[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_107":108,"_109":110},"routes/embed.x.$handle.$name.$",{"_9":10,"_11":12,"_13":14,"_15":16,"_17":18,"_19":20,"_21":-5},"handle","Tribeca_Rep","name","tribeca-chat","path","greeting.ts","project",{"_22":45,"_11":12,"_33":46,"_47":-5,"_48":42,"_49":50,"_51":50,"_52":53,"_54":-5,"_55":56,"_57":58,"_59":32,"_60":-5,"_61":-5,"_62":63,"_64":65,"_66":67,"_68":69,"_70":71,"_72":-5,"_73":74,"_75":74,"_76":77,"_78":77,"_79":77,"_80":74,"_81":74,"_82":83,"_84":85,"_86":87,"_88":42},"file",{"_22":23,"_11":14,"_24":25,"_26":-5,"_27":28,"_29":30,"_31":32,"_33":34,"_35":36,"_37":38,"_39":40,"_41":42,"_43":-5,"_13":44},"permalink","https://www.val.town/x/Tribeca_Rep/tribeca-chat/code/greeting.ts","extension","id","4c7717e6-107b-4ab6-a89a-4b8d6f9190e1","fileId","3f2669a2-a64a-11f1-8729-1607ee4eb77e","parentId","branchId","7f3f9c05-d6ad-413b-a869-bc1a5f60764f","branchVersion",703,"latest",true,"createdAt",["D",1789146259476],"version",81,"fileType","script","content","// The opening line.\n//\n// WHY THIS IS NOT A CONSTANT ANY MORE\n// tawk opened all 100 exported chats with \"Ask us if we deliver in your area!\n// What city do you live in?\" and it worked — 92% of real visitors replied. But\n// 52% of those first replies were four words or fewer, because the question\n// asks for something we can already know, and 48% of chats never got past that\n// exchange (82s median, against 600s for chats that reached buying intent).\n//\n// So the opener now spends its one turn ANSWERING rather than asking: coverage\n// confirmed, the real delivered price for their region, and an offer to do the\n// signup here. The next reply is then a decision instead of data entry.\n//\n// LOCATION COMES FROM THE ZIP, NEVER FROM AN IP. context.ts explains why in\n// full: city-from-IP means posting every visitor's address to a third-party\n// geolocator, which is the data flow moving off tawk was meant to remove, and\n// it is wrong often enough to put the wrong price in front of someone. A ZIP is\n// volunteered by the customer or already known to WooCommerce, and it is exact.\n// No ZIP means no quote — the generic opener is used instead.\n//\n// PRICES ARE NEVER WRITTEN HERE. They come from tribeca-profile, which is the\n// one place pricing is allowed to live. The ?v= pin is load-bearing: Val Town\n// caches a remote module by URL, so without bumping it this file would keep\n// quoting last week's prices with nothing in the logs.\nimport {\n COOLERS,\n GLASS_5GAL,\n money,\n PLASTIC_5GAL,\n UNSELLABLE_REGIONS,\n} from \"https://esm.town/v/Tribeca_Rep/tribeca-profile/profile.ts?v=2026-09-07d\";\n\n/**\n * THE OPENER. One question, answerable in one word.\n *\n * Chosen 9/1/2026 against measured evidence. tawk opened every one of its 100\n * exported chats with a question — \"what city do you live in?\" — and 92% of\n * real visitors answered; 52% of those replies were four words or fewer,\n * because a cheap question is a cheap thing to answer.\n *\n * This file previously opened by ANSWERING instead: coverage confirmed and the\n * delivered price for their region, in about forty words. Better information,\n * worse bait — 21% replied. Two problems with it. It needs a composed sentence\n * to accept, where tawk needed one word. And leading with the price satisfies\n * the visitor: they came to find out what it costs, they now know, and they can\n * leave.\n *\n * The price has not gone anywhere — the model still has the whole profile and\n * quotes it the moment anyone asks. It is just no longer the first thing said.\n */\nexport const FALLBACK_GREETING = \"Were you interested in water delivery?\";\n\ntype Region = keyof typeof PLASTIC_5GAL;\n\n/**\n * THE CHEAPEST TWO-BOTTLE ALKALINE PRICE WE ACTUALLY CHARGE, computed rather\n * than typed.\n *\n * The variants below say \"starts at\", and \"starts at\" has to be TRUE for every\n * region or it is a bait price — a New Yorker quoted $24.98 who is then billed\n * $31.92 was misled by the first sentence we ever said to them. Taking the\n * minimum from the shared table means it stays true when the table changes, and\n * it keeps this file's rule intact: prices are never written here.\n */\nconst FROM_ALKALINE = Math.min(\n ...Object.values(PLASTIC_5GAL).map((r: any) => Number(r.alkaline)),\n);\n\n/**\n * OPENER SPLIT TEST.\n *\n * Reply rates measured 9/1/2026, first day live. Small numbers, but the shape\n * was consistent enough to act on:\n *\n * \"Hi! Ask us about delivery in your area, pricing, or your account.\" 15/32\n * region-priced (\"We deliver in New Jersey — ... $26.98\") 5/6\n * \"Were you interested in water delivery?\" 0/8\n * \"Do you have glass bottles delivered currently?\" 0/13\n *\n * The two bare questions were at zero. The priced ones were not, and the\n * region-priced figure is confounded — a known ZIP usually means a signed-in\n * customer, a warmer audience — but a price is at least something a visitor has\n * to react to, where a yes/no question is free to ignore.\n *\n * SO IT IS NO LONGER ONE GUESS. Every variant here goes out to a fifth of\n * visitors with an unknown ZIP, and `visitors.opener_variant` records which,\n * so reply rate per variant is a query rather than an argument.\n *\n * NOT IN THE TEST, deliberately: the unsellable-area line, which is a factual\n * correction and not a pitch, and the glass-page opener, which is there by a\n * specific decision about that page. Both would need their own experiment.\n */\nexport const OPENER_VARIANTS: ReadonlyArray<{ id: string; text: string }> = [\n // RETIRED 9/6/2026, all four under a 4% reply-rate floor after 90 days'\n // measurement: \"question\" (the original control, 0.9%), \"discovery\" (3.4%),\n // \"price\" (2.9%), and \"setup\" (1.6%). Historical rows keep their ids —\n // nothing here deletes past data, only stops new assignment.\n //\n // RETIRED 9/11/2026, the next two, on the same reasoning: \"human\" (4.1%)\n // and \"discount\" (3.4%) both landed below the two survivors' reply rates\n // — has_delivery 5.9%, price_zip 5.5% — once the full-period all-time\n // numbers were in, not just the early read. The set[0] fallback in\n // openerVariantFor (for a request with no visitorId yet) lands on\n // has_delivery, which is still just \"whichever variant is safe to show,\"\n // not a claim that index 0 is special.\n //\n // THE SAME SHAPE THAT WORKS ON THE GLASS PAGE (\"Do you have glass bottles\n // delivered currently?\"), generalised — added 9/3/2026 at the office's\n // request. It asks something a ZIP cannot answer: whether this is a switch\n // from another delivery service or someone who has never had one at all,\n // and the two deserve different next lines — a switch is a comparison, a\n // first-timer is an explanation. Also a closed question, like the control,\n // but \"do you have X\" invites a story where \"were you interested in X\" only\n // invites yes or no.\n {\n id: \"has_delivery\",\n text: \"Do you have water delivery currently?\",\n },\n\n // The winning shape generalised: the fact first, then a question that is one\n // word to answer and hands us the one thing we need.\n {\n id: \"price_zip\",\n text: `Delivery starts at ${money(FROM_ALKALINE)} a month for two 5-gallon bottles. `\n + `What's your ZIP and I'll check your exact price?`,\n },\n];\n\n/**\n * THE GLASS PAGE GETS ITS OWN TEST, because it is a different problem.\n *\n * It is the busiest product page on the site — 100 visitors in the first tracked\n * sample — and the fastest bounce at 76 seconds average. Its opener is also the\n * worst performer measured so far, at 0 replies from 13. The reason is almost\n * certainly the price: glass looks four to five times plastic until you know\n * that $30 a bottle of it is a refundable deposit, and the current line asks a\n * question about their existing service without touching that at all.\n *\n * So the variants split on WHAT TO LEAD WITH: the question we ask now, the\n * deposit that makes the number make sense, or the offer to check their area.\n *\n * THE PROFILE'S \"NEVER VOLUNTEER GLASS\" RULE IS NOT BROKEN HERE. That rule is\n * about not steering someone towards a $134.98 decision they never asked for.\n * Somebody standing on the glass product page has asked. It still holds\n * everywhere else, which is why this set is reachable only from that page.\n */\n// RETIRED 9/4/2026 — glass_price led with the full itemized breakdown and\n// scored 0 replies from 95 shown, the worst of this test despite having the\n// largest sample. Leading with ANY price on this page loses.\n//\n// RETIRED 9/6/2026 — glass_current (1.3%), glass_deposit (2.1%), and\n// glass_discount (0.9%) all fell under a 4% reply-rate floor after 90 days'\n// measurement, leaving glass_check (2.8%, still under the floor but the best\n// of the four and the only one left) as the page's sole opener. A one-element\n// set is deliberate here, not a placeholder for more variants: openerVariantFor\n// still runs its hash-and-mod against it, which is a no-op at length 1 but\n// means this needs no special-casing if a new variant is added later.\n// Historical rows for every retired id are untouched, and pageNudge's\n// alreadyPriced check still recognizes glass_price/glass_deposit for any\n// straggler session that saw them before retirement.\nexport const GLASS_VARIANTS: ReadonlyArray<{ id: string; text: string }> = [\n // No price, no product — just the offer. The counterweight: if the numbers\n // are what scares people off this page, the variant that omits them entirely\n // is the one that tells us.\n {\n id: \"glass_check\",\n text: \"Want me to check whether we deliver glass bottles to you?\",\n },\n];\n\n/**\n * Which variant this visitor gets — STABLE, from their id, not random.\n *\n * A visitor who reloads or moves to a second page must see the same opener.\n * Math.random() here would reassign them on every page view, so one person\n * could be counted under three variants and the whole comparison would measure\n * nothing. Hashing the id also means no state to store and no coordination\n * between isolates.\n *\n * djb2-ish, forced unsigned. The ids are random hex, so this distributes\n * evenly without anything cleverer.\n */\nexport function openerVariantFor(\n visitorId: unknown,\n set: ReadonlyArray<{ id: string; text: string }> = OPENER_VARIANTS,\n): { id: string; text: string } {\n // No id yet means the very first request of a session, before presence has\n // handed one back. set[0] is just whichever variant happens to be first —\n // there is nothing special about that position — and the row is written\n // under that variant's own id either way, so this never pollutes the numbers.\n const s = String(visitorId ?? \"\");\n if (!s) return set[0];\n let h = 0;\n for (let i = 0; i < s.length; i++) h = ((h * 31 + s.charCodeAt(i)) >>> 0);\n return set[h % set.length];\n}\n\n/**\n * ZIP to the pricing region the store itself uses.\n *\n * DELIBERATELY CONSERVATIVE, and it returns null far more often than a ZIP map\n * would. Every range below is one we actually serve; anything outside them —\n * upstate New York, Jacksonville, the Florida panhandle, most of California —\n * returns null and gets the generic opener.\n *\n * The failure modes are not symmetrical. A visitor who gets the generic\n * greeting has lost nothing. A visitor told \"we deliver in your area\" who then\n * finds out we do not has been lied to by a robot, and quoting the wrong\n * region's price is worse still, because the number is the thing they remember.\n */\n/**\n * Areas the store will happily sell to and we cannot actually deliver to.\n *\n * NAMED, not just flagged. There are two of these now, and the opener has to\n * say which one — telling someone in Melbourne that we get asked about\n * \"Greater Philadelphia\" a lot is worse than saying nothing.\n */\nfunction unsellableAreaForZip(z: number): string | null {\n // PENNSYLVANIA REMOVED 9/1/2026 — WE DO DELIVER THERE.\n //\n // 15000-19699 used to return \"Greater Philadelphia Area\" and every PA\n // visitor was opened with \"it needs a team member to set up rather than the\n // website\". The office confirmed we deliver plastic and glass to PA; the\n // store's inability to take the order online was never the same thing as\n // not having a route.\n //\n // PA now falls through to null, the same as any ZIP we hold no price for.\n // That is the correct behaviour while the PA plastic price is unconfirmed:\n // the generic opener, no refusal and no invented number.\n // Sells and prices fine on the website; the route does not exist. Confirmed\n // by the office 9/1/2026.\n // \"area\" is not decoration — the opener says \"the ${which}\", and \"the Greater\n // Melbourne a lot\" is a sentence no person would write.\n if (z >= 32900 && z <= 32999) return \"Greater Melbourne area\";\n return null;\n}\n\nexport function pricingRegionForZip(zip: unknown): Region | \"UNSELLABLE\" | null {\n const z = Number(String(zip ?? \"\").trim().slice(0, 5));\n if (!z || Number.isNaN(z)) return null;\n\n // Checked FIRST. These ranges sit inside otherwise-served states, so a later\n // range test would happily quote them a price for a truck that never comes.\n if (unsellableAreaForZip(z)) return \"UNSELLABLE\";\n\n if (z >= 6000 && z <= 6999) return \"Connecticut\";\n if (z >= 7000 && z <= 8999) return \"New Jersey\";\n\n // NYC, Long Island, Westchester and Rockland only. 12000+ is upstate and not\n // served, so the range deliberately stops short of it.\n if (z >= 10001 && z <= 11999) return \"New York\";\n\n // DC, Maryland and Northern Virginia.\n if (z >= 20000 && z <= 22999) return \"Washington DC Metro (MD, VA)\";\n\n // Los Angeles and Orange County. Northern California is not served.\n if (z >= 90000 && z <= 92899) return \"California\";\n\n // Florida, by route rather than by state. The panhandle (324xx-325xx) and\n // Jacksonville (320xx-322xx) are absent because we do not go there.\n if (z >= 32700 && z <= 32899) return \"Florida - Tampa Metro / Pinellas / Central Florida\"; // Orlando, Kissimmee\n if (z >= 33000 && z <= 33499) return \"Florida - Miami Metro (Dade/Broward/Palm Beach)\";\n if (z >= 33500 && z <= 34299) return \"Florida - Tampa Metro / Pinellas / Central Florida\"; // Tampa, St Pete, Sarasota, Lakeland\n // Added 9/7/2026, same route: the 342xx block above already covered\n // Sarasota and Manatee County, but Pasco County's own ZIPs split across the\n // boundary — its southern towns are 335xx-342xx (already matched) while its\n // northern and western half (Land O' Lakes, New Port Richey, Holiday,\n // Wesley Chapel's own 34638) and all of Hernando County (Brooksville,\n // Spring Hill) sit in 346xx, past the old 34299 ceiling. Confirmed served\n // by the office the same day as Manatee and Pasco.\n if (z >= 34600 && z <= 34699) return \"Florida - Tampa Metro / Pinellas / Central Florida\"; // Pasco (north/west), Hernando\n\n return null;\n}\n\n// MOVED TO tribeca-profile ON 9/3/2026, and imported at the top of this file.\n//\n// It lived here, privately, while this was the only file that formatted a\n// price. It is not any more — the order-confirmation line in main.ts hit the\n// identical bug on 9/2/2026 and had to defend itself separately with string\n// concatenation. Five bots read prices out of the shared profile; the\n// formatter belongs beside them, so a fix made once is a fix made everywhere.\n\n/** Short label for the opener — the full keys name routes, not places. */\nfunction placeLabel(region: Region): string {\n if (String(region).startsWith(\"Florida\")) return \"your part of Florida\";\n if (region === \"Washington DC Metro (MD, VA)\") return \"the DC metro\";\n return String(region);\n}\n\n/**\n * Build the opening line.\n *\n * Quotes `cartTotal`, not `pageTotal`. The store's product page under-quotes by\n * $2 in every region except California — a known defect recorded in\n * tribeca-profile — and cartTotal is what the customer is actually charged.\n * Opening with the cheaper number would make the first thing we ever tell them\n * the thing they later find out was wrong.\n *\n * The 25% discount is deliberately NOT here. It is a closer, not an opener:\n * offering money off before anyone has hesitated discounts a sale already made.\n */\n/** The glass product page, by the two URL shapes the store actually uses. */\nfunction isGlassPage(pageUrl: unknown): boolean {\n const u = String(pageUrl ?? \"\").toLowerCase();\n return u.includes(\"glass-bottles\") || (u.includes(\"glass\") && u.includes(\"/product/\"));\n}\n\n/**\n * ARRIVED HERE FROM THE LOGIN PAGE — the same pattern the phone console's\n * cameFrom() already uses to describe this in the detail panel, pulled out so\n * the opener and nudge can act on it rather than only display it.\n *\n * WHY IT MATTERS: someone trying to sign in has already told us the one thing\n * the discovery openers exist to find out. \"How do you buy your water at the\n * moment?\" or \"Do you have water delivery currently?\" asked of a person\n * actively logging into their TriBeCa account is not a question — it is the\n * bot not noticing what is right in front of it.\n */\nfunction cameFromLoginPage(referrer: unknown): boolean {\n return /wp-login\\.php|\\/wp-admin\\/|[?&]action=login/i.test(String(referrer ?? \"\"));\n}\n\n/**\n * RETURNS THE VARIANT ALONGSIDE THE TEXT, not just the line.\n *\n * The console already stores `shown_opener`, so the wording alone could be\n * grouped on — but only until the day someone fixes a comma in one variant,\n * at which point that variant silently splits into two rows and the history\n * before the edit stops matching the history after it. A short stable id\n * survives rewording, which is the whole point of running this for weeks.\n *\n * The fixed branches return an id too (\"unsellable\", \"glass\") rather than\n * null, so the report accounts for every greeting rather than leaving a\n * blank bucket nobody can interpret.\n */\nexport function openingGreeting(\n zip: unknown,\n pageUrl?: unknown,\n visitorId?: unknown,\n isCustomer?: boolean,\n referrer?: unknown,\n): { text: string; variant: string } {\n // A KNOWN PAYING CUSTOMER GETS NO PITCH, CHECKED FIRST.\n //\n // Every branch below exists to sell — a discount, a ZIP-check, an offer to\n // \"get you set up\". None of that belongs in front of someone who already\n // has a route: Brian Grossbard, signed into the customer portal with a\n // delivery on the 12th, does not need to hear \"order through me and I'll\n // take 25% off your first delivery\". It reads as the bot not knowing who he\n // is, when the whole point of the Salesforce match is that we do.\n //\n // CHECKED BEFORE the unsellable-region branch, because it doesn't apply —\n // an existing customer already has a working route, whatever the ZIP table\n // says about new signups in that area. Checked before the glass-page branch\n // for the same reason: there is nothing to upsell them into.\n //\n // ONE FIXED LINE, not run through the split test. The A/B openers exist to\n // learn what converts a stranger; a customer is not being converted, so\n // mixing them into that measurement would blur the result for both.\n if (isCustomer) {\n return {\n variant: \"customer\",\n text: \"Hey! What can I help with?\",\n };\n }\n\n // CAME FROM THE LOGIN PAGE — the discovery question has already been\n // answered by the fact of this visit, so it is not asked again.\n //\n // CHECKED AFTER isCustomer, not merged into it. A login attempt is real\n // evidence of an existing relationship, but it is not the same evidence: a\n // confirmed Salesforce match means we KNOW the account, where this means\n // someone is TRYING to reach one — possibly failing, possibly a brand new\n // WooCommerce signup with no delivery history at all. It earns a different\n // line, not the sf_account block's full no-pitch treatment.\n if (!isCustomer && cameFromLoginPage(referrer)) {\n return {\n variant: \"from_login\",\n text: \"Looks like you were headed to sign in — need a hand with that, \"\n + \"or is there something else I can help with?\",\n };\n }\n\n const region = pricingRegionForZip(zip);\n\n // GLASS IS THE ONE PAGE THAT GETS ITS OWN OPENER.\n //\n // The profile forbids volunteering glass — it is four to five times the price\n // with a $30-per-bottle deposit, and offering it unprompted turns a $26.98\n // decision into a $134.98 one. That rule is about not RAISING it. Someone\n // standing on the glass product page has raised it themselves, so asking\n // about it is answering where they are rather than upselling.\n //\n // Asked as \"currently\" on purpose: it separates an existing glass customer\n // from someone considering it, and either answer is one word.\n //\n // CHECKED AFTER the unsellable branch below, not before. Someone in\n // Philadelphia or Melbourne standing on the glass page cannot be sold\n // anything at all, and asking about their glass delivery before telling them\n // we do not deliver to them is the wrong order to say two things in.\n if (region === \"UNSELLABLE\") {\n // Named per ZIP, NOT UNSELLABLE_REGIONS[0]. That index worked while\n // Philadelphia was the only entry; adding Melbourne on 9/1/2026 would have\n // started telling people in Florida about Pennsylvania.\n const z = Number(String(zip ?? \"\").trim().slice(0, 5));\n const which = unsellableAreaForZip(z) ?? UNSELLABLE_REGIONS[0] ?? \"your area\";\n return {\n variant: \"unsellable\",\n text: `Hi! We get asked about the ${which} a lot — it needs a team member to set up `\n + `rather than the website. Tell me what you need and I'll pass it straight on.`,\n };\n }\n\n // JUST PAID. No pitch, and above all no discount — the variants below\n // include \"25% off your first delivery\", which is the worst sentence we\n // could put in front of somebody who has this second paid full price.\n //\n // This line is only ever seen if they open the widget THEMSELVES; the widget\n // does not auto-open on the receipt page.\n const u0 = String(pageUrl ?? \"\").toLowerCase();\n if (u0.includes(\"order-received\") || u0.includes(\"/thank-you\")) {\n return { variant: \"post_order\", text: \"Anything you need on that order, I'm here.\" };\n }\n\n if (isGlassPage(pageUrl)) {\n // Its own split, on its own set. The glass page is the busiest and the\n // fastest-bouncing on the site, and its single opener is the worst\n // performer measured — so it gets tested rather than guessed at again.\n const g = openerVariantFor(visitorId, GLASS_VARIANTS);\n return { variant: g.id, text: g.text };\n }\n\n // Deliberately the SAME question whether or not we know their region. Knowing\n // the ZIP is still worth everything — it decides the unsellable branch above,\n // it feeds pageNudge, and it means the price is right the moment they ask —\n // but a visitor who has not spoken yet does not need it recited at them.\n //\n // THIS IS THE BRANCH UNDER TEST, and it is the great majority of visitors —\n // most arrive from a search with no ZIP known to anyone. Whatever wins here\n // is the opener the business runs on.\n const v = openerVariantFor(visitorId);\n return { text: v.text, variant: v.id };\n}\n\n/**\n * The second line, sent only when the opener got no reply.\n *\n * WHY IT IS KEYED ON THE PAGE. Measured over the first 394 tracked visitors,\n * what someone is looking at separates them far more sharply than anything else\n * we know:\n *\n * glass bottles 100 visitors, 76s average — the busiest product page and\n * the fastest bounce\n * home 77 visitors, 154s\n * plastic 5gal 25 visitors, 187s\n * contact us 23 visitors, 339s — five minutes filling a form\n * checkout / cart 10 visitors, 334s — five minutes stuck\n *\n * The two long-dwell pages are people already trying to reach you, and the\n * busiest one is people leaving. A generic \"still there?\" earns nothing from\n * any of them, so each gets the one fact that would actually move it.\n *\n * Returns null where a nudge would be wrong rather than merely useless — after\n * checkout the sale is already made, and chasing someone who has just paid\n * reads as a system that has not noticed.\n */\nexport function pageNudge(\n pageUrl: unknown,\n openerVariant?: unknown,\n isCustomer?: boolean,\n referrer?: unknown,\n): string | null {\n // SAME RULE AS THE OPENER: nothing below this line is for someone we\n // already deliver to. Checked first and independent of page, because every\n // branch that follows is keyed on which product page a STRANGER is\n // standing on — checkout, glass, coolers — and a known customer trailing\n // off on any of them still does not need a pitch, just a door left open.\n if (isCustomer) {\n return \"Let me know if anything comes up — happy to help.\";\n }\n\n // SAME LOGIC AS THE OPENER, for someone who did not answer it. Repeating\n // \"do you have delivery currently\" a second time is not just redundant —\n // it never made sense the first time, so a follow-up must not fall through\n // to it either. See cameFromLoginPage on the opener side for the reasoning.\n if (cameFromLoginPage(referrer)) {\n return \"Still having trouble signing in? Happy to help, or I can flag it for the office.\";\n }\n\n const u = String(pageUrl ?? \"\").toLowerCase();\n if (!u) return null;\n\n // THE FOLLOW-UP MUST NOT REPEAT THE OPENER.\n //\n // Caught live 9/1/2026 on the glass page: the opener said \"Glass is $69.99\n // for one bottle or $134.98 for two, delivery included\", and twenty-five\n // seconds later the nudge said the same three numbers again in a longer\n // sentence. Two messages, one fact, nobody having said a word in between —\n // which reads as a machine repeating itself, and is worse than silence.\n //\n // It happened because the openers became variable and this function did not\n // know which one had been used. It does now.\n const alreadyPriced = openerVariant === \"glass_price\" || openerVariant === \"glass_deposit\";\n\n // ALREADY BOUGHT — SAY NOTHING. Removed 9/1/2026 on the office's call.\n //\n // There used to be a cheerful \"Thanks for the order!\" here. It is the wrong\n // instinct: someone who has just paid is reading their receipt, and an\n // unprompted message at that moment is a business interrupting a customer\n // for its own benefit. The widget is still sitting there if they want it.\n if (u.includes(\"order-received\") || u.includes(\"/thank-you\")) return null;\n\n // Stuck in checkout for minutes. This is the highest-value interruption we\n // have, and the offer is to take the work away rather than to help them do it.\n if (u.includes(\"/checkout\") || u.includes(\"/cart\")) {\n return \"If checkout's being awkward I can just set it up here — I only need your \"\n + \"address and how many bottles.\";\n }\n\n // The busiest product page and the shortest visit. The deposit is almost\n // certainly why: the price looks enormous until you know $30 a bottle of it\n // is refundable, so that is the fact that goes first.\n if (u.includes(\"glass-bottles\") || (u.includes(\"glass\") && u.includes(\"/product/\"))) {\n // They have already been given the numbers. Say the thing the price does\n // not say — that most of what looks expensive comes back — and keep it to\n // one line.\n if (alreadyPriced) {\n // A DIFFERENT closing question too. Both glass openers that carry a price\n // end with \"want me to check your area?\", so ending the follow-up the\n // same way asks the same thing twice in a row — which is the smaller\n // version of the bug this branch exists to fix.\n return \"Worth knowing — refills drop to \" + money(GLASS_5GAL.ongoingOneBottle)\n + \" once your empties start going back. Want me to just set it up?\";\n }\n\n // ONE BOTTLE, and the deposit named as one-time and refundable — the same\n // change as glass_price above, for the same reason. The two-bottle total\n // was doing the damage: it is the biggest number on the page and it landed\n // before anyone had explained that most of it comes back.\n return \"On the glass — \" + money(GLASS_5GAL.waterPerBottle) + \" a bottle, \"\n + money(GLASS_5GAL.deliveryFee) + \" delivery per drop, and a \"\n + money(GLASS_5GAL.depositPerBottle) + \" deposit you get back when the empty \"\n + \"goes back. So \" + money(GLASS_5GAL.oneBottleTotal) + \" to start and \"\n + money(GLASS_5GAL.ongoingOneBottle) + \" after. Want me to check your area?\";\n }\n\n if (u.includes(\"cooler\") || u.includes(\"dispenser\")) {\n // All three named, cheapest first. Quoting only \"$8\" for every unit is the\n // mistake the profile now warns about — it was the single \"from\" figure\n // until 9/1/2026 and the other two units were being under-quoted.\n return \"Coolers are \" + money(COOLERS[\"cold only\"]) + \" a month cold only, \"\n + money(COOLERS[\"hot and cold\"]) + \" hot and cold, \"\n + money(COOLERS[\"bottom-load\"]) + \" bottom-load — service is free on all of them. \"\n + \"Want one on a delivery?\";\n }\n\n // Five minutes on a contact form is someone who has decided to get in touch\n // and is doing it the slow way.\n if (u.includes(\"contact\")) {\n return \"No need for the form — just tell me here and I'll sort it or pass it \"\n + \"straight on.\";\n }\n\n if (u.includes(\"/product/\") || u.includes(\"spring-water\") || u.includes(\"water-delivery\")) {\n return \"Want me to check your street and what it'd come to? Just need your ZIP.\";\n }\n\n // Home page and everything else — the widest catch-all, and so the moment\n // \"things get silent\" for most visitors.\n //\n // LEADS WITH THE SAME DISCOVERY QUESTION AS THE OPENER, not a pitch — added\n // 9/3/2026 alongside the has_delivery opener, and for the same reason PACE\n // exists in the AI's own instructions: find out where someone is before\n // asking for the sale. A stranger who ignored a fact-and-ZIP opener is not\n // more likely to answer a second one; a plain question about their own\n // situation is easier to answer and does not read as a second sales attempt.\n //\n // MUST NOT REPEAT THE OPENER — same rule as glass. If has_delivery was\n // already shown and went unanswered, asking it again is the exact bug that\n // was fixed there: two messages, one question, nobody having said a word in\n // between. Falls back to the previous fact-first line instead.\n if (openerVariant === \"has_delivery\") {\n return \"No contract, and you can skip or change any delivery. What's your ZIP and \"\n + \"I'll check your area?\";\n }\n return \"Do you have water delivery currently?\";\n}\n","deleted",false,"label",[14],"53b5cfec-2084-4c81-ba38-78b35a1e8625",["D",1788226289223],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","Chat widget, AI answering, human takeover, Salesforce","imageUrl","userId","c5acd3b3-60a7-4deb-bc6d-528f83012ea9","pinnedFileId","db056860-a5a4-11f1-8f50-1607ee4eb77e","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","019f8fe2-cdfc-775d-a845-a69af13d39cb","tursoName","53b5cfec20844c81ba3878b35a1e8625","tursoHostname","53b5cfec20844c81ba3878b35a1e8625-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789484509295],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_9":10,"_106":102},"org",{"_9":10,"_99":100,"_101":102,"_55":56,"_103":104,"_105":42},"branch",{"_35":89,"_11":90,"_22":28,"_91":45,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":-5,"_97":12,"_98":10},"isOwned",709,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","displayName","M A","avatar","https://lh3.googleusercontent.com/a/ACg8ocI9znhXn2hWp1wjjyWzeuOyoO_JrDgJo8aD0umFraR7bAG3UAmc=s96-c","tier","free","canRestrictVals","profileImageUrl","ctx",{"_82":-5,"_140":141,"_142":143,"_144":145,"_146":42,"_147":42,"_148":42,"_149":32,"_150":-5,"_151":-7,"_152":-7,"_153":63,"_154":42,"_155":42,"_156":-5,"_157":42,"_158":42,"_159":42,"_160":-5},"ENV",{"_111":112,"_113":114,"_115":90,"_116":117,"_118":119,"_120":121,"_122":123,"_124":125,"_126":127,"_128":129,"_130":131,"_132":133,"_134":135,"_136":137,"_138":139},"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"]