: priceBook,\n issue_date: new Date().toISOString().slice(0, 10),\n po_number: poNumber,\n order_status: 1, // draft first, promote after sanity check\n fulfillment_required: true,\n // Critical: invoices must only go out at the point of delivery, per\n // Chris's normal process — never at order creation/acceptance. Without\n // this explicit false, Breww defaults to true and emails the customer\n // an invoice the moment the order is created, regardless of whether\n // anything has actually been fulfilled yet.\n allow_automatic_emailing_to_customer: false,\n order_lines: orderLines,\n };\n const res = await fetch(BREWW_BASE + \"orders/\", {\n method: \"POST\",\n headers: BREWW_HEADERS,\n body: JSON.stringify(payload),\n });\n return res.json();\n}\n\nasync function sendCustomerConfirmation(\n token: string,\n email: any,\n orderNumber: string,\n poNumber: string | null,\n) {\n const poLine = poNumber ? ` Your reference: ${poNumber}.` : \"\";\n const comment =\n `Thanks for your order.${poLine} This has been logged as order ${orderNumber} ` +\n `and will be added to our next processing run.\\n\\n` +\n `If anything above looks off, just reply to this email and we'll sort it out.`;\n\n await fetch(\n `https://graph.microsoft.com/v1.0/users/${MAILBOX}/messages/${email.id}/reply`,\n {\n method: \"POST\",\n headers: {\n Authorization: `Bearer ${token}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({ comment }),\n },\n );\n}\n\nasync function acceptOrder(orderId: number) {\n await fetch(`${BREWW_BASE}orders/${orderId}/`, {\n method: \"PATCH\",\n headers: BREWW_HEADERS,\n body: JSON.stringify({ order_status: 3 }),\n });\n}\n\nasync function sendSummaryEmail(\n token: string,\n processed: ProcessedItem[],\n review: ReviewItem[],\n) {\n if (processed.length === 0 && review.length === 0) return; // nothing to report\n\n const rows = (arr: string[]) => arr.map((r) => `

${r}

B2B order processing run — ${new Date().toISOString()}

\n ${\n processed.length\n ? `

Auto-accepted (${processed.length})

${processedRows}`\n : \"\"\n }\n ${\n review.length\n ? `

Diff · v31