[{"_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","veltson","name","amazon-pharmacy-rfp-demo","path","catalog.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/veltson/amazon-pharmacy-rfp-demo/code/catalog.ts","extension","id","b61b655f-a54a-4350-b9e3-9d9d46da6eed","fileId","8cdee466-a1be-11f1-a1eb-1607ee4eb77e","parentId","branchId","93f67de8-2473-491d-8a83-4d82ebfd56cc","branchVersion",89,"latest",true,"createdAt",["D",1788200604204],"version",21,"fileType","script","content","// Medication catalog + spoken-number formatting for the Amazon Pharmacy RFP demo.\n// The assistant and the printable table at GET /catalog both read this one constant,\n// so the PDF handed to the customer can never drift from what the agent knows.\n\n// ---- spoken-number formatting ----------------------------------------------\n// The RFP grades this directly: \"$123\" is \"one-hundred twenty-three dollars\", but a\n// verification code is \"one two three\". Precomputing the words server-side is the\n// only reliable way to keep the two registers apart.\nconst ONES = [\"zero\",\"one\",\"two\",\"three\",\"four\",\"five\",\"six\",\"seven\",\"eight\",\"nine\",\"ten\",\n \"eleven\",\"twelve\",\"thirteen\",\"fourteen\",\"fifteen\",\"sixteen\",\"seventeen\",\"eighteen\",\"nineteen\"];\nconst TENS = [\"\",\"\",\"twenty\",\"thirty\",\"forty\",\"fifty\",\"sixty\",\"seventy\",\"eighty\",\"ninety\"];\n\n// ---- Spanish wording ---------------------------------------------------------\n// A second wording set behind the SAME functions, not a fork. Every exported formatter\n// below takes `lang` LAST and defaults it to \"en\", so a call site that was never updated\n// still produces byte-identical English. The English bodies are untouched.\n//\n// Spanish is not a find-and-replace over the English tables, because Spanish cardinals\n// agree with the noun they count. Getting that wrong produces \"veintiuno dólares\", which\n// is exactly the kind of thing this RFP grades.\nexport type Lang = \"en\" | \"es\";\n\n// Which form of \"one\" (and of the hundreds) a number takes is decided by what follows it:\n// \"m\" directly before a masculine noun un dólar, veintiún dólares, doscientos\n// \"f\" directly before a feminine noun una unidad, veintiuna, doscientas\n// \"neutral\" standing alone, counting uno, veintiuno, treinta y uno\n// Every Spanish helper takes one explicitly rather than guessing, because guessing is how\n// the wrong form ships.\nexport type EsForm = \"m\" | \"f\" | \"neutral\";\n\nconst ES_ONES = [\"cero\",\"uno\",\"dos\",\"tres\",\"cuatro\",\"cinco\",\"seis\",\"siete\",\"ocho\",\"nueve\",\"diez\",\n \"once\",\"doce\",\"trece\",\"catorce\",\"quince\",\"dieciséis\",\"diecisiete\",\"dieciocho\",\"diecinueve\"];\nconst ES_TENS = [\"\",\"\",\"veinte\",\"treinta\",\"cuarenta\",\"cincuenta\",\"sesenta\",\"setenta\",\"ochenta\",\"noventa\"];\nconst ES_HUNDREDS = [\"\",\"ciento\",\"doscientos\",\"trescientos\",\"cuatrocientos\",\"quinientos\",\n \"seiscientos\",\"setecientos\",\"ochocientos\",\"novecientos\"];\n\nconst esOne = (form: EsForm) => form === \"m\" ? \"un\" : form === \"f\" ? \"una\" : \"uno\";\n\nfunction intWordsEs(n: number, form: EsForm): string {\n if (n < 0) return \"menos \" + intWordsEs(-n, form);\n if (n === 1) return esOne(form);\n if (n < 20) return ES_ONES[n];\n if (n < 30) {\n // The twenties are single words, and three of them carry an accent that the generic\n // tens-plus-ones rule below would silently drop.\n const r = n % 10;\n if (!r) return \"veinte\";\n if (r === 1) return form === \"m\" ? \"veintiún\" : form === \"f\" ? \"veintiuna\" : \"veintiuno\";\n return ({ 2: \"veintidós\", 3: \"veintitrés\", 6: \"veintiséis\" } as Record)[r]\n ?? \"veinti\" + ES_ONES[r];\n }\n if (n < 100) {\n const t = ES_TENS[Math.floor(n / 10)], r = n % 10;\n return r ? `${t} y ${intWordsEs(r, form)}` : t;\n }\n if (n < 1000) {\n const h = Math.floor(n / 100), r = n % 100;\n // 100 exactly is \"cien\"; anything above it switches to \"ciento\".\n if (h === 1 && !r) return \"cien\";\n // 200 through 900 agree in gender. \"cien\" and \"ciento\" do not.\n const head = (h > 1 && form === \"f\") ? ES_HUNDREDS[h].replace(/os$/, \"as\") : ES_HUNDREDS[h];\n return r ? `${head} ${intWordsEs(r, form)}` : head;\n }\n if (n < 1e6) {\n const th = Math.floor(n / 1000), r = n % 1000;\n // \"mil\", never \"un mil\". The multiplier itself is masculine regardless of the noun\n // being counted, because it is agreeing with \"mil\".\n const head = th === 1 ? \"mil\" : `${intWordsEs(th, \"m\")} mil`;\n return r ? `${head} ${intWordsEs(r, form)}` : head;\n }\n return String(n);\n}\n\n// The decimal separator read aloud. Peninsular Spanish says \"coma\"; US Latin American and\n// Mexican usage says \"punto\". The demo audience is US-based Amazon Pharmacy and the target\n// register is Mexican Spanish, so \"punto\". Change this one line if the customer disagrees.\nconst ES_DECIMAL_SEPARATOR = \"punto\";\n\n// Units carry their own gender so strengthWords can agree without a lookup at the call\n// site. An unknown unit has no entry, and the caller falls back to the neutral form rather\n// than inventing an agreement.\nconst ES_UNITS: Record = {\n mg: { plural: \"miligramos\", singular: \"miligramo\", gender: \"m\" },\n mcg: { plural: \"microgramos\", singular: \"microgramo\", gender: \"m\" },\n \"units/mL\": { plural: \"unidades por mililitro\", singular: \"unidad por mililitro\", gender: \"f\" },\n};\n\nfunction moneyWordsEs(amount: number): string {\n const cents = Math.round(amount * 100);\n const d = Math.floor(cents / 100), c = cents % 100;\n // \"dólar\" is masculine, so 1 apocopates to \"un\" and 21 to \"veintiún\".\n const dollars = `${intWordsEs(d, \"m\")} ${d === 1 ? \"dólar\" : \"dólares\"}`;\n if (!c) return dollars;\n // \"con\" rather than \"y\". Both are heard; \"con\" is the form a Mexican pharmacy or\n // checkout reads a price in, and mixing the two across a call is worse than either.\n return `${dollars} con ${intWordsEs(c, \"m\")} ${c === 1 ? \"centavo\" : \"centavos\"}`;\n}\n\nfunction decimalWordsEs(x: number, form: EsForm): string {\n const s = String(x);\n if (!s.includes(\".\")) return intWordsEs(x, form);\n const [w, f] = s.split(\".\");\n // The integer part is NOT adjacent to the noun once a fraction follows it, so it takes\n // the neutral form: \"uno punto cinco miligramos\", never \"un punto cinco\".\n const frac = f.split(\"\").map((d) => intWordsEs(Number(d), \"neutral\")).join(\" \");\n return `${intWordsEs(Number(w), \"neutral\")} ${ES_DECIMAL_SEPARATOR} ${frac}`;\n}\n\nexport function intWords(n: number, lang: Lang = \"en\", form: EsForm = \"m\"): string {\n if (lang === \"es\") return intWordsEs(n, form);\n if (n < 0) return \"negative \" + intWords(-n);\n if (n < 20) return ONES[n];\n if (n < 100) {\n const t = TENS[Math.floor(n / 10)], r = n % 10;\n return r ? `${t}-${ONES[r]}` : t;\n }\n if (n < 1000) {\n const h = `${ONES[Math.floor(n / 100)]}-hundred`, r = n % 100;\n return r ? `${h} ${intWords(r)}` : h;\n }\n if (n < 1e6) {\n const th = `${intWords(Math.floor(n / 1000))} thousand`, r = n % 1000;\n return r ? `${th} ${intWords(r)}` : th;\n }\n return String(n);\n}\n\n// 50 -> \"fifty dollars\"; 3.1 -> \"three dollars and ten cents\"\n// es: 50 -> \"cincuenta dólares\"; 1 -> \"un dólar\"; 3.1 -> \"tres dólares con diez centavos\"\nexport function moneyWords(amount: number, lang: Lang = \"en\"): string {\n if (lang === \"es\") return moneyWordsEs(amount);\n const cents = Math.round(amount * 100);\n const d = Math.floor(cents / 100), c = cents % 100;\n const dollars = `${intWords(d)} dollar${d === 1 ? \"\" : \"s\"}`;\n if (!c) return dollars;\n return `${dollars} and ${intWords(c)} cent${c === 1 ? \"\" : \"s\"}`;\n}\n\n// 12.5 -> \"twelve point five\" (dose register, never \"twelve and a half\")\n// es: 12.5 -> \"doce punto cinco\"\nexport function decimalWords(x: number, lang: Lang = \"en\", form: EsForm = \"m\"): string {\n if (lang === \"es\") return decimalWordsEs(x, form);\n const s = String(x);\n if (!s.includes(\".\")) return intWords(x);\n const [w, f] = s.split(\".\");\n return `${intWords(Number(w))} point ${f.split(\"\").map((d) => ONES[Number(d)]).join(\" \")}`;\n}\n\n// \"936722\" -> \"nine three six seven two two\" (verification-code register)\n// es: \"936722\" -> \"nueve tres seis siete dos dos\"\nexport const digitWords = (s: string, lang: Lang = \"en\") =>\n s.split(\"\").filter((c) => /\\d/.test(c))\n .map((d) => lang === \"es\" ? ES_ONES[Number(d)] : ONES[Number(d)]).join(\" \");\n\nexport const unitLong = (u: string, lang: Lang = \"en\") =>\n lang === \"es\" ? (ES_UNITS[u]?.plural ?? u)\n : u === \"mg\" ? \"milligrams\" : u === \"mcg\" ? \"micrograms\" : u;\n\n// \"one milligrams\" is the kind of thing an evaluator writes down, and this RFP grades\n// spoken number formatting explicitly. Only a strength of exactly 1 takes the singular;\n// 0.5 and 1.5 are both plural in speech, as is every other value. Same rule in Spanish,\n// where 1 additionally apocopates: \"un miligramo\", \"cero punto cinco miligramos\".\nconst singularise = (u: string) =>\n u.endsWith(\"s\") ? u.slice(0, -1) : u;\nexport const strengthWords = (v: number, unit: string, lang: Lang = \"en\") => {\n if (lang === \"es\") {\n const e = ES_UNITS[unit];\n if (!e) return `${decimalWordsEs(v, \"neutral\")} ${unit}`;\n return `${decimalWordsEs(v, e.gender)} ${v === 1 ? e.singular : e.plural}`;\n }\n return `${decimalWords(v)} ${v === 1 ? singularise(unitLong(unit)) : unitLong(unit)}`;\n};\n\nexport const strengthSpoken = (v: number, unit: string, lang: Lang = \"en\") =>\n `${decimalWords(v, lang, \"neutral\")} ${unitLong(unit, lang)}`;\n\n// ---- catalog ----------------------------------------------------------------\nexport type Drug = {\n name: string; generic: string; category: string; form: string; unit: string;\n schedule: \"weekly\" | \"daily\";\n strengths: number[]; rx: boolean;\n stock: \"in_stock\" | \"limited\" | \"backordered\";\n supplies: number[];\n insurance: Record;\n cash: Record;\n couponMax: number; limited: number[]; mlPerDose: number | null;\n};\n\nconst W = \"weekly\" as const, DA = \"daily\" as const;\nconst WK = [28, 56, 84], DY = [30, 60, 90], AB = [10, 20, 30];\nconst PEN = \"single-dose pen injection\", TAB = \"tablet\", CAP = \"capsule\";\nconst GLP = \"Weight management & GLP-1\", DIA = \"Diabetes\", CV = \"Cardiovascular\";\nconst RESP = \"Respiratory & allergy\", MH = \"Mental health\", GI = \"Digestive\";\nconst THY = \"Thyroid\", PAIN = \"Pain & inflammation\", ABX = \"Antibiotics\";\nconst OTH = \"Other prescription\", OTC = \"Over-the-counter\";\n\nfunction D(\n name: string, generic: string, category: string, form: string, unit: string,\n schedule: \"weekly\" | \"daily\", strengths: number[], rx: boolean,\n stock: \"in_stock\" | \"limited\" | \"backordered\", supplies: number[],\n ins: number[], cash: number[], couponMax = 0, limited: number[] = [], mlPerDose: number | null = null,\n): Drug {\n const zip = (vals: number[]) => Object.fromEntries(supplies.map((s, i) => [String(s), vals[i]]));\n return { name, generic, category, form, unit, schedule, strengths, rx, stock, supplies,\n insurance: zip(ins), cash: zip(cash), couponMax, limited, mlPerDose };\n}\n\nexport const CATALOG: Drug[] = [\n D(\"Zepbound\",\"tirzepatide\",GLP,PEN,\"mg\",W,[2.5,5,7.5,10,12.5,15],true,\"in_stock\",WK,[25,38,50],[649,1249,1799],25,[15],0.5),\n D(\"Mounjaro\",\"tirzepatide\",GLP,PEN,\"mg\",W,[2.5,5,7.5,10,12.5,15],true,\"in_stock\",WK,[25,40,55],[1069,2079,3049],25,[],0.5),\n D(\"Wegovy\",\"semaglutide\",GLP,PEN,\"mg\",W,[0.25,0.5,1,1.7,2.4],true,\"limited\",WK,[25,45,65],[1349,2599,3799],0,[2.4],0.5),\n D(\"Ozempic\",\"semaglutide\",GLP,\"multi-dose pen injection\",\"mg\",W,[0.25,0.5,1,2],true,\"in_stock\",WK,[25,42,58],[998,1939,2879],25,[],0.75),\n D(\"Trulicity\",\"dulaglutide\",GLP,PEN,\"mg\",W,[0.75,1.5,3,4.5],true,\"in_stock\",WK,[25,40,55],[987,1899,2799],25,[],0.5),\n D(\"Saxenda\",\"liraglutide\",GLP,PEN,\"mg\",DA,[0.6,1.2,1.8,2.4,3],true,\"in_stock\",DY,[25,45,65],[1349,2599,3799],25),\n D(\"Rybelsus\",\"semaglutide\",GLP,TAB,\"mg\",DA,[3,7,14],true,\"in_stock\",DY,[25,45,65],[998,1939,2879],25),\n D(\"Metformin\",\"metformin\",DIA,\"extended-release tablet\",\"mg\",DA,[500,850,1000],true,\"in_stock\",DY,[3.1,5.5,8],[9,16,22]),\n D(\"Glipizide\",\"glipizide\",DIA,TAB,\"mg\",DA,[5,10],true,\"in_stock\",DY,[4,7,10],[11,19,26]),\n D(\"Januvia\",\"sitagliptin\",DIA,TAB,\"mg\",DA,[25,50,100],true,\"in_stock\",DY,[45,85,120],[621,1199,1749]),\n D(\"Jardiance\",\"empagliflozin\",DIA,TAB,\"mg\",DA,[10,25],true,\"in_stock\",DY,[40,75,110],[634,1229,1799]),\n D(\"Farxiga\",\"dapagliflozin\",DIA,TAB,\"mg\",DA,[5,10],true,\"in_stock\",DY,[40,75,110],[617,1199,1749]),\n D(\"Lantus\",\"insulin glargine\",DIA,\"pen injection\",\"units/mL\",DA,[100],true,\"in_stock\",DY,[35,65,95],[292,569,829]),\n D(\"Lisinopril\",\"lisinopril\",CV,TAB,\"mg\",DA,[5,10,20,40],true,\"in_stock\",DY,[4,7,10],[11,19,26]),\n D(\"Amlodipine\",\"amlodipine\",CV,TAB,\"mg\",DA,[2.5,5,10],true,\"in_stock\",DY,[4,7,10],[12,21,28]),\n D(\"Losartan\",\"losartan\",CV,TAB,\"mg\",DA,[25,50,100],true,\"in_stock\",DY,[4,7,10],[13,23,31]),\n D(\"Metoprolol\",\"metoprolol succinate\",CV,\"extended-release tablet\",\"mg\",DA,[25,50,100,200],true,\"in_stock\",DY,[4,7,10],[14,25,34]),\n D(\"Carvedilol\",\"carvedilol\",CV,TAB,\"mg\",DA,[3.125,6.25,12.5,25],true,\"in_stock\",DY,[4,7,10],[13,23,31]),\n D(\"Atorvastatin\",\"atorvastatin\",CV,TAB,\"mg\",DA,[10,20,40,80],true,\"in_stock\",DY,[1.9,3.4,5],[12,21,29]),\n D(\"Rosuvastatin\",\"rosuvastatin\",CV,TAB,\"mg\",DA,[5,10,20,40],true,\"in_stock\",DY,[4,7,10],[15,26,35]),\n D(\"Hydrochlorothiazide\",\"hydrochlorothiazide\",CV,TAB,\"mg\",DA,[12.5,25,50],true,\"in_stock\",DY,[3,5,7],[9,16,21]),\n D(\"Clopidogrel\",\"clopidogrel\",CV,TAB,\"mg\",DA,[75],true,\"in_stock\",DY,[4,7,10],[14,25,34]),\n D(\"Eliquis\",\"apixaban\",CV,TAB,\"mg\",DA,[2.5,5],true,\"in_stock\",DY,[45,85,125],[594,1149,1679]),\n D(\"Xarelto\",\"rivaroxaban\",CV,TAB,\"mg\",DA,[10,15,20],true,\"in_stock\",DY,[45,85,125],[578,1119,1639]),\n D(\"Albuterol HFA\",\"albuterol sulfate\",RESP,\"metered-dose inhaler\",\"mcg\",DA,[90],true,\"backordered\",DY,[15,28,40],[68,129,189]),\n D(\"Fluticasone\",\"fluticasone propionate\",RESP,\"nasal spray\",\"mcg\",DA,[50],true,\"in_stock\",DY,[8,14,20],[24,42,58]),\n D(\"Montelukast\",\"montelukast\",RESP,TAB,\"mg\",DA,[4,5,10],true,\"in_stock\",DY,[4,7,10],[16,28,38]),\n D(\"Symbicort\",\"budesonide/formoterol\",RESP,\"metered-dose inhaler\",\"mcg\",DA,[80,160],true,\"in_stock\",DY,[45,85,125],[349,679,989]),\n D(\"Sertraline\",\"sertraline\",MH,TAB,\"mg\",DA,[25,50,100],true,\"in_stock\",DY,[4,7,10],[13,23,31]),\n D(\"Escitalopram\",\"escitalopram\",MH,TAB,\"mg\",DA,[5,10,20],true,\"in_stock\",DY,[4,7,10],[14,25,34]),\n D(\"Fluoxetine\",\"fluoxetine\",MH,CAP,\"mg\",DA,[10,20,40],true,\"in_stock\",DY,[4,7,10],[12,21,28]),\n D(\"Bupropion XL\",\"bupropion\",MH,\"extended-release tablet\",\"mg\",DA,[150,300],true,\"in_stock\",DY,[6,11,15],[22,39,53]),\n D(\"Trazodone\",\"trazodone\",MH,TAB,\"mg\",DA,[50,100,150],true,\"in_stock\",DY,[4,7,10],[12,21,28]),\n D(\"Duloxetine\",\"duloxetine\",MH,\"delayed-release capsule\",\"mg\",DA,[20,30,60],true,\"in_stock\",DY,[6,11,15],[20,35,48]),\n D(\"Omeprazole\",\"omeprazole\",GI,\"delayed-release capsule\",\"mg\",DA,[20,40],true,\"in_stock\",DY,[4,7,10],[15,26,35]),\n D(\"Pantoprazole\",\"pantoprazole\",GI,\"delayed-release tablet\",\"mg\",DA,[20,40],true,\"in_stock\",DY,[4,7,10],[16,28,38]),\n D(\"Famotidine\",\"famotidine\",GI,TAB,\"mg\",DA,[10,20,40],true,\"in_stock\",DY,[4,7,10],[11,19,26]),\n D(\"Levothyroxine\",\"levothyroxine\",THY,TAB,\"mcg\",DA,[25,50,75,100,125],true,\"in_stock\",DY,[4,7,10],[14,25,34]),\n D(\"Gabapentin\",\"gabapentin\",PAIN,CAP,\"mg\",DA,[100,300,400,600,800],true,\"in_stock\",DY,[5,9,12],[17,30,41]),\n D(\"Meloxicam\",\"meloxicam\",PAIN,TAB,\"mg\",DA,[7.5,15],true,\"in_stock\",DY,[4,7,10],[12,21,28]),\n D(\"Naproxen\",\"naproxen\",PAIN,TAB,\"mg\",DA,[250,375,500],true,\"in_stock\",DY,[4,7,10],[13,23,31]),\n D(\"Prednisone\",\"prednisone\",PAIN,TAB,\"mg\",DA,[1,5,10,20],true,\"in_stock\",DY,[4,7,10],[10,17,23]),\n D(\"Amoxicillin\",\"amoxicillin\",ABX,CAP,\"mg\",DA,[250,500,875],true,\"in_stock\",AB,[5,9,12],[16,28,38]),\n D(\"Azithromycin\",\"azithromycin\",ABX,TAB,\"mg\",DA,[250,500],true,\"in_stock\",AB,[6,11,15],[24,42,58]),\n D(\"Doxycycline\",\"doxycycline hyclate\",ABX,CAP,\"mg\",DA,[50,100],true,\"limited\",AB,[8,14,20],[28,49,67],0,[50]),\n D(\"Cephalexin\",\"cephalexin\",ABX,CAP,\"mg\",DA,[250,500],true,\"in_stock\",AB,[6,11,15],[19,33,45]),\n D(\"Tamsulosin\",\"tamsulosin\",OTH,CAP,\"mg\",DA,[0.4],true,\"in_stock\",DY,[4,7,10],[14,25,34]),\n D(\"Finasteride\",\"finasteride\",OTH,TAB,\"mg\",DA,[1,5],true,\"in_stock\",DY,[4,7,10],[15,26,35]),\n D(\"Sildenafil\",\"sildenafil\",OTH,TAB,\"mg\",DA,[20,25,50,100],true,\"in_stock\",DY,[6,11,15],[18,32,44]),\n D(\"Aspirin\",\"aspirin\",OTC,TAB,\"mg\",DA,[81,325],false,\"in_stock\",DY,[0,0,0],[4,7,9]),\n D(\"Ibuprofen\",\"ibuprofen\",OTC,TAB,\"mg\",DA,[200,400,600,800],false,\"in_stock\",DY,[0,0,0],[6,10,14]),\n D(\"Cetirizine\",\"cetirizine\",OTC,TAB,\"mg\",DA,[5,10],false,\"in_stock\",DY,[0,0,0],[8,14,19]),\n D(\"Loratadine\",\"loratadine\",OTC,TAB,\"mg\",DA,[10],false,\"in_stock\",DY,[0,0,0],[8,14,19]),\n];\n\n// ---- fuzzy drug-name matching ------------------------------------------------\n// \"sepbamd\" and \"Cetpound\" must both surface Zepbound. Raw edit distance is too weak\n// (sepbamd vs zepbound differs by ~4 of 8 characters), so we also compare consonant\n// skeletons under sound-alike folding and keep whichever score is better.\nfunction lev(a: string, b: string): number {\n const m = a.length, n = b.length;\n let prev = Array.from({ length: n + 1 }, (_, j) => j);\n for (let i = 1; i <= m; i++) {\n const cur = [i];\n for (let j = 1; j <= n; j++) {\n cur[j] = Math.min(prev[j] + 1, cur[j - 1] + 1, prev[j - 1] + (a[i - 1] === b[j - 1] ? 0 : 1));\n }\n prev = cur;\n }\n return prev[n];\n}\n\nconst clean = (s: string) => s.toLowerCase().replace(/[^a-z]/g, \"\");\n\nfunction skeletons(w: string): string[] {\n const base = clean(w).replace(/ph/g, \"f\").replace(/ck/g, \"k\").replace(/x/g, \"ks\");\n const out = new Set();\n for (const c of [\"s\", \"k\"]) {\n out.add(base.replace(/z/g, \"s\").replace(/c/g, c).replace(/q/g, \"k\")\n .replace(/y/g, \"i\").replace(/[aeiou]/g, \"\"));\n }\n return [...out];\n}\n\nconst ratio = (a: string, b: string) => (a && b) ? 1 - lev(a, b) / Math.max(a.length, b.length) : 0;\n\n// A skeleton match is weaker evidence than a spelling match, and is now scored as such.\n// Both words have had their vowels stripped, so a short name is down to five or six\n// consonants, and at that length two coincidentally shared letters look like a real match.\n//\n// That is exactly how a real call produced its worst answer. Deepgram heard \"Zepbound\" as\n// \"set down\", whose skeleton \"stdwn\" is two edits from Cetirizine's \"strsn\", scoring a flat\n// 0.600 and clearing the floor on the nose. The agent then asked a caller who wanted a\n// GLP-1 whether they meant an antihistamine. Raw similarity between those two words is\n// 0.300; the entire score came from the skeleton path. Discounting that path restores the\n// gap without squeezing the floor into the 0.600 to 0.625 sliver between \"setdown\" and\n// Amazon's \"Cetpound\", which would have been a threshold with no room to breathe.\nconst SKELETON_WEIGHT = 0.9;\n\nfunction nameScore(spoken: string, d: Drug): number {\n const s = clean(spoken);\n let best = Math.max(ratio(s, clean(d.name)), ratio(s, clean(d.generic)));\n for (const sk of skeletons(spoken)) {\n for (const t of [d.name, d.generic]) {\n for (const tk of skeletons(t)) best = Math.max(best, ratio(sk, tk) * SKELETON_WEIGHT);\n }\n }\n return best;\n}\n\n// A confident wrong suggestion is worse than no suggestion. Under the old flat 0.34\n// filter, keyboard mash came back specific and plausible: \"Qwertyuiop\" surfaced Losartan\n// and \"Zylaxopram\" surfaced Escitalopram, so the agent named a real cardiovascular or\n// psychiatric drug at someone who had said nothing resembling one. Two gates now.\n//\n// BEST_FLOOR is the bar the single strongest candidate must clear before we name anything\n// at all. Below it matchDrug returns no suggestions and didYouMean falls through to its\n// \"nothing resembles that name, ask them to spell it\" branch. 0.6 is the ceiling this\n// scorer allows: Amazon's own \"Cetpound\" test case peaks at 0.625 against Zepbound, while\n// \"Zylaxopram\" peaks at 0.571, so the two sit on either side of 0.6 with nothing between.\n//\n// KEEP_FLOOR and KEEP_MARGIN then decide which of the remaining candidates are close\n// enough to the leader to be worth reading out. The margin is what keeps \"fosinopril\" to\n// Lisinopril alone instead of Lisinopril plus a coincidental Aspirin at exactly 0.5.\nconst NAME_BEST_FLOOR = 0.6, NAME_KEEP_FLOOR = 0.5, NAME_KEEP_MARGIN = 0.9;\n\n// Below BEST_FLOOR there is still a band worth speaking, and the thing that separates it from\n// noise is not the score, it is the MARGIN to the runner-up. A real mishearing lands on one\n// drug and leaves the rest behind. A random string sits roughly equidistant from everything,\n// because edit distance to a word you never said is just a function of length.\n//\n// From the live corpus, leader then runner-up:\n// \"cepham\" Cephalexin 0.500, next 0.333, margin 0.167 <- a real word, worth asking about\n// \"blorfizan\" Ibuprofen 0.600, next 0.450, margin 0.150 <- already clears BEST_FLOOR\n// \"pant\" Lantus 0.500, next 0.450, margin 0.050 <- genuinely ambiguous, stays silent\n// \"Zylaxopram\" Meloxicam 0.514, TIED at 0.514, margin 0 <- noise\n// \"Qwertyuiop\" Xarelto 0.360, TIED at 0.360, margin 0 <- noise\n// \"grunklewax\" Trulicity 0.338, TIED at 0.338, margin 0 <- noise\n//\n// So a floor alone cannot separate these: Zylaxopram outscores cepham. The margin does it\n// cleanly, and every nonsense case in the corpus fails on margin rather than on score.\n// A hit in this band is returned as `tentative` so the caller is asked a softer question and\n// never hears a drug named as though we were confident about it.\nconst NAME_TENTATIVE_FLOOR = 0.45, NAME_TENTATIVE_MARGIN = 0.12;\n\n// Above this score, with this much daylight to the runner-up, the match is certain enough\n// that asking \"did you mean X?\" is itself the wrong answer. On a real call the caller said\n// a word one edit off Zepbound and got asked to confirm it anyway, which is the tic that\n// makes an agent feel deaf. Past these two bars we just answer, and naming the drug in the\n// answer is its own confirmation because the caller hears which one we picked.\n//\n// Both numbers come from the corpus, not from taste. 0.87 sits in the gap between the\n// highest-scoring REAL medication we do not stock (canagliflozin, 0.846 against Farxiga)\n// and the lowest mishearing we want accepted silently (ohzempic, 0.875). That gap is the\n// whole safety argument: no real medication can be silently rewritten into a different one.\n// Citalopram against Escitalopram at 0.833 and lansoprazole against Pantoprazole at 0.833\n// sit under the bar for the same reason, and those are the pairs that would actually hurt.\n//\n// The margin is what holds back a genuine tie. Every mishearing of \"tirzepatide\" scores\n// 0.909 against BOTH Zepbound and Mounjaro, and silently picking one would be indefensible;\n// a 0 margin sends it to confirmation instead. Checked by brute force over every single\n// edit of every catalog name and generic: of the 26,648 variants that clear both bars, zero\n// resolve to the wrong catalog drug.\nconst AUTO_ACCEPT_SCORE = 0.87, AUTO_ACCEPT_MARGIN = 0.15;\n\n// ---- brand names -------------------------------------------------------------\n// Callers say brand names, and a brand name is not a mispronunciation, so the fuzzy\n// scorer has no business guessing at one. Left to itself it put several real brands above\n// the 0.6 floor against the wrong drug: Zyrtec came back Losartan, Claritin came back\n// Losartan, Lexapro came back Eliquis, Prozac came back Omeprazole, Plavix came back\n// Eliquis, Cymbalta came back Wegovy. Answering an allergy question with a blood pressure\n// drug is the worst failure this demo can produce, so brands resolve by table before the\n// scorer ever runs.\n//\n// Two outcomes, and only two. A brand whose molecule we dispense maps to that catalog\n// entry and resolves as an exact match. A brand whose molecule we do not dispense\n// resolves to a flat \"we do not carry it\", because admitting a gap is always better than\n// naming a drug the caller did not ask for.\n//\n// Nothing in either table is a therapeutic substitution. Every mapping is the same\n// molecule under a different label, which is why Lopressor is absent: it is metoprolol\n// tartrate, immediate release, and the catalog stocks the succinate extended-release, so\n// treating them as one label would be a formulation change dressed up as a synonym.\n// Keys are clean()ed, so \"Z-pack\" and \"Toprol XL\" arrive here as zpack and toprolxl.\nconst BRAND_TO_CATALOG: Record = {\n // Diabetes\n glucophage: \"Metformin\", glucotrol: \"Glipizide\",\n // Cardiovascular\n norvasc: \"Amlodipine\", cozaar: \"Losartan\", prinivil: \"Lisinopril\", zestril: \"Lisinopril\",\n toprol: \"Metoprolol\", toprolxl: \"Metoprolol\", coreg: \"Carvedilol\",\n lipitor: \"Atorvastatin\", crestor: \"Rosuvastatin\",\n microzide: \"Hydrochlorothiazide\", hctz: \"Hydrochlorothiazide\", plavix: \"Clopidogrel\",\n // Respiratory and allergy\n ventolin: \"Albuterol HFA\", proair: \"Albuterol HFA\", flonase: \"Fluticasone\",\n singulair: \"Montelukast\",\n // Mental health\n zoloft: \"Sertraline\", lexapro: \"Escitalopram\", prozac: \"Fluoxetine\",\n wellbutrin: \"Bupropion XL\", wellbutrinxl: \"Bupropion XL\", desyrel: \"Trazodone\",\n cymbalta: \"Duloxetine\",\n // Digestive\n prilosec: \"Omeprazole\", protonix: \"Pantoprazole\", pepcid: \"Famotidine\",\n // Thyroid\n synthroid: \"Levothyroxine\", levoxyl: \"Levothyroxine\",\n // Pain and inflammation\n neurontin: \"Gabapentin\", mobic: \"Meloxicam\", aleve: \"Naproxen\", naprosyn: \"Naproxen\",\n deltasone: \"Prednisone\", advil: \"Ibuprofen\", motrin: \"Ibuprofen\",\n // Antibiotics\n amoxil: \"Amoxicillin\", zithromax: \"Azithromycin\", zpack: \"Azithromycin\",\n zpak: \"Azithromycin\", vibramycin: \"Doxycycline\", keflex: \"Cephalexin\",\n // Other prescription\n flomax: \"Tamsulosin\", proscar: \"Finasteride\", propecia: \"Finasteride\",\n viagra: \"Sildenafil\",\n // Over-the-counter\n zyrtec: \"Cetirizine\", claritin: \"Loratadine\",\n};\n\n// Brands whose molecule is genuinely absent from the catalog. Mapped to the generic name\n// so the agent can name what it is we do not have, rather than only rejecting the brand.\n// The controlled substances here are worth a flat refusal on their own merits: a clean\n// \"we do not carry that\" is both the truthful answer and the right compliance answer.\nconst BRAND_NOT_CARRIED: Record = {\n tylenol: \"acetaminophen\", nexium: \"esomeprazole\", coumadin: \"warfarin\",\n zocor: \"simvastatin\", diovan: \"valsartan\", celexa: \"citalopram\",\n paxil: \"paroxetine\", effexor: \"venlafaxine\", lyrica: \"pregabalin\",\n humalog: \"insulin lispro\", ambien: \"zolpidem\", xanax: \"alprazolam\",\n ativan: \"lorazepam\", klonopin: \"clonazepam\", adderall: \"amphetamine salts\",\n vyvanse: \"lisdexamfetamine\", oxycontin: \"oxycodone\",\n percocet: \"oxycodone with acetaminophen\",\n};\n\n// Transcription shapes seen coming out of Deepgram on real calls, pinned so the scorer\n// never gets a vote on them. Only shapes the scorer actually gets wrong belong here. Every\n// other GLP-1 mishearing tried (\"zip bound\", \"sep bound\", \"we go vee\", \"mon jaro\", \"truly\n// city\", \"rebelsus\", \"ozempick\" and the rest) already lands on the right drug on its own,\n// and pinning those would be dead weight that hides a future regression rather than\n// catching it.\n//\n// Deliberately NOT here: \"sepbamd\" and \"Cetpound\". Both are Amazon's own stated test cases\n// and both must keep asking \"did you mean Zepbound?\" rather than resolving silently. They\n// are far enough off that a silent wrong resolution would be unrecoverable, which is the\n// opposite of what pinning them would do.\nconst MISHEARD_TO_CATALOG: Record = {\n // \"Zepbound\" heard as \"set down\" on call 01a0446d, which scored 0.600 against Cetirizine\n // and produced \"did you mean cetirizine, the allergy medicine?\" to a GLP-1 caller.\n setdown: \"Zepbound\",\n // Same call, next turn. Scores 0.778, which is real but under the auto-accept bar, so\n // without this it asks a second time for a word that is one edit from the drug.\n stepbound: \"Zepbound\",\n};\n\n// A typo in any table would silently fall through to the fuzzy scorer, which is the\n// exact behaviour the tables exist to prevent, so it fails loudly at load instead.\nfor (const [table, name] of [[BRAND_TO_CATALOG, \"BRAND_TO_CATALOG\"], [MISHEARD_TO_CATALOG, \"MISHEARD_TO_CATALOG\"]] as const) {\n for (const [key, target] of Object.entries(table)) {\n if (!CATALOG.some((d) => d.name === target)) {\n throw new Error(`${name}[\"${key}\"] points at \"${target}\", which is not a CATALOG name`);\n }\n }\n}\n// An alias that already matches a catalog name or generic exactly is unreachable, because\n// the exact pass runs first. Better to reject it than to leave a line that looks load\n// bearing and is not: \"zep bound\" cleans to \"zepbound\" and needs no entry.\nfor (const [key, table] of [\n ...Object.keys(BRAND_TO_CATALOG).map((k) => [k, \"BRAND_TO_CATALOG\"] as const),\n ...Object.keys(BRAND_NOT_CARRIED).map((k) => [k, \"BRAND_NOT_CARRIED\"] as const),\n ...Object.keys(MISHEARD_TO_CATALOG).map((k) => [k, \"MISHEARD_TO_CATALOG\"] as const),\n]) {\n if (key !== clean(key)) {\n throw new Error(`${table} key \"${key}\" is not clean()ed, so it can never be looked up`);\n }\n if (CATALOG.some((d) => key === clean(d.name) || key === clean(d.generic))) {\n throw new Error(`${table}[\"${key}\"] is unreachable: the exact catalog pass already matches it`);\n }\n}\nconst aliasKeys = [...Object.keys(BRAND_TO_CATALOG), ...Object.keys(BRAND_NOT_CARRIED), ...Object.keys(MISHEARD_TO_CATALOG)];\nfor (const key of aliasKeys) {\n if (aliasKeys.indexOf(key) !== aliasKeys.lastIndexOf(key)) {\n throw new Error(`alias key \"${key}\" appears in more than one alias table`);\n }\n}\n\n// The mishear entries are load bearing in a way the brand entries are not: they are the\n// only reason \"setdown\" and \"stepbound\" resolve at all, and they mask whatever the scorer\n// would have said. The checks above cannot tell us when that stops being true, because they\n// all test the key STRING and never run the scorer. So this runs it.\n//\n// Warn rather than throw. A redundant mishear entry is harmless at runtime, since the table\n// is consulted first and simply agrees with the scorer, and taking the endpoint down mid\n// demo over a tuning change that made things BETTER would be the wrong trade. A typo in the\n// table is a different matter and still throws above.\nfor (const [key, target] of Object.entries(MISHEARD_TO_CATALOG)) {\n const scored = CATALOG.map((d) => ({ d, sc: nameScore(key, d) })).sort((a, b) => b.sc - a.sc);\n const best = scored[0], margin = best.sc - (scored[1]?.sc ?? 0);\n const wouldAutoAccept = best.sc >= AUTO_ACCEPT_SCORE && margin >= AUTO_ACCEPT_MARGIN;\n const stat = `${best.sc.toFixed(3)}, margin ${margin.toFixed(3)}`;\n // Every entry reports, with no silent case. An entry that the scorer merely AGREES with is\n // still load bearing if the scorer would only offer it as a suggestion, because the whole\n // point of pinning \"stepbound\" was to stop it being read back as a question.\n if (best.d.name !== target) {\n console.warn(\n `MISHEARD_TO_CATALOG[\"${key}\"]: LOAD BEARING for correctness. Without it the scorer ` +\n `would say ${best.sc < NAME_BEST_FLOOR ? \"nothing at all\" : best.d.name} (${stat}).`,\n );\n } else if (!wouldAutoAccept) {\n console.warn(\n `MISHEARD_TO_CATALOG[\"${key}\"]: LOAD BEARING for auto-accept. The scorer reaches ` +\n `${target} but only well enough to ask (${stat}), so removing this would bring the ` +\n `\"did you mean\" question back.`,\n );\n } else {\n console.warn(\n `MISHEARD_TO_CATALOG[\"${key}\"]: REDUNDANT. The scorer reaches ${target} and accepts it ` +\n `on its own (${stat}). Safe to delete.`,\n );\n }\n}\n\nexport type DrugMatch = {\n exact: Drug | null;\n suggestions: Drug[];\n // Set only when the caller named a brand we can identify but do not dispense.\n notCarried?: { requested: string; generic: string };\n // Set when the caller named a generic that several products in the catalog share. This is\n // a successful match to more than one thing, not a failed match, and reads differently.\n sharedGeneric?: string;\n // Set when the single suggestion came from the low-confidence band. It is a guess worth\n // putting to the caller, not a match, and it is phrased that way.\n tentative?: boolean;\n};\n\nexport function matchDrug(spoken: unknown): DrugMatch {\n const s = clean(String(spoken ?? \"\"));\n if (!s) return { exact: null, suggestions: [] };\n // A real catalog entry always wins, so no alias key can ever shadow a stocked drug.\n // A product name belongs to exactly one row, so an exact name match is unambiguous.\n const byName = CATALOG.find((d) => s === clean(d.name));\n if (byName) return { exact: byName, suggestions: [] };\n // A generic is not unique, and this used to be the most dangerous line in the file. The\n // old single loop matched name or generic together and returned the first row it hit, so\n // \"semaglutide\" silently became Wegovy while Ozempic and Rybelsus went unmentioned, and\n // \"tirzepatide\" silently became Zepbound over Mounjaro. That is a wrong drug named with\n // total confidence and no opening for the caller to correct it, and it was worse than the\n // misheard version because the exact path never reaches the margin rule that catches\n // \"irzepatide\". A generic shared by several products now asks which one.\n const byGeneric = CATALOG.filter((d) => s === clean(d.generic));\n if (byGeneric.length === 1) return { exact: byGeneric[0], suggestions: [] };\n if (byGeneric.length > 1) {\n return { exact: null, suggestions: byGeneric, sharedGeneric: byGeneric[0].generic };\n }\n const branded = BRAND_TO_CATALOG[s] ?? MISHEARD_TO_CATALOG[s];\n if (branded) {\n const d = CATALOG.find((c) => c.name === branded)!;\n return { exact: d, suggestions: [] };\n }\n const absent = BRAND_NOT_CARRIED[s];\n if (absent) {\n return {\n exact: null, suggestions: [],\n notCarried: { requested: String(spoken ?? \"\").trim(), generic: absent },\n };\n }\n const scored = CATALOG.map((d) => ({ d, sc: nameScore(s, d) })).sort((a, b) => b.sc - a.sc);\n const best = scored[0]?.sc ?? 0;\n if (best < NAME_BEST_FLOOR) {\n // One clear leader well ahead of the field is worth putting to the caller even down here.\n // A tie, at any score, is noise and stays unspoken.\n const lead = best - (scored[1]?.sc ?? 0);\n if (best >= NAME_TENTATIVE_FLOOR && lead >= NAME_TENTATIVE_MARGIN) {\n return { exact: null, suggestions: [scored[0].d], tentative: true };\n }\n return { exact: null, suggestions: [] };\n }\n // Certain enough to just answer. Confirming a word this close reads as not having heard\n // the caller, and the answer names the drug anyway, so they hear which one we picked.\n if (best >= AUTO_ACCEPT_SCORE && best - (scored[1]?.sc ?? 0) >= AUTO_ACCEPT_MARGIN) {\n return { exact: scored[0].d, suggestions: [] };\n }\n const keep = Math.max(NAME_KEEP_FLOOR, best * NAME_KEEP_MARGIN);\n const ranked = scored.filter((r) => r.sc >= keep).slice(0, 3).map((r) => r.d);\n return { exact: null, suggestions: ranked };\n}\n\n// Takes the whole match rather than just the suggestions, so that the not-carried branch\n// cannot be dropped by a call site that only remembered to pass an array.\nexport const didYouMean = (m: DrugMatch) => {\n if (m.notCarried) {\n return {\n error: \"medication_not_carried\",\n medication_requested: m.notCarried.requested,\n generic: m.notCarried.generic,\n did_you_mean: [],\n // Deliberately no pharmacist offer. Not stocking something is a catalog fact, not a\n // clinical question, and routing it to a pharmacist both wastes the transfer and\n // implies there is something to discuss. Stated as a hard check on the test card.\n message: `Amazon Pharmacy does not carry ${m.notCarried.requested}, or its generic ${m.notCarried.generic}. Say plainly that we do not have it. Do not offer a different medication in its place, do not guess at a substitute, and do not offer to connect them to a pharmacist, because this is not a clinical question. Then ask if there is anything else you can help with.`,\n };\n }\n if (m.sharedGeneric) {\n const names = m.suggestions.map((d) => d.name);\n const spokenList = names.length > 1\n ? `${names.slice(0, -1).join(\", \")} and ${names[names.length - 1]}`\n : names[0];\n return {\n error: \"confirm_which_product\",\n generic: m.sharedGeneric,\n did_you_mean: m.suggestions.map((d) => ({ name: d.name, generic: d.generic, category: d.category })),\n message: `${m.sharedGeneric} is the active ingredient in ${intWords(names.length)} products we carry: ${spokenList}. They are different products with different strengths and prices, so do not pick one for the caller. Read the names back and ask which one they are on, then call again with that product name.`,\n };\n }\n const sug = m.suggestions;\n return {\n error: \"confirm_medication_name\",\n did_you_mean: sug.map((d) => ({ name: d.name, generic: d.generic, category: d.category })),\n tentative: m.tentative ?? false,\n message: !sug.length\n ? \"Nothing in the catalog resembles that name. Ask the caller to spell it or say it once more.\"\n : m.tentative\n // A long shot, but a named long shot beats asking someone to spell a word they have\n // already said. Offer it once, softly, and fall back to spelling if they say no.\n ? `Not a close match, but ${sug[0].name} is the nearest thing we carry. Put it to them as a guess rather than a finding, something like \"I'm not sure I caught that, did you mean ${sug[0].name}?\". If they say no, then ask them to spell it. Do not ask them to spell it before offering this.`\n : `That was not an exact match to anything in the catalog. Ask the caller to confirm which one they meant, for example \"did you mean ${sug[0].name}?\", then call again with the confirmed name. Never assume.`,\n };\n};\n\n// ---- day supply --------------------------------------------------------------\n// Weekly injectables dispense in whole pens, so a \"90-day\" request is really an\n// 84-day supply. Snapping happens here so the model never reasons about it.\nexport function resolveSupply(d: Drug, requested: number): number {\n const under = d.supplies.filter((s) => s <= requested);\n return under.length ? Math.max(...under) : Math.min(...d.supplies);\n}\n\nexport function supplyNote(d: Drug, requested: number, actual: number, lang: Lang = \"en\"): string | null {\n if (requested === actual) return null;\n if (lang === \"es\") {\n // Prose, not a wording swap. \"dispensed\" is \"se surte\" here rather than a literal\n // \"se dispensa\", because \"surtir una receta\" is the verb a Mexican pharmacy uses, and\n // an injection pen is a \"pluma\".\n if (d.schedule === \"weekly\" && d.mlPerDose) {\n return `${d.name} se surte en plumas semanales completas, así que el suministro más cercano a ${intWords(requested, \"es\", \"m\")} días es de ${intWords(actual, \"es\", \"m\")} días.`;\n }\n return `El suministro disponible más cercano a ${intWords(requested, \"es\", \"m\")} días es de ${intWords(actual, \"es\", \"m\")} días.`;\n }\n if (d.schedule === \"weekly\" && d.mlPerDose) {\n return `${d.name} is dispensed in whole weekly pens, so the closest supply to ${intWords(requested)} days is ${intWords(actual)} days.`;\n }\n return `The closest available supply to ${intWords(requested)} days is ${intWords(actual)} days.`;\n}\n","deleted",false,"label",[14],"d174b348-9a0a-41bf-9755-a00af31e650f",["D",1787797376557],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","Amazon Pharmacy RFP demo tool server (consolidated)","imageUrl","userId","019ca0d6-7e9d-7708-aea8-6df17bfdceea","pinnedFileId","37350edc-a1be-11f1-bfeb-1607ee4eb77e","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","019ca0d6-7e9d-7708-aea8-700f071be19f","tursoName","d174b3489a0a41bf9755a00af31e650f","tursoHostname","d174b3489a0a41bf9755a00af31e650f-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789534872841],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_9":10,"_105":101},"org",{"_9":10,"_99":-5,"_100":101,"_55":56,"_102":103,"_104":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",138,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","displayName","avatar","https://img.clerk.com/eyJ0eXBlIjoiZGVmYXVsdCIsImlpZCI6Imluc18yTm10cGhQUklwU0tqZHNiOHNnNGd2ZjJURFUiLCJyaWQiOiJ1c2VyXzNBR2hoWFBtaWhlZUdZZlg1TnNKZ2Uwc1YyQyJ9","tier","free","canRestrictVals","profileImageUrl","ctx",{"_82":-5,"_139":140,"_141":142,"_143":144,"_145":42,"_146":42,"_147":42,"_148":32,"_149":-5,"_150":-7,"_151":-7,"_152":63,"_153":42,"_154":42,"_155":-5,"_156":42,"_157":42,"_158":42,"_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"]