Datasets:

PeterKol
/

llm-questionnaire-protocol-sensitivity

/**
 * Shared pieces for the pilot run (2026-09-30). Outputs go to the bundle folder and to
 * <tmp>/pilot-2026-09-30/.
 *
 * Scoring: scoring-key.csv (scale and the five answer weights of every item) and scripts/score-from-key.mjs.
 * Up to 2026-10-05 the analysis called the product's own scoring code instead; the key was exported from that
 * code and the two were compared on every stored answer vector (README, "Scorer parity"). Item texts and
 * option labels come from items-en.csv (and items-en-licence-pending.csv, if present); the analysis does not
 * need them, only the model-calling scripts do.
 */
import fs from 'fs'
import path from 'path'
import crypto from 'crypto'
import os from 'os'
import { fileURLToPath } from 'url'
import { loadKey, scoreFromKey } from './score-from-key.mjs'

// Public bundle: paths are relative to the bundle (override with PROTOCOL_DIR).
const HERE = path.dirname(fileURLToPath(import.meta.url))
export const BUNDLE = (process.env.PROTOCOL_DIR ? path.resolve(process.env.PROTOCOL_DIR) : path.resolve(HERE, '..')).replace(/\\/g, '/')
export const VAULT = BUNDLE // name kept from the original scripts: the folder that holds runs/, runs-full/ and the result files
export const SCRATCH = path.join(process.env.PROTOCOL_SCRATCH || os.tmpdir(), 'pilot-2026-09-30')
export const HF_DIR = process.env.PROTOCOL_HF_DIR ? path.resolve(process.env.PROTOCOL_HF_DIR) : path.join(SCRATCH, 'hf') // the human dataset CSVs (see README)
export const RUNS = path.join(VAULT, 'runs')
fs.mkdirSync(RUNS, { recursive: true })
fs.mkdirSync(SCRATCH, { recursive: true })

// ---------- the scorer ----------
export const KEY_FILE = path.join(BUNDLE, 'scoring-key.csv')
const KEY = loadKey(KEY_FILE)
export const scoreSubmission = (slug, answers) => scoreFromKey(KEY, slug, answers)

// ---------- the three tests ----------
export const TESTS = {
  big_five: {
    slug: 'big-five', pack: 'en.test.big-five.json', key: 'big_five', csv: 'big_five.csv',
    scales: ['O', 'C', 'E', 'A', 'N'], nItems: 50,
  },
  riasec: {
    slug: 'riasec', pack: 'en.test.riasec.json', key: 'riasec', csv: 'riasec.csv',
    scales: ['R', 'I', 'A', 'S', 'E', 'C'], nItems: 60,
  },
  dark_triad: {
    slug: 'dark-triad', pack: 'en.test.dark-triad.json', key: 'dark_triad', csv: 'dark_triad.csv',
    scales: ['narcissism', 'machiavellianism', 'psychopathy'], nItems: 18,
  },
}
export const TEST_IDS = Object.keys(TESTS)

/** Per item: scale, whether it is reverse-keyed (from the weights, not from a comment), and the text and option labels
 *  exactly as the English pack served them to a person (null when the item-text files are absent). */
export function loadItems(testId) {
  const t = TESTS[testId]
  const k = KEY[t.slug]
  if (!k || k.items.length !== t.nItems) throw new Error(`${testId}: key has ${k ? k.items.length : 0} items, dataset has ${t.nItems}`)
  const texts = {}
  for (const f of ['items-en.csv', 'items-en-licence-pending.csv']) {
    const p = path.join(BUNDLE, f)
    if (!fs.existsSync(p)) continue
    const { head, rows } = readCsv(p)
    const c = Object.fromEntries(head.map((h, i) => [h, i]))
    for (const r of rows) if (r.length === head.length) texts[`${r[c.test]}/${r[c.item]}`] = { text: r[c.text], options: [0, 1, 2, 3, 4].map((o) => r[c[`option_${o}`]]) }
  }
  return k.items.map((it, i) => {
    const w = it.weights
    const tx = texts[`${t.slug}/${i + 1}`]
    return { i, text: tx ? tx.text : null, options: tx ? tx.options : null, scale: it.scale, reverse: w[0] > w[w.length - 1] }
  })
}

// ---------- CSV ----------
export function readCsv(file) {
  const text = fs.readFileSync(file, 'utf8')
  const rows = []
  let row = [], field = '', q = false
  for (let i = 0; i < text.length; i++) {
    const c = text[i]
    if (q) {
      if (c === '"') { if (text[i + 1] === '"') { field += '"'; i++ } else q = false } else field += c
    } else if (c === '"') q = true
    else if (c === ',') { row.push(field); field = '' }
    else if (c === '\n') { row.push(field.replace(/\r$/, '')); rows.push(row); row = []; field = '' }
    else field += c
  }
  if (field.length || row.length) { row.push(field); rows.push(row) }
  const head = rows.shift()
  return { head, rows }
}

export function loadHumans(testId) {
  const t = TESTS[testId]
  const { head, rows } = readCsv(path.join(HF_DIR, t.csv))
  const idx = Object.fromEntries(head.map((h, i) => [h, i]))
  const qCols = Array.from({ length: t.nItems }, (_, k) => idx[`q${k + 1}`])
  if (qCols.some((c) => c == null)) throw new Error(`${testId}: missing q columns`)
  return rows.filter((r) => r.length === head.length).map((r) => ({
    id: Number(r[idx.response_id]),
    locale: r[idx.locale],
    ym: r[idx.year_month],
    duration: r[idx.duration_seconds] === '' ? null : Number(r[idx.duration_seconds]),
    top: r[idx.top_result],
    stored: Object.fromEntries(t.scales.map((s) => [s, r[idx[`score_${s}`]] === '' ? null : Number(r[idx[`score_${s}`]])])),
    q: qCols.map((c) => Number(r[c])),
  }))
}

// ---------- small stats ----------
export const mean = (a) => a.reduce((x, y) => x + y, 0) / a.length
export const sd = (a) => { const m = mean(a); return Math.sqrt(a.reduce((s, x) => s + (x - m) ** 2, 0) / (a.length - 1)) }
export const median = (a) => { const s = [...a].sort((x, y) => x - y); const n = s.length; return n % 2 ? s[(n - 1) / 2] : (s[n / 2 - 1] + s[n / 2]) / 2 }
export const quantile = (sorted, p) => { const i = (sorted.length - 1) * p; const lo = Math.floor(i), hi = Math.ceil(i); return sorted[lo] + (sorted[hi] - sorted[lo]) * (i - lo) }
export function pearson(x, y) {
  const mx = mean(x), my = mean(y)
  let sxy = 0, sxx = 0, syy = 0
  for (let i = 0; i < x.length; i++) { sxy += (x[i] - mx) * (y[i] - my); sxx += (x[i] - mx) ** 2; syy += (y[i] - my) ** 2 }
  return sxx && syy ? sxy / Math.sqrt(sxx * syy) : NaN
}
/** mid-rank percentile of a value inside a sample (ties count half). */
export function midrankPct(sortedSample, v) {
  let lo = 0, hi = sortedSample.length
  while (lo < hi) { const m = (lo + hi) >> 1; if (sortedSample[m] < v) lo = m + 1; else hi = m }
  const below = lo
  hi = sortedSample.length
  let j = lo
  while (j < hi && sortedSample[j] === v) j++
  return ((below + 0.5 * (j - below)) / sortedSample.length) * 100
}
export const r1 = (x) => (x == null || Number.isNaN(x) ? null : Math.round(x * 10) / 10)
export const r2 = (x) => (x == null || Number.isNaN(x) ? null : Math.round(x * 100) / 100)
export const r3 = (x) => (x == null || Number.isNaN(x) ? null : Math.round(x * 1000) / 1000)

export function sha256File(p) { return crypto.createHash('sha256').update(fs.readFileSync(p)).digest('hex') }

/** What the numbers in an output file were computed from: the key, the scorer, the human CSVs. */
export function provenance() {
  const files = ['scoring-key.csv', 'scripts/score-from-key.mjs']
  return {
    sha256: Object.fromEntries(files.map((f) => [f, sha256File(path.join(BUNDLE, f)).slice(0, 16)])),
    humanCsv: Object.fromEntries(Object.values(TESTS).map((t) => [t.csv, sha256File(path.join(HF_DIR, t.csv)).slice(0, 16)])),
    hfCommit: 'a39a58649a58202a02bb7eaf826609c2fe8bca84',
  }
}

	/**
	* Shared pieces for the pilot run (2026-09-30). Outputs go to the bundle folder and to
	* <tmp>/pilot-2026-09-30/.
	*
	* Scoring: scoring-key.csv (scale and the five answer weights of every item) and scripts/score-from-key.mjs.
	* Up to 2026-10-05 the analysis called the product's own scoring code instead; the key was exported from that
	* code and the two were compared on every stored answer vector (README, "Scorer parity"). Item texts and
	* option labels come from items-en.csv (and items-en-licence-pending.csv, if present); the analysis does not
	* need them, only the model-calling scripts do.
	*/
	import fs from 'fs'
	import path from 'path'
	import crypto from 'crypto'
	import os from 'os'
	import { fileURLToPath } from 'url'
	import { loadKey, scoreFromKey } from './score-from-key.mjs'

	// Public bundle: paths are relative to the bundle (override with PROTOCOL_DIR).
	const HERE = path.dirname(fileURLToPath(import.meta.url))
	export const BUNDLE = (process.env.PROTOCOL_DIR ? path.resolve(process.env.PROTOCOL_DIR) : path.resolve(HERE, '..')).replace(/\\/g, '/')
	export const VAULT = BUNDLE // name kept from the original scripts: the folder that holds runs/, runs-full/ and the result files
	export const SCRATCH = path.join(process.env.PROTOCOL_SCRATCH \|\| os.tmpdir(), 'pilot-2026-09-30')
	export const HF_DIR = process.env.PROTOCOL_HF_DIR ? path.resolve(process.env.PROTOCOL_HF_DIR) : path.join(SCRATCH, 'hf') // the human dataset CSVs (see README)
	export const RUNS = path.join(VAULT, 'runs')
	fs.mkdirSync(RUNS, { recursive: true })
	fs.mkdirSync(SCRATCH, { recursive: true })

	// ---------- the scorer ----------
	export const KEY_FILE = path.join(BUNDLE, 'scoring-key.csv')
	const KEY = loadKey(KEY_FILE)
	export const scoreSubmission = (slug, answers) => scoreFromKey(KEY, slug, answers)

	// ---------- the three tests ----------
	export const TESTS = {
	big_five: {
	slug: 'big-five', pack: 'en.test.big-five.json', key: 'big_five', csv: 'big_five.csv',
	scales: ['O', 'C', 'E', 'A', 'N'], nItems: 50,
	},
	riasec: {
	slug: 'riasec', pack: 'en.test.riasec.json', key: 'riasec', csv: 'riasec.csv',
	scales: ['R', 'I', 'A', 'S', 'E', 'C'], nItems: 60,
	},
	dark_triad: {
	slug: 'dark-triad', pack: 'en.test.dark-triad.json', key: 'dark_triad', csv: 'dark_triad.csv',
	scales: ['narcissism', 'machiavellianism', 'psychopathy'], nItems: 18,
	},
	}
	export const TEST_IDS = Object.keys(TESTS)

	/** Per item: scale, whether it is reverse-keyed (from the weights, not from a comment), and the text and option labels
	* exactly as the English pack served them to a person (null when the item-text files are absent). */
	export function loadItems(testId) {
	const t = TESTS[testId]
	const k = KEY[t.slug]
	if (!k \|\| k.items.length !== t.nItems) throw new Error(`${testId}: key has ${k ? k.items.length : 0} items, dataset has ${t.nItems}`)
	const texts = {}
	for (const f of ['items-en.csv', 'items-en-licence-pending.csv']) {
	const p = path.join(BUNDLE, f)
	if (!fs.existsSync(p)) continue
	const { head, rows } = readCsv(p)
	const c = Object.fromEntries(head.map((h, i) => [h, i]))
	for (const r of rows) if (r.length === head.length) texts[`${r[c.test]}/${r[c.item]}`] = { text: r[c.text], options: [0, 1, 2, 3, 4].map((o) => r[c[`option_${o}`]]) }
	}
	return k.items.map((it, i) => {
	const w = it.weights
	const tx = texts[`${t.slug}/${i + 1}`]
	return { i, text: tx ? tx.text : null, options: tx ? tx.options : null, scale: it.scale, reverse: w[0] > w[w.length - 1] }
	})
	}

	// ---------- CSV ----------
	export function readCsv(file) {
	const text = fs.readFileSync(file, 'utf8')
	const rows = []
	let row = [], field = '', q = false
	for (let i = 0; i < text.length; i++) {
	const c = text[i]
	if (q) {
	if (c === '"') { if (text[i + 1] === '"') { field += '"'; i++ } else q = false } else field += c
	} else if (c === '"') q = true
	else if (c === ',') { row.push(field); field = '' }
	else if (c === '\n') { row.push(field.replace(/\r$/, '')); rows.push(row); row = []; field = '' }
	else field += c
	}
	if (field.length \|\| row.length) { row.push(field); rows.push(row) }
	const head = rows.shift()
	return { head, rows }
	}

	export function loadHumans(testId) {
	const t = TESTS[testId]
	const { head, rows } = readCsv(path.join(HF_DIR, t.csv))
	const idx = Object.fromEntries(head.map((h, i) => [h, i]))
	const qCols = Array.from({ length: t.nItems }, (_, k) => idx[`q${k + 1}`])
	if (qCols.some((c) => c == null)) throw new Error(`${testId}: missing q columns`)
	return rows.filter((r) => r.length === head.length).map((r) => ({
	id: Number(r[idx.response_id]),
	locale: r[idx.locale],
	ym: r[idx.year_month],
	duration: r[idx.duration_seconds] === '' ? null : Number(r[idx.duration_seconds]),
	top: r[idx.top_result],
	stored: Object.fromEntries(t.scales.map((s) => [s, r[idx[`score_${s}`]] === '' ? null : Number(r[idx[`score_${s}`]])])),
	q: qCols.map((c) => Number(r[c])),
	}))
	}

	// ---------- small stats ----------
	export const mean = (a) => a.reduce((x, y) => x + y, 0) / a.length
	export const sd = (a) => { const m = mean(a); return Math.sqrt(a.reduce((s, x) => s + (x - m) ** 2, 0) / (a.length - 1)) }
	export const median = (a) => { const s = [...a].sort((x, y) => x - y); const n = s.length; return n % 2 ? s[(n - 1) / 2] : (s[n / 2 - 1] + s[n / 2]) / 2 }
	export const quantile = (sorted, p) => { const i = (sorted.length - 1) * p; const lo = Math.floor(i), hi = Math.ceil(i); return sorted[lo] + (sorted[hi] - sorted[lo]) * (i - lo) }
	export function pearson(x, y) {
	const mx = mean(x), my = mean(y)
	let sxy = 0, sxx = 0, syy = 0
	for (let i = 0; i < x.length; i++) { sxy += (x[i] - mx) * (y[i] - my); sxx += (x[i] - mx) 2; syy += (y[i] - my) 2 }
	return sxx && syy ? sxy / Math.sqrt(sxx * syy) : NaN
	}
	/** mid-rank percentile of a value inside a sample (ties count half). */
	export function midrankPct(sortedSample, v) {
	let lo = 0, hi = sortedSample.length
	while (lo < hi) { const m = (lo + hi) >> 1; if (sortedSample[m] < v) lo = m + 1; else hi = m }
	const below = lo
	hi = sortedSample.length
	let j = lo
	while (j < hi && sortedSample[j] === v) j++
	return ((below + 0.5 * (j - below)) / sortedSample.length) * 100
	}
	export const r1 = (x) => (x == null \|\| Number.isNaN(x) ? null : Math.round(x * 10) / 10)
	export const r2 = (x) => (x == null \|\| Number.isNaN(x) ? null : Math.round(x * 100) / 100)
	export const r3 = (x) => (x == null \|\| Number.isNaN(x) ? null : Math.round(x * 1000) / 1000)

	export function sha256File(p) { return crypto.createHash('sha256').update(fs.readFileSync(p)).digest('hex') }

	/** What the numbers in an output file were computed from: the key, the scorer, the human CSVs. */
	export function provenance() {
	const files = ['scoring-key.csv', 'scripts/score-from-key.mjs']
	return {
	sha256: Object.fromEntries(files.map((f) => [f, sha256File(path.join(BUNDLE, f)).slice(0, 16)])),
	humanCsv: Object.fromEntries(Object.values(TESTS).map((t) => [t.csv, sha256File(path.join(HF_DIR, t.csv)).slice(0, 16)])),
	hfCommit: 'a39a58649a58202a02bb7eaf826609c2fe8bca84',
	}
	}