[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_256":257,"_258":259},"routes/_app.x.$handle.$name",{"_99":98,"_31":44,"_108":89,"_26":109,"_13":14,"_110":-5,"_11":111,"_112":113,"_114":115,"_116":117,"_118":119,"_20":-5,"_120":-5,"_121":-5,"_122":24,"_123":124,"_125":24,"_126":-5,"_127":-5,"_128":129},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_42":43,"_31":44,"_37":45,"_46":-5,"_47":24,"_48":49,"_50":49,"_51":52,"_53":-5,"_18":54,"_55":56,"_57":58,"_59":-5,"_60":-5,"_61":62,"_63":64,"_65":66,"_67":68,"_69":70,"_71":-5,"_72":73,"_74":73,"_75":76,"_77":76,"_78":76,"_79":73,"_80":73,"_81":82,"_83":84,"_85":86,"_87":24},"version",10,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":-7,"_23":24,"_25":24,"_26":27,"_28":27},"file",{"_29":30,"_31":32,"_33":34,"_35":36,"_37":38,"_39":14,"_40":41,"_23":24},"previous","deleted",false,"deletedHere","path","core/prompt.ts","latestPath","fileId","807ab770-81c2-11f1-ba7e-1607ee4eb77e","name","prompt.ts","fileType","script","content","import type { Metrics } from \"./types.ts\";\n\nexport const SYSTEM_PROMPT = `Tu es un coach en recrutement. Tu analyses un entretien de recrutement mené par un employé de Riot et tu donnes à l'INTERVIEWER un feedback factuel et actionnable sur sa technique d'entretien. Ton seul sujet est l'interviewer : tu n'évalues JAMAIS le candidat et tu ne dois jamais exprimer ou laisser entendre s'il faut l'embaucher.\n\nSTYLE\nRéponds en français (conserve les anglicismes d'usage). Sois direct et bienveillant. Base-toi uniquement sur le transcript : n'invente rien. Ne signale que ce qui mérite de l'être — jamais de contrôles négatifs (\"aucune question discriminatoire\", \"rien à signaler\", \"pas de problème détecté\").\n\nMÉTRIQUES MESURÉES\nLes métriques fournies dans le message utilisateur ont été calculées à partir du transcript et sont exactes. Cite-les quand c'est pertinent. Tu ne dois JAMAIS calculer, estimer, approximer ou inventer un chiffre — ni pourcentage, ni compte, ni durée. Si tu veux exprimer quelque chose de chiffré qui n'est pas dans les métriques fournies, dis-le qualitativement.\n\nFORMAT DU TRANSCRIPT\nChaque ligne est préfixée par un ID stable et un rôle, par exemple :\n [L13] [INTERVIEWER Louis Cibot]: So you'd agree that's the right call, wouldn't you?\nLes rôles sont INTERVIEWER, CANDIDATE, OTHER_INTERNAL, OTHER_EXTERNAL.\n\nRÈGLES D'ANCRAGE\n- Chaque élément que tu émets DOIT citer \"source_line_id\" — l'ID exact d'une ligne [INTERVIEWER].\n- Ne cite jamais une ligne CANDIDATE, OTHER_INTERNAL ou OTHER_EXTERNAL. Ces éléments seront supprimés.\n- \"verbatim_quote\" doit être un extrait exact de cette même ligne, idéalement 5 à 40 mots. Ne recolle jamais plusieurs lignes. Ne paraphrase jamais à l'intérieur de la citation.\n- Tu peux raccourcir une citation avec [...], mais la majorité des mots doivent rester ceux, exacts, de la ligne citée.\n\nCONFIDENTIALITÉ\nCe feedback peut être lu par d'autres personnes que l'interviewer. Avant de citer le transcript :\n- Ne retranscris JAMAIS : les circonstances de départ d'un salarié (rupture conventionnelle, licenciement, difficultés de performance), les rémunérations ou évaluations de salariés Riot, des informations personnelles sur des tiers nommés.\n- Si l'interviewer a divulgué une info de ce type pendant l'entretien, signale-le — c'est important — dans \"improvements\" ou \"risk_flags\" (severity \"medium\"), mais sans répéter le contenu : décris seulement la catégorie (ex : \"détails confidentiels sur le départ d'un salarié\") et remplace le passage sensible dans la citation par [confidentiel], en gardant les mots neutres autour pour que la citation reste reconnaissable.\n- Dans toutes les citations, garde uniquement ce qui illustre le comportement de l'interviewer ; coupe le reste avec [...].\n\nLES 4 PRINCIPES D'UN BON INTERVIEWER CHEZ RIOT\nÉvalue l'entretien sur ces 4 principes, dans cet ordre :\n1. HIGH STANDARDS — L'entretien est exigeant et rigoureux. Cadre l'entretien (agenda, durée annoncés). Pose des questions précises. Creuse les réponses vagues jusqu'à obtenir des exemples et des chiffres (\"What\", \"How\", \"Tell me more\").\n2. NO BLIND SPOTS — L'évaluation est complète et sans biais. Écoute plus qu'il ne parle (cible : ratio de parole de l'interviewer ≤ 0.30 — appuie-toi sur la métrique fournie, ne recalcule rien). Ne souffle pas les réponses (pas de questions orientées ou fermées). N'exprime pas de jugement pendant l'entretien. Explore les zones de doute au lieu de les éviter. Aucune question discriminatoire (âge, famille, grossesse, origine, nationalité, religion, santé, handicap, situation maritale...) — toute ligne touchant une caractéristique protégée va aussi dans \"risk_flags\", même posée gentiment ou en passant.\n3. CONTAGIOUS ENTHUSIASM — L'entretien donne envie de rejoindre Riot. Énergie et conviction. Vend Riot et le rôle, pas seulement de l'évaluation. Laisse de la place aux questions du candidat et y répond avec clarté.\n4. NEED FOR SPEED — L'entretien est efficace. Va droit au but, gère son temps. Annonce des next steps clairs et rapides en fin d'entretien.\n\nSORTIE\nRetourne UNIQUEMENT du JSON valide. Pas de fences markdown, pas de prose avant ou après. Ce schéma exact :\n\n{\n \"summary\": \"3 phrases max sur la conduite de l'entretien. Aucun signal d'embauche.\",\n \"score\": 7,\n \"principles\": [\n { \"principle\": \"HIGH STANDARDS\", \"verdict\": \"✅\", \"evidence\": [ { \"source_line_id\": \"L11\", \"verbatim_quote\": \"...\" } ] },\n { \"principle\": \"NO BLIND SPOTS\", \"verdict\": \"🟡\", \"evidence\": [ { \"source_line_id\": \"L13\", \"verbatim_quote\": \"...\" } ] },\n { \"principle\": \"CONTAGIOUS ENTHUSIASM\", \"verdict\": \"✅\", \"evidence\": [] },\n { \"principle\": \"NEED FOR SPEED\", \"verdict\": \"❌\", \"evidence\": [ { \"source_line_id\": \"L40\", \"verbatim_quote\": \"...\" } ] }\n ],\n \"strengths\": [\n { \"point\": \"Ce qu'il a bien fait\", \"source_line_id\": \"L11\", \"verbatim_quote\": \"...\" }\n ],\n \"improvements\": [\n { \"point\": \"Ce qu'il faut changer\",\n \"why_it_matters\": \"Le coût concret de faire comme ça\",\n \"source_line_id\": \"L13\",\n \"verbatim_quote\": \"...\",\n \"suggested_alternative\": \"Au lieu de X, essaie Y — une meilleure formulation de cette question précise\" }\n ],\n \"risk_flags\": [\n { \"severity\": \"high\", \"issue\": \"Question sur les projets de famille\", \"source_line_id\": \"L27\", \"verbatim_quote\": \"...\" }\n ],\n \"priority_action\": \"L'action numéro 1, concrète, pour le prochain entretien.\"\n}\n\n- \"score\" : note globale sur 10 de la conduite de l'entretien (jamais du candidat).\n- \"principles\" : exactement les 4 principes, dans l'ordre, verdict \"✅\" (solide), \"🟡\" (perfectible) ou \"❌\" (problème net), avec 1 à 2 citations d'appui chacun quand une ligne pertinente existe (tableau \"evidence\" vide sinon).\n- Émets au plus 2 \"strengths\" et au plus 2 \"improvements\" — les plus importants seulement.\n- \"risk_flags\" UNIQUEMENT quand une ligne touche clairement une caractéristique protégée ou une divulgation confidentielle — un tableau vide est la réponse correcte et fréquente. \"severity\" vaut \"high\" ou \"medium\".\n- Si l'entretien est bien mené, dis-le dans \"summary\" et garde des \"improvements\" mineurs ; ne fabrique pas de critique.`;\n\nexport function buildUserMessage(metrics: Metrics, interviewerName: string, transcriptText: string): string {\n const lines = [\n \"## Measured metrics (these are facts — do not recompute or estimate any number)\",\n `Interviewer talk ratio: ${metrics.interviewer_talk_ratio}`,\n `Candidate talk ratio: ${metrics.candidate_talk_ratio}`,\n `Interviewer turns: ${metrics.interviewer_turns}`,\n `Interviewer questions asked: ${metrics.interviewer_questions}`,\n `Average words per interviewer turn: ${metrics.avg_words_per_interviewer_turn}`,\n `Longest interviewer monologue (words): ${metrics.longest_interviewer_monologue_words}`,\n ];\n if (metrics.interview_span_minutes !== null) {\n lines.push(`Interview length (minutes): ${metrics.interview_span_minutes}`);\n }\n lines.push(\"\", \"## Interviewer\", interviewerName, \"\", \"## Transcript\", transcriptText);\n return lines.join(\"\\n\");\n}\n","createdAt",["D",1784280780409],"branchVersion","branchId","86c684dc-fb6b-40db-85f4-e12036ab3e09","id","f0ac4d89-2364-4c3c-a35d-9e744274b9f6","interview-coach",["D",1784280622128],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","Fathom interview → Claude coaching (interviewer) → Notion DB","imageUrl","4704801c-db4b-4f2a-8ad7-932a7d734595","pinnedFileId","af5858b8-81c2-11f1-bffa-1607ee4eb77e","pinPreview",true,"pinMigrated","screenshotVersion","likeCount",0,"orgId","019f6f4e-7661-7305-979e-3c621fe95d18","tursoName","f0ac4d8923644c3ca35d9e744274b9f6","tursoHostname","f0ac4d8923644c3ca35d9e744274b9f6-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789527031073],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_99":98,"_107":103},"org",{"_99":98,"_100":101,"_102":103,"_18":54,"_104":105,"_106":24},"branch",{"_13":88,"_31":89,"_42":41,"_90":43,"_91":-5,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":44,"_97":98},"isOwned",18,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","lcibot","handle","displayName","Louis Cibot","avatar","https://lh3.googleusercontent.com/a/ACg8ocIJUslEt6CPs22Hts-0n6afiQpKVLbtvn-iRhJvA1gVqdthrY0=s96-c","tier","free","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_42":43,"_31":44,"_37":251,"_46":-5,"_47":24,"_48":49,"_50":49,"_51":52,"_53":-5,"_18":54,"_55":56,"_57":58,"_59":-5,"_60":-5,"_61":62,"_63":64,"_65":66,"_67":68,"_69":252,"_71":-5,"_72":73,"_74":73,"_75":76,"_77":76,"_78":76,"_79":73,"_80":73,"_81":253,"_83":254,"_85":255,"_87":24},"tree",[158,168],"files",[158,159,160,161,162,163,164,165,166,167,168,169],"counts",{"_148":62,"_149":145,"_150":62,"_151":152,"_153":154},"contents",[130,131],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},{"_42":142,"_31":143,"_29":144,"_135":-5,"_40":41,"_39":145,"_137":58,"_37":146,"_13":62,"_33":139,"_35":-5,"_23":24,"_140":-5,"_26":147},{"_42":132,"_31":133,"_29":134,"_135":-5,"_40":41,"_39":136,"_137":58,"_37":138,"_13":62,"_33":139,"_35":-5,"_23":24,"_140":-5,"_26":141},"d9587335-a3d2-4004-967b-cc7b073acb08","glue","54c03fa6-81ac-448f-a060-75ffa88db4d9","parentId",9,"latest",["D",1784280744388],"directory","label",[133],"87b75345-444c-40b8-b4a0-eeea89d08d77","core","01fedcb8-c298-46eb-adb9-de5f71997ea1",1,["D",1784280644634],[143],"pulls","branches","forks","envVars",{"_155":145,"_157":62},"connections",{"_155":62,"_156":62},"count","unfilledCount","emptyCount",{"_42":142,"_31":143,"_29":144,"_135":-5,"_40":41,"_39":145,"_137":58,"_37":248,"_13":62,"_33":139,"_35":-5,"_23":24,"_140":-5,"_175":249,"_26":250},{"_42":240,"_31":241,"_29":242,"_135":144,"_40":41,"_39":243,"_137":58,"_37":244,"_13":62,"_33":34,"_35":245,"_23":24,"_140":-5,"_175":246,"_26":247},{"_42":232,"_31":233,"_29":234,"_135":144,"_40":41,"_39":235,"_137":58,"_37":236,"_13":62,"_33":34,"_35":237,"_23":24,"_140":-5,"_175":238,"_26":239},{"_42":224,"_31":225,"_29":226,"_135":144,"_40":41,"_39":227,"_137":58,"_37":228,"_13":62,"_33":34,"_35":229,"_23":24,"_140":-5,"_175":230,"_26":231},{"_42":216,"_31":217,"_29":218,"_135":144,"_40":41,"_39":219,"_137":58,"_37":220,"_13":62,"_33":34,"_35":221,"_23":24,"_140":-5,"_175":222,"_26":223},{"_42":212,"_31":32,"_29":30,"_135":144,"_40":41,"_39":14,"_137":24,"_37":213,"_13":62,"_33":34,"_35":36,"_23":24,"_140":-5,"_175":214,"_26":215},{"_42":204,"_31":205,"_29":206,"_135":144,"_40":41,"_39":207,"_137":58,"_37":208,"_13":62,"_33":34,"_35":209,"_23":24,"_140":-5,"_175":210,"_26":211},{"_42":196,"_31":197,"_29":198,"_135":144,"_40":41,"_39":199,"_137":58,"_37":200,"_13":62,"_33":34,"_35":201,"_23":24,"_140":-5,"_175":202,"_26":203},{"_42":189,"_31":190,"_29":191,"_135":144,"_40":41,"_39":145,"_137":58,"_37":192,"_13":62,"_33":34,"_35":193,"_23":24,"_140":-5,"_175":194,"_26":195},{"_42":181,"_31":182,"_29":183,"_135":144,"_40":41,"_39":184,"_137":58,"_37":185,"_13":62,"_33":34,"_35":186,"_23":24,"_140":-5,"_175":187,"_26":188},{"_42":132,"_31":133,"_29":134,"_135":-5,"_40":41,"_39":136,"_137":58,"_37":178,"_13":62,"_33":139,"_35":-5,"_23":24,"_140":-5,"_175":179,"_26":180},{"_42":170,"_31":171,"_29":172,"_135":134,"_40":41,"_39":136,"_137":58,"_37":173,"_13":62,"_33":34,"_35":174,"_23":24,"_140":-5,"_175":176,"_26":177},"1eaa13c5-8474-496d-889c-2989a7c34ada","env.ts","6b002ccc-81c2-11f1-acbc-1607ee4eb77e",["D",1784280744388],"export function env(key: string): string | undefined {\n const deno = (globalThis as any).Deno;\n if (deno?.env?.get) return deno.env.get(key);\n return (globalThis as any).process?.env?.[key];\n}\n\nexport function requireEnv(key: string): string {\n const value = env(key);\n if (!value) throw new Error(`Missing required environment variable: ${key}`);\n return value;\n}\n\nexport function isDryRun(): boolean {\n return (env(\"DRY_RUN\") ?? \"true\").toLowerCase() !== \"false\";\n}\n","children",[],[133,171],["D",1784280744388],[169],[133],"0d278a26-f187-448b-8b73-7ab04be04858","validate.ts","4f082a2e-81c2-11f1-83e5-1607ee4eb77e",5,["D",1784280697463],"import type { Citation, Coaching, DroppedItem, LabelledLine, Principle } from \"./types.ts\";\n\nconst QUOTE_TOKEN_OVERLAP_THRESHOLD = 0.6;\nconst MIN_TOKENS_FOR_OVERLAP_CHECK = 4;\nconst PRINCIPLE_VERDICTS = new Set([\"✅\", \"🟡\", \"❌\"]);\n\nexport function stripFences(raw: unknown): string {\n return String(raw ?? \"\")\n .trim()\n .replace(/^```(?:json)?\\s*/i, \"\")\n .replace(/```\\s*$/, \"\")\n .trim();\n}\n\nfunction normalizeQuote(text: unknown): string {\n return String(text ?? \"\")\n .normalize(\"NFKD\") // decompose accents; the ASCII filter below drops the marks\n .toLowerCase()\n .replace(/[^a-z0-9@.]+/g, \" \")\n .replace(/\\s+/g, \" \")\n .trim();\n}\n\nfunction significantTokens(text: unknown): string[] {\n return normalizeQuote(text).split(\" \").filter((token) => token.length > 2);\n}\n\nexport function quoteLooksRelated(quote: unknown, sourceText: unknown): boolean {\n const q = normalizeQuote(quote);\n const s = normalizeQuote(sourceText);\n if (!q || !s) return false;\n if (s.includes(q) || q.includes(s)) return true;\n\n const quoteTokens = significantTokens(quote);\n if (quoteTokens.length < MIN_TOKENS_FOR_OVERLAP_CHECK) return false;\n const sourceTokens = new Set(significantTokens(sourceText));\n const hits = quoteTokens.filter((token) => sourceTokens.has(token)).length;\n return hits / quoteTokens.length >= QUOTE_TOKEN_OVERLAP_THRESHOLD;\n}\n\nfunction normalizeLineId(lineId: unknown): string {\n return String(lineId ?? \"\").trim().replace(/^\\[|\\]$/g, \"\").toUpperCase();\n}\n\nexport function validateCoaching(rawText: string, labelled: LabelledLine[]): Coaching {\n let parsed: any;\n try {\n parsed = JSON.parse(stripFences(rawText));\n } catch (error) {\n throw new Error(\n `Claude did not return valid JSON: ${(error as Error).message}\\n---\\n${String(rawText).slice(0, 500)}`\n );\n }\n\n const byId = new Map(labelled.map((line) => [line.id.toUpperCase(), line]));\n const dropped: DroppedItem[] = [];\n\n const keep = (items: unknown, bucket: string): T[] =>\n (Array.isArray(items) ? items : []).filter((item: any) => {\n const id = normalizeLineId(item.source_line_id);\n const line = byId.get(id);\n if (!line) {\n dropped.push({ bucket, item, reason: \"unknown_line_id\" });\n return false;\n }\n if (line.role !== \"INTERVIEWER\") {\n dropped.push({ bucket, item, reason: \"not_an_interviewer_line\" });\n return false;\n }\n if (!quoteLooksRelated(item.verbatim_quote, line.text)) {\n dropped.push({ bucket, item, reason: \"quote_not_in_line\" });\n return false;\n }\n item.source_line_id = id;\n return true;\n });\n\n const principles: Principle[] = (Array.isArray(parsed.principles) ? parsed.principles : [])\n .map((p: any) => ({\n principle: String(p.principle ?? \"\").trim(),\n verdict: PRINCIPLE_VERDICTS.has(p.verdict) ? p.verdict : \"🟡\",\n evidence: keep(p.evidence, \"principles\"),\n }))\n .filter((p: Principle) => p.principle);\n\n const score = Number.isFinite(parsed.score)\n ? Math.min(10, Math.max(0, Math.round(parsed.score)))\n : null;\n\n return {\n summary: String(parsed.summary ?? \"\").trim(),\n score,\n principles,\n strengths: keep(parsed.strengths, \"strengths\"),\n improvements: keep(parsed.improvements, \"improvements\"),\n risk_flags: keep(parsed.risk_flags, \"risk_flags\"),\n priority_action: String(parsed.priority_action ?? \"\").trim(),\n dropped,\n };\n}\n",[],[143,182],"14d0463c-bfb3-40ff-9846-dc002b60cc8b","types.ts","2f8aff1e-81c2-11f1-9145-1607ee4eb77e",["D",1784280644634],"export interface Speaker {\n display_name?: string;\n email?: string;\n matched_calendar_invitee_email?: string;\n matched_calendar_invitee?: { email?: string };\n}\n\nexport interface TranscriptEntry {\n speaker?: Speaker;\n text?: string;\n timestamp?: string;\n}\n\nexport interface Meeting {\n recording_id?: string | number;\n recording_start_time?: string;\n created_at?: string;\n url?: string;\n share_url?: string;\n transcript?: TranscriptEntry[];\n}\n\nexport interface Participant {\n email: string;\n name: string;\n words: number;\n}\n\nexport type IdentifyResult =\n | { ok: false; reason: \"transcript_too_short\" | \"no_internal_speaker\" | \"no_external_speaker\" }\n | { ok: true; interviewer: Participant; candidate: Participant; otherInternal: Participant[] };\n\nexport type Role = \"INTERVIEWER\" | \"CANDIDATE\" | \"OTHER_INTERNAL\" | \"OTHER_EXTERNAL\";\n\nexport interface LabelledLine {\n id: string;\n role: Role;\n email: string;\n name: string;\n text: string;\n line: string;\n}\n\nexport interface Metrics {\n interviewer_talk_ratio: number;\n candidate_talk_ratio: number;\n interviewer_turns: number;\n interviewer_questions: number;\n avg_words_per_interviewer_turn: number;\n longest_interviewer_monologue_words: number;\n interview_span_minutes: number | null;\n}\n\nexport interface Citation {\n source_line_id: string;\n verbatim_quote: string;\n}\n\nexport interface Strength extends Citation {\n point: string;\n}\n\nexport interface Improvement extends Citation {\n point: string;\n why_it_matters?: string;\n suggested_alternative?: string;\n}\n\nexport interface RiskFlag extends Citation {\n severity: string;\n issue: string;\n}\n\nexport interface Principle {\n principle: string;\n verdict: string;\n evidence: Citation[];\n}\n\nexport interface DroppedItem {\n bucket: string;\n item: unknown;\n reason: \"unknown_line_id\" | \"not_an_interviewer_line\" | \"quote_not_in_line\";\n}\n\nexport interface Coaching {\n summary: string;\n score: number | null;\n principles: Principle[];\n strengths: Strength[];\n improvements: Improvement[];\n risk_flags: RiskFlag[];\n priority_action: string;\n dropped: DroppedItem[];\n}\n\nexport interface RenderContext {\n interviewerName: string;\n metrics: Metrics;\n fathomUrl: string;\n}\n",[],[143,190],"c012561e-229b-4daa-9415-54270e3f97a4","svix.ts","5f532e10-81c2-11f1-b54e-1607ee4eb77e",7,["D",1784280724802],"const TOLERANCE_SECONDS = 300;\n\nexport interface SvixHeaders {\n id: string | null;\n timestamp: string | null;\n signature: string | null;\n}\n\nexport type SvixResult =\n | { ok: true }\n | { ok: false; reason: \"missing_headers\" | \"bad_timestamp\" | \"timestamp_out_of_tolerance\" | \"bad_signature\" };\n\nfunction base64ToBytes(b64: string): Uint8Array {\n const bin = atob(b64);\n const bytes = new Uint8Array(bin.length);\n for (let i = 0; i < bin.length; i++) bytes[i] = bin.charCodeAt(i);\n return bytes;\n}\n\nfunction bytesToBase64(bytes: Uint8Array): string {\n let bin = \"\";\n for (const b of bytes) bin += String.fromCharCode(b);\n return btoa(bin);\n}\n\n// Constant-time comparison. The length guard is deliberate: comparing lengths\n// leaks nothing useful, while looping over mismatched lengths would.\nfunction timingSafeEqual(a: Uint8Array, b: Uint8Array): boolean {\n if (a.length !== b.length) return false;\n let diff = 0;\n for (let i = 0; i < a.length; i++) diff |= a[i] ^ b[i];\n return diff === 0;\n}\n\nexport async function verifySvixSignature(\n secret: string,\n rawBody: string,\n headers: SvixHeaders,\n nowSeconds: number = Math.floor(Date.now() / 1000),\n): Promise {\n const { id, timestamp, signature } = headers;\n if (!id || !timestamp || !signature) return { ok: false, reason: \"missing_headers\" };\n\n if (!/^\\d+$/.test(timestamp.trim())) return { ok: false, reason: \"bad_timestamp\" };\n const ts = Number(timestamp.trim());\n if (Math.abs(nowSeconds - ts) > TOLERANCE_SECONDS) {\n return { ok: false, reason: \"timestamp_out_of_tolerance\" };\n }\n\n const key = base64ToBytes(secret.replace(/^whsec_/, \"\"));\n const cryptoKey = await crypto.subtle.importKey(\n \"raw\",\n key,\n { name: \"HMAC\", hash: \"SHA-256\" },\n false,\n [\"sign\"],\n );\n const signedContent = `${id}.${ts}.${rawBody}`;\n const mac = await crypto.subtle.sign(\"HMAC\", cryptoKey, new TextEncoder().encode(signedContent));\n const expected = new TextEncoder().encode(bytesToBase64(new Uint8Array(mac)));\n\n // webhook-signature may carry several space-separated entries like \"v1,\"\n for (const entry of signature.split(\" \")) {\n const candidate = entry.includes(\",\") ? entry.slice(entry.indexOf(\",\") + 1) : entry;\n if (timingSafeEqual(new TextEncoder().encode(candidate), expected)) return { ok: true };\n }\n return { ok: false, reason: \"bad_signature\" };\n}\n",[],[143,197],"a5db2322-903b-4efb-9006-de04d3c86b24","render.ts","59f9eecc-81c2-11f1-9c86-1607ee4eb77e",6,["D",1784280715813],"import type { Coaching, RenderContext } from \"./types.ts\";\n\nexport function renderFeedback(coaching: Coaching, context: RenderContext): string {\n const parts: string[] = [];\n parts.push(\"**🤖 Coaching IA de l'interviewer**\");\n parts.push(\n `_Feedback sur la conduite de cet entretien par **${context.interviewerName}**. ` +\n `Ce n'est pas une évaluation du candidat et ne comporte aucune recommandation d'embauche._`\n );\n parts.push(\"\");\n parts.push(coaching.summary);\n if (coaching.score != null) {\n parts.push(\"\");\n parts.push(`**Note globale : ${coaching.score}/10**`);\n }\n\n if (coaching.risk_flags.length) {\n parts.push(\"\");\n parts.push(\"### ⚠️ Signalements\");\n for (const flag of coaching.risk_flags) {\n parts.push(`- **${flag.severity.toUpperCase()}** — ${flag.issue}`);\n parts.push(` > \"${flag.verbatim_quote}\" _(${flag.source_line_id})_`);\n }\n }\n\n const principles = coaching.principles || [];\n if (principles.length) {\n parts.push(\"\");\n parts.push(\"### Évaluation des 4 principes\");\n for (const p of principles) {\n parts.push(`- ${p.verdict} **${p.principle}**`);\n for (const ev of p.evidence || []) {\n parts.push(` > \"${ev.verbatim_quote}\" _(${ev.source_line_id})_`);\n }\n }\n }\n\n if (coaching.strengths.length) {\n parts.push(\"\");\n parts.push(\"### Points forts\");\n for (const item of coaching.strengths) {\n parts.push(`- ${item.point}`);\n parts.push(` > \"${item.verbatim_quote}\" _(${item.source_line_id})_`);\n }\n }\n\n if (coaching.improvements.length) {\n parts.push(\"\");\n parts.push(\"### Axes d'amélioration\");\n for (const item of coaching.improvements) {\n parts.push(`- **${item.point}** — ${item.why_it_matters || \"\"}`.trimEnd());\n parts.push(` > \"${item.verbatim_quote}\" _(${item.source_line_id})_`);\n if (item.suggested_alternative) parts.push(` Essaie plutôt : _\"${item.suggested_alternative}\"_`);\n }\n }\n\n if (coaching.priority_action) {\n parts.push(\"\");\n parts.push(`**🎯 Action prioritaire pour le prochain entretien :** ${coaching.priority_action}`);\n }\n\n parts.push(\"\");\n parts.push(\n `_Mesuré : l'interviewer a prononcé ${Math.round(context.metrics.interviewer_talk_ratio * 100)} % des mots ` +\n `et posé ${context.metrics.interviewer_questions} questions._`\n );\n if (context.fathomUrl) parts.push(`_Enregistrement : ${context.fathomUrl}_`);\n\n return parts.join(\"\\n\");\n}\n",[],[143,205],"a1e04e6f-31f3-4471-b1e9-bfe6ab76f6f4",["D",1784280780409],[],[143,32],"89e8e41d-5c79-4691-8aa7-08816d3ab1f0","participants.ts","35d7d716-81c2-11f1-896f-1607ee4eb77e",3,["D",1784280655207],"import { isInternal, speakerEmail, countWords } from \"./lib.ts\";\nimport type { IdentifyResult, Meeting, Participant } from \"./types.ts\";\n\nexport const MIN_TRANSCRIPT_ENTRIES = 10;\n\nexport function identifyParticipants(meeting: Meeting | null | undefined): IdentifyResult {\n const transcript = Array.isArray(meeting && meeting.transcript) ? meeting!.transcript! : [];\n if (transcript.length < MIN_TRANSCRIPT_ENTRIES) {\n return { ok: false, reason: \"transcript_too_short\" };\n }\n\n const byEmail = new Map();\n for (const entry of transcript) {\n const email = speakerEmail(entry);\n if (!email) continue;\n const existing = byEmail.get(email) || {\n email,\n name: (entry.speaker && entry.speaker.display_name) || \"Unknown\",\n words: 0,\n };\n existing.words += countWords(entry.text);\n byEmail.set(email, existing);\n }\n\n const speakers = [...byEmail.values()];\n const byWordsDesc = (a: Participant, b: Participant) => b.words - a.words;\n const internal = speakers.filter((s) => isInternal(s.email)).sort(byWordsDesc);\n const external = speakers.filter((s) => !isInternal(s.email)).sort(byWordsDesc);\n\n if (internal.length === 0) return { ok: false, reason: \"no_internal_speaker\" };\n if (external.length === 0) return { ok: false, reason: \"no_external_speaker\" };\n\n return {\n ok: true,\n interviewer: internal[0],\n candidate: external[0],\n otherInternal: internal.slice(1),\n };\n}\n",[],[143,217],"414d9a48-4f3e-4f03-89f6-c38d5f2590a4","notion-blocks.ts","69ae91ec-81c2-11f1-9851-1607ee4eb77e",8,["D",1784280742135],"// Converts the rendered French Markdown into Notion blocks.\n// Supported subset (all that render.ts emits): ### headings, \"- \" bullets,\n// \"> \" quotes (possibly indented), plain paragraphs, and inline **bold** / _italic_.\n\nconst MAX_TEXT_LENGTH = 2000; // Notion limit per rich_text element\n\nexport interface RichText {\n type: \"text\";\n text: { content: string };\n annotations?: { bold?: boolean; italic?: boolean };\n}\n\nexport interface NotionBlock {\n object: \"block\";\n type: string;\n [key: string]: unknown;\n}\n\nfunction chunk(content: string): string[] {\n const chunks: string[] = [];\n for (let i = 0; i < content.length; i += MAX_TEXT_LENGTH) {\n chunks.push(content.slice(i, i + MAX_TEXT_LENGTH));\n }\n return chunks.length ? chunks : [\"\"];\n}\n\nexport function parseInline(text: string): RichText[] {\n const parts: RichText[] = [];\n // Tokenize on **bold** and _italic_ spans; everything else is plain.\n const pattern = /(\\*\\*[^*]+\\*\\*|_[^_]+_)/g;\n let lastIndex = 0;\n for (const match of text.matchAll(pattern)) {\n const index = match.index ?? 0;\n if (index > lastIndex) {\n for (const c of chunk(text.slice(lastIndex, index))) {\n parts.push({ type: \"text\", text: { content: c } });\n }\n }\n const token = match[0];\n if (token.startsWith(\"**\")) {\n for (const c of chunk(token.slice(2, -2))) {\n parts.push({ type: \"text\", text: { content: c }, annotations: { bold: true } });\n }\n } else {\n for (const c of chunk(token.slice(1, -1))) {\n parts.push({ type: \"text\", text: { content: c }, annotations: { italic: true } });\n }\n }\n lastIndex = index + token.length;\n }\n if (lastIndex < text.length) {\n for (const c of chunk(text.slice(lastIndex))) {\n parts.push({ type: \"text\", text: { content: c } });\n }\n }\n return parts.length ? parts : [{ type: \"text\", text: { content: \"\" } }];\n}\n\nfunction block(type: string, text: string): NotionBlock {\n return { object: \"block\", type, [type]: { rich_text: parseInline(text) } };\n}\n\nexport function markdownToNotionBlocks(markdown: string): NotionBlock[] {\n const blocks: NotionBlock[] = [];\n for (const rawLine of markdown.split(\"\\n\")) {\n const line = rawLine.trim();\n if (!line) continue;\n if (line.startsWith(\"### \")) {\n blocks.push(block(\"heading_3\", line.slice(4)));\n } else if (line.startsWith(\"> \")) {\n blocks.push(block(\"quote\", line.slice(2)));\n } else if (line.startsWith(\"- \")) {\n blocks.push(block(\"bulleted_list_item\", line.slice(2)));\n } else {\n blocks.push(block(\"paragraph\", line));\n }\n }\n return blocks;\n}\n",[],[143,225],"e3ff3f8f-9ee2-42a0-896a-7b0f11aff754","metrics.ts","41c79fac-81c2-11f1-8bfd-1607ee4eb77e",4,["D",1784280675234],"import { speakerEmail, countWords, timestampToSeconds, round, isInternal } from \"./lib.ts\";\nimport type { LabelledLine, Metrics, Role, TranscriptEntry } from \"./types.ts\";\n\ninterface Turn {\n email: string;\n words: number;\n entries: TranscriptEntry[];\n}\n\nexport function toTurns(transcript: TranscriptEntry[]): Turn[] {\n const turns: Turn[] = [];\n for (const entry of transcript) {\n const email = speakerEmail(entry);\n const words = countWords(entry.text);\n const last = turns[turns.length - 1];\n if (last && last.email === email) {\n last.words += words;\n last.entries.push(entry);\n } else {\n turns.push({ email, words, entries: [entry] });\n }\n }\n return turns;\n}\n\nexport function labelTranscript(\n transcript: TranscriptEntry[],\n interviewerEmail: string,\n candidateEmail: string,\n): LabelledLine[] {\n return transcript.map((entry, index) => {\n const email = speakerEmail(entry);\n let role: Role;\n if (email === interviewerEmail) role = \"INTERVIEWER\";\n else if (email === candidateEmail) role = \"CANDIDATE\";\n else if (isInternal(email)) role = \"OTHER_INTERNAL\";\n else role = \"OTHER_EXTERNAL\";\n\n const id = `L${index + 1}`;\n const name = (entry.speaker && entry.speaker.display_name) || \"Unknown\";\n const text = String(entry.text ?? \"\");\n return { id, role, email, name, text, line: `[${id}] [${role} ${name}]: ${text}` };\n });\n}\n\nexport function computeMetrics(\n transcript: TranscriptEntry[],\n interviewerEmail: string,\n candidateEmail: string,\n): Metrics {\n const entries = Array.isArray(transcript) ? transcript : [];\n\n let interviewerWords = 0;\n let candidateWords = 0;\n let totalWords = 0;\n let interviewerQuestions = 0;\n\n for (const entry of entries) {\n const email = speakerEmail(entry);\n const words = countWords(entry.text);\n totalWords += words;\n if (email === interviewerEmail) {\n interviewerWords += words;\n if (String(entry.text ?? \"\").includes(\"?\")) interviewerQuestions += 1;\n } else if (email === candidateEmail) {\n candidateWords += words;\n }\n }\n\n const interviewerTurns = toTurns(entries).filter((t) => t.email === interviewerEmail);\n const longestMonologue = interviewerTurns.reduce((max, t) => Math.max(max, t.words), 0);\n\n const first = entries.length ? timestampToSeconds(entries[0].timestamp) : null;\n const last = entries.length ? timestampToSeconds(entries[entries.length - 1].timestamp) : null;\n const span = first !== null && last !== null ? Math.round((last - first) / 60) : null;\n\n return {\n interviewer_talk_ratio: totalWords ? round(interviewerWords / totalWords, 2) : 0,\n candidate_talk_ratio: totalWords ? round(candidateWords / totalWords, 2) : 0,\n interviewer_turns: interviewerTurns.length,\n interviewer_questions: interviewerQuestions,\n avg_words_per_interviewer_turn: interviewerTurns.length\n ? round(interviewerWords / interviewerTurns.length, 1)\n : 0,\n longest_interviewer_monologue_words: longestMonologue,\n interview_span_minutes: span,\n };\n}\n",[],[143,233],"9e574286-21e0-4ed3-9d03-3816c875a57d","lib.ts","3304c904-81c2-11f1-9c86-1607ee4eb77e",2,["D",1784280650418],"import type { TranscriptEntry } from \"./types.ts\";\n\nexport const RIOT_EMAIL_DOMAIN = \"tryriot.com\";\n\nexport function normalizeEmail(email: unknown): string {\n return String(email ?? \"\").trim().toLowerCase();\n}\n\nexport function isInternal(email: unknown): boolean {\n const normalized = normalizeEmail(email);\n const atCount = (normalized.match(/@/g) || []).length;\n if (atCount !== 1) return false;\n return normalized.split(\"@\").pop() === RIOT_EMAIL_DOMAIN;\n}\n\nexport function speakerEmail(entry: TranscriptEntry | null | undefined): string {\n const speaker = (entry && entry.speaker) || {};\n return normalizeEmail(\n speaker.matched_calendar_invitee_email ||\n speaker.email ||\n (speaker.matched_calendar_invitee && speaker.matched_calendar_invitee.email)\n );\n}\n\nexport function countWords(text: unknown): number {\n return String(text ?? \"\").trim().split(/\\s+/).filter(Boolean).length;\n}\n\nexport function timestampToSeconds(ts: unknown): number | null {\n const raw = String(ts ?? \"\").trim();\n if (!raw) return null;\n const segments = raw.split(\":\");\n if (segments.some((seg) => !/^\\d+$/.test(seg))) return null;\n const parts = segments.map(Number);\n if (parts.length === 3) return parts[0] * 3600 + parts[1] * 60 + parts[2];\n if (parts.length === 2) return parts[0] * 60 + parts[1];\n return null;\n}\n\nexport function round(n: number, dp: number): number {\n const factor = 10 ** dp;\n return Math.round(n * factor) / factor;\n}\n",[],[143,241],["D",1784280644634],[159,160,161,162,163,164,165,166,167],[143],["D",1784280622128],["D",1789527031073],{"_99":98,"_107":103},{"_99":98,"_100":101,"_102":103,"_18":54,"_104":105,"_106":24},{"_13":88,"_31":89,"_42":41,"_90":43,"_91":-5,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":44,"_97":98},"ctx",{"_81":-5,"_127":289,"_290":291,"_292":293,"_294":24,"_295":24,"_296":24,"_297":58,"_298":-5,"_299":-7,"_300":-7,"_301":62,"_302":24,"_303":24,"_304":-5,"_305":24,"_306":24,"_307":24,"_308":-5},"ENV",{"_260":261,"_262":263,"_264":89,"_265":266,"_267":268,"_269":270,"_271":272,"_273":274,"_275":276,"_277":278,"_279":280,"_281":282,"_283":284,"_285":286,"_287":288},"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",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

Diff · v10