import fs from 'fs'
const all = JSON.parse(fs.readFileSync(process.argv[2], 'utf8'))
const norm = s => s.toLowerCase().replace(/[^a-z0-9 ]/g, ' ').replace(/\s+/g, ' ').trim()

// Product management only. The exclusion list is deliberately short: because
// INCLUDE already pins the profession ("product manager" / "product management"),
// a word like "developer" or "engineer" in one of these titles names the product
// domain, not the job — "Staff Product Manager, Developer Experience" is a PM.
// An earlier, wider EXCLUDE dropped 14 real PMs for exactly that reason.
// "Product Marketing Manager" never matches INCLUDE anyway; it is listed for safety.
const INCLUDE = /\bproduct manager\b|\bproduct management\b/i
const EXCLUDE = /\bproduct marketing\b|\bintern\b|\binternship\b/i

const seen = new Set(), jobs = []
for (const j of all.filter(x => INCLUDE.test(x.title) && !EXCLUDE.test(x.title))) {
  const k = j.co + '|' + norm(j.title) + '|' + j.content.length
  if (!seen.has(k)) { seen.add(k); jobs.push(j) }
}
const unesc = s => s.replace(/&lt;/g,'<').replace(/&gt;/g,'>').replace(/&amp;/g,'&')
  .replace(/&quot;/g,'"').replace(/&#39;/g,"'").replace(/&nbsp;/g,' ')
const texts = jobs.map(j => unesc(j.content).replace(/<[^>]+>/g,' ').replace(/\s+/g,' '))

// Left out on purpose, and said so on the page:
//   Linear     — 0.7% loose, 0.2% once "linear regression/algebra/model" is excluded.
//                Too rare to report and dirty at that size.
//   Aha!       — one posting. The trailing "!" also makes a word-boundary match
//                unreliable.
//   platform   — 85%, but it is ordinary English in a tech posting rather than a
//                term a candidate can act on. Reporting it would pad the table.
// "discovery" is matched only in its qualified forms: bare "discovery" hits 17.6%
// against 4.0% for product/customer/user/continuous discovery, and the remainder
// is generic English.
const T = [
 ['Craft & practice','roadmap',/\broadmap/i],
 ['Craft & practice','product strategy',/\bproduct strategy\b/i],
 ['Craft & practice','prioritization',/\bprioriti[sz]/i],
 ['Craft & practice','user research',/\buser research\b|\bcustomer research\b/i],
 ['Craft & practice','customer interviews',/\bcustomer interviews?\b|\buser interviews?\b/i],
 ['Craft & practice','product discovery',/\b(product|customer|user|continuous)\s+discovery\b|\bdiscovery (process|work|phase)\b/i],
 ['Craft & practice','requirements / PRD',/\bPRDs?\b|\bproduct requirements?\b/i],
 ['Craft & practice','user stories',/\buser stor(y|ies)\b/i],
 ['Craft & practice','go-to-market',/\bgo[- ]to[- ]market\b|\bGTM\b/i],
 ['Craft & practice','stakeholder management',/\bstakeholder/i],
 ['Craft & practice','cross-functional',/\bcross[- ]functional\b/i],

 ['Metrics & analysis','KPIs',/\bKPIs?\b/i],
 ['Metrics & analysis','OKRs',/\bOKRs?\b/i],
 ['Metrics & analysis','A/B testing',/\bA\/B test/i],
 ['Metrics & analysis','experimentation',/\bexperiment/i],
 ['Metrics & analysis','retention',/\bretention\b/i],
 ['Metrics & analysis','churn',/\bchurn\b/i],
 ['Metrics & analysis','conversion',/\bconversion\b/i],
 ['Metrics & analysis','engagement',/\bengagement\b/i],
 ['Metrics & analysis','north star metric',/\bnorth star\b/i],
 ['Metrics & analysis','funnel',/\bfunnel\b/i],
 ['Metrics & analysis','cohort analysis',/\bcohort/i],
 ['Metrics & analysis','NPS',/\bNPS\b/],

 ['Method','Agile',/\bAgile\b/i],
 ['Method','Scrum',/\bScrum\b/i],
 ['Method','sprint',/\bsprints?\b/i],
 ['Method','Kanban',/\bKanban\b/i],
 ['Method','lean',/\blean (startup|product)\b/i],

 ['Tools','Jira',/\bJira\b/i],
 ['Tools','Figma',/\bFigma\b/i],
 ['Tools','SQL',/\bSQL\b/],
 ['Tools','Amplitude',/\bAmplitude\b/],
 ['Tools','Mixpanel',/\bMixpanel\b/i],
 ['Tools','Looker',/\bLooker\b/i],
 ['Tools','Tableau',/\bTableau\b/i],
 ['Tools','Confluence',/\bConfluence\b/i],
 ['Tools','Productboard',/\bProductboard\b/i],

 ['Domain & context','B2B',/\bB2B\b/i],
 ['Domain & context','B2C',/\bB2C\b/i],
 ['Domain & context','SaaS',/\bSaaS\b/i],
 ['Domain & context','API',/\bAPIs?\b/],
 ['Domain & context','AI / machine learning',/\bmachine learning\b|\bartificial intelligence\b|\bAI\/ML\b/i],
 ['Domain & context','enterprise',/\benterprise\b/i],
 ['Domain & context','MBA',/\bMBA\b/],
]
const n = texts.length
const moe = p => +(1.96 * Math.sqrt((p/100)*(1-p/100)/n) * 100).toFixed(1)
const rows = T.map(([cat, term, re]) => {
  const c = texts.filter(t => re.test(t)).length
  const pct = +(100*c/n).toFixed(1)
  return { cat, term, n: c, pct, moe: moe(pct) }
}).sort((a,b) => b.n - a.n)
const out = { role: 'Product Manager', n, companies: [...new Set(jobs.map(j=>j.co))].sort(),
  collected: new Date(Date.now() - new Date().getTimezoneOffset() * 60000).toISOString().slice(0,10),
  medianWords: texts.map(t=>t.split(' ').length).sort((a,b)=>a-b)[Math.floor(n/2)], rows }
fs.writeFileSync(process.argv[3], JSON.stringify(out, null, 1))
console.log(`n=${n} from ${out.companies.length} companies, median ${out.medianWords} words`)
rows.forEach((r,i)=>console.log(`  ${String(i+1).padStart(2)}. ${r.term.padEnd(24)} ${String(r.pct).padStart(5)}% ±${r.moe}  (${r.n})`))