scripts/generate_ux_feedback.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
EU-Retail-UX-Feedback-Live - Synthetic live UX feedback dataset generator
=========================================================================

This script produces a realistic, *synthetic* real-time user-experience (UX)
feedback dataset for a large European e-commerce retail company (1000+ employees)
operating in the United Kingdom, France and Germany.

The pipeline has two stages:

  Stage 1 - RAW generation:
      Simulates records as they would arrive from real collection systems
      (e-commerce website, mobile app, customer service platform). RAW records
      contain direct personal data (user email, name, phone number, address)
      to illustrate what must NEVER be published.

  Stage 2 - GDPR anonymisation (pseudonymisation):
      * Deletes all direct identifiers (email, name, phone, address).
      * Replaces the real user identifier with a pseudonymous ID
        (HMAC-SHA256 over the raw user id with a secret salt).
      * Replaces session identifiers with pseudonymous session tokens.
      * Scrubs free-text feedback for residual PII patterns
        (emails, phone numbers, UK/FR/DE postcodes, common personal data).
      * Every record is encoded as UTF-8 and tagged with a BCP-47 language
        identifier (en-GB, fr-FR, de-DE).

Only the anonymised output is ever uploaded to the Hugging Face Hub.

Versioning
----------
The dataset is refreshed every 30 minutes. Each refresh produces an immutable,
cumulative snapshot (v1.0.0, v1.1.0, ...) with a SHA-256 checksum recorded in
a version manifest so every published version is auditable and rollback-able.
"""

from __future__ import annotations

import hashlib
import hmac
import json
import os
import random
import re
import secrets
import string
from datetime import datetime, timedelta, timezone

import pandas as pd

# ---------------------------------------------------------------------------
# Configuration
# ---------------------------------------------------------------------------
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
DATA_DIR = os.path.join(BASE_DIR, "data")
SNAPSHOT_DIR = os.path.join(BASE_DIR, "snapshots")
VERSION_DIR = os.path.join(BASE_DIR, "versions")
RAW_DIR = os.path.join(BASE_DIR, "raw_deleted_never_upload")

PSEUDO_SALT = os.environ.get("UX_PSEUDO_SALT", "EU-Retail-UX-2026-kdf-salt")

DIMENSIONS = [
    "page_load_speed",
    "search_relevance",
    "product_recommendations",
    "checkout_flow",
    "delivery_experience",
    "after_sales_support",
]

MARKETS = [
    {"country_code": "GB", "region": "United Kingdom", "language": "en-GB"},
    {"country_code": "FR", "region": "France", "language": "fr-FR"},
    {"country_code": "DE", "region": "Germany", "language": "de-DE"},
]

DEVICE_TYPES = ["desktop", "mobile", "tablet"]
SOURCE_CHANNELS = ["website", "mobile_app", "customer_service"]

DIMENSION_LABELS = {
    "page_load_speed": "Page load speed",
    "search_relevance": "Search relevance",
    "product_recommendations": "Product recommendations",
    "checkout_flow": "Checkout flow",
    "delivery_experience": "Delivery experience",
    "after_sales_support": "After-sales support",
}

# ---------------------------------------------------------------------------
# Multi-language feedback content (UTF-8)
# ---------------------------------------------------------------------------
# Each entry: lang -> dimension -> rating band -> list of template sentences.
FEEDBACK = {
    "en-GB": {
        "page_load_speed": {
            "pos": [
                "The site loaded instantly and the product pages felt really snappy.",
                "Page loading speed has improved a lot, everything opens quickly.",
                "Lovely fast experience, the homepage loaded in under a second.",
                "Search and category pages load very quickly on both mobile and desktop.",
            ],
            "neu": [
                "Loading speed is acceptable but could be a bit faster on the homepage.",
                "Pages take a few seconds to load on my connection, not terrible though.",
                "The site works fine most of the time, with the occasional slow image.",
                "Reasonable loading times, though the product gallery is a little heavy.",
            ],
            "neg": [
                "The checkout page took ages to load and I nearly gave up.",
                "Product pages are painfully slow on mobile data, very frustrating.",
                "The site kept spinning during the sale and pages timed out.",
                "Loading speed is poor, images appear seconds after the text.",
            ],
        },
        "search_relevance": {
            "pos": [
                "Search results were spot on, exactly what I was looking for.",
                "The search function understands what I type and returns great matches.",
                "Great search relevance, the filters helped me narrow down quickly.",
                "I found the exact item in seconds thanks to smart search suggestions.",
            ],
            "neu": [
                "Search results are okay but a couple of items were not relevant.",
                "The search works, though it sometimes ignores my spelling mistakes.",
                "Results were acceptable, but I had to scroll to find the right size.",
                "Search is decent, the category filters helped a little.",
            ],
            "neg": [
                "Search returned mostly irrelevant products, I wasted time scrolling.",
                "The search engine seems to ignore my keywords entirely.",
                "Search relevance is poor, none of the results matched my query.",
                "I searched a specific brand and got unrelated items first.",
            ],
        },
        "product_recommendations": {
            "pos": [
                "The recommendations on the homepage were genuinely useful.",
                "Loved the 'you might also like' section, it suggested perfect items.",
                "Recommendations matched my taste and I added two extra products.",
                "The personalised recommendations felt relevant and helpful.",
            ],
            "neu": [
                "Recommendations were hit and miss, some useful and some not.",
                "The suggested items are fine but nothing really caught my eye.",
                "Reasonable recommendations, though a few were repeats of my history.",
                "The 'related items' were okay, I expected something more tailored.",
            ],
            "neg": [
                "The recommendations are always the same items I already viewed.",
                "Suggested products were completely unrelated to what I bought.",
                "Recommendations feel generic and irrelevant, I ignored them all.",
                "Poor recommendations, they kept pushing items out of stock.",
            ],
        },
        "checkout_flow": {
            "pos": [
                "Checkout was smooth, only a few steps and payment worked first time.",
                "The checkout flow is simple and I paid with Apple Pay in seconds.",
                "Very easy checkout, the address was pre-filled and everything was clear.",
                "Smooth and quick checkout, exactly how online shopping should feel.",
            ],
            "neu": [
                "Checkout worked but I had to create an account to finish.",
                "The flow is fine, though there were a couple of extra steps.",
                "Checkout was straightforward but the page reloaded a few times.",
                "Decent checkout, the delivery options could have been clearer.",
            ],
            "neg": [
                "Checkout kept failing at the payment step, very frustrating.",
                "I had to re-enter my card details three times before it worked.",
                "The checkout flow is too long and I almost abandoned the order.",
                "Payment errors and no clear error message, a really poor experience.",
            ],
        },
        "delivery_experience": {
            "pos": [
                "Delivery arrived earlier than promised and was well packaged.",
                "Excellent delivery, tracking updated regularly and on time.",
                "My parcel arrived the next day, brilliant service.",
                "Delivery was quick and the courier was friendly and careful.",
            ],
            "neu": [
                "Delivery took a couple of days longer than the estimate.",
                "The parcel arrived fine but the tracking stopped updating.",
                "Delivery was okay, packaging was a little damaged on the outside.",
                "Delivery was within the expected window, nothing special.",
            ],
            "neg": [
                "My order was delayed by a week with no explanation at all.",
                "Tracking never updated and the parcel was left in the rain.",
                "The delivery arrived damaged and the box was completely crushed.",
                "Poor delivery experience, the courier did not even knock.",
            ],
        },
        "after_sales_support": {
            "pos": [
                "Customer support resolved my issue quickly and politely.",
                "The after-sales team replied within minutes and fixed everything.",
                "Excellent support, the agent went above and beyond to help me.",
                "My refund was processed fast and the support chat was helpful.",
            ],
            "neu": [
                "Support answered my query but it took two days.",
                "The agent was polite but could not solve my problem directly.",
                "Support was okay, though I had to repeat my details a few times.",
                "Reasonable help, but the reply was a bit generic.",
            ],
            "neg": [
                "I waited on hold for 40 minutes and then got cut off.",
                "After-sales support was useless, nobody responded to my email.",
                "My complaint was ignored for weeks and I got no refund.",
                "The support agent was unhelpful and I had to contact them three times.",
            ],
        },
    },
    "fr-FR": {
        "page_load_speed": {
            "pos": [
                "Le site se charge instantanément, c'est vraiment très fluide.",
                "Les pages produits s'ouvrent très rapidement, bravo.",
                "La vitesse de chargement est excellente, même sur mobile.",
                "Le site est rapide et agréable à utiliser au quotidien.",
            ],
            "neu": [
                "Le chargement est correct mais un peu lent sur la page d'accueil.",
                "Les pages mettent quelques secondes à s'afficher, ce n'est pas terrible.",
                "La vitesse est acceptable même si les images sont parfois lourdes.",
                "Les temps de chargement sont corrects, sans plus.",
            ],
            "neg": [
                "La page de paiement a mis une éternité à charger, j'ai failli abandonner.",
                "Les pages produits sont très lentes en 4G, c'est frustrant.",
                "Le site a planté pendant les soldes, les pages ne se chargeaient pas.",
                "Le chargement est vraiment lent, les images arrivent trop tard.",
            ],
        },
        "search_relevance": {
            "pos": [
                "Les résultats de recherche étaient parfaits, exactement ce que je cherchais.",
                "La recherche comprend bien ce que je tape et propose d'excellents résultats.",
                "La pertinence des résultats est excellente, les filtres m'ont bien aidé.",
                "J'ai trouvé l'article en quelques secondes grâce aux suggestions.",
            ],
            "neu": [
                "Les résultats sont corrects mais certains articles n'étaient pas pertinents.",
                "La recherche fonctionne mais ignore parfois mes fautes de frappe.",
                "Les résultats étaient acceptables, il a fallu chercher un peu.",
                "La recherche est correcte, sans plus.",
            ],
            "neg": [
                "La recherche m'a renvoyé surtout des produits hors sujet.",
                "Le moteur de recherche semble ignorer complètement mes mots-clés.",
                "La pertinence est mauvaise, aucun résultat ne correspondait à ma demande.",
                "J'ai cherché une marque précise et j'ai eu des articles sans rapport.",
            ],
        },
        "product_recommendations": {
            "pos": [
                "Les recommandations sur la page d'accueil étaient vraiment utiles.",
                "J'ai adoré la section 'vous aimerez aussi', les suggestions étaient parfaites.",
                "Les recommandations correspondaient à mes goûts, j'ai ajouté deux articles.",
                "Les recommandations personnalisées étaient pertinentes et utiles.",
            ],
            "neu": [
                "Les recommandations sont inégales, certaines utiles et d'autres non.",
                "Les suggestions sont correctes mais rien ne m'a vraiment attiré.",
                "Des recommandations raisonnables, bien que répétitives.",
                "Les articles associés étaient corrects, je m'attendais à mieux.",
            ],
            "neg": [
                "Les recommandations sont toujours les mêmes articles déjà consultés.",
                "Les produits suggérés n'avaient rien à voir avec mon achat.",
                "Les recommandations sont génériques et sans intérêt.",
                "Mauvaises recommandations, on me proposait des articles en rupture.",
            ],
        },
        "checkout_flow": {
            "pos": [
                "Le paiement s'est fait en quelques clics, très simple et rapide.",
                "Le processus de commande est fluide, j'ai payé avec Apple Pay.",
                "Commande très facile, l'adresse était pré-remplie et tout était clair.",
                "Un parcours d'achat rapide et agréable, bravo.",
            ],
            "neu": [
                "La commande a fonctionné mais j'ai dû créer un compte pour finir.",
                "Le parcours est correct avec quelques étapes de plus que prévu.",
                "Le paiement était simple mais la page s'est rechargée plusieurs fois.",
                "Un achat correct, les options de livraison étaient peu claires.",
            ],
            "neg": [
                "Le paiement a échoué plusieurs fois, c'est très frustrant.",
                "J'ai dû saisir mes coordonnées bancaires trois fois.",
                "Le processus de commande est trop long, j'ai failli abandonner.",
                "Erreurs de paiement sans message clair, mauvaise expérience.",
            ],
        },
        "delivery_experience": {
            "pos": [
                "La livraison est arrivée en avance et bien emballée.",
                "Excellente livraison, le suivi était à jour et ponctuel.",
                "Mon colis est arrivé le lendemain, service impeccable.",
                "Livraison rapide et livreur très aimable.",
            ],
            "neu": [
                "La livraison a pris quelques jours de plus que prévu.",
                "Le colis est arrivé mais le suivi ne se mettait plus à jour.",
                "Livraison correcte, l'emballage était un peu abîmé.",
                "Livraison dans les délais annoncés, sans surprise.",
            ],
            "neg": [
                "Ma commande a été retardée d'une semaine sans aucune explication.",
                "Le suivi n'a jamais été mis à jour et le colis a été laissé sous la pluie.",
                "La livraison est arrivée endommagée, le carton était écrasé.",
                "Mauvaise expérience, le livreur n'a même pas sonné.",
            ],
        },
        "after_sales_support": {
            "pos": [
                "Le service client a répondu rapidement et résolu mon problème.",
                "L'équipe après-vente m'a répondu en quelques minutes.",
                "Excellent support, l'agent a fait tout son possible pour m'aider.",
                "Mon remboursement a été traité très vite, service réactif.",
            ],
            "neu": [
                "Le support a répondu à ma demande mais après deux jours.",
                "L'agent était poli mais n'a pas pu résoudre mon problème.",
                "Le support était correct, même si j'ai dû répéter mes informations.",
                "Une aide correcte, mais la réponse était un peu générique.",
            ],
            "neg": [
                "J'ai attendu 40 minutes au téléphone puis on m'a raccroché au nez.",
                "Le service après-vente est inutile, personne ne répond à mes e-mails.",
                "Ma réclamation a été ignorée pendant des semaines.",
                "L'agent n'était pas serviable, j'ai dû le contacter trois fois.",
            ],
        },
    },
    "de-DE": {
        "page_load_speed": {
            "pos": [
                "Die Seite lädt sofort, wirklich sehr flüssig.",
                "Die Produktseiten öffnen sich sehr schnell, super.",
                "Die Ladezeit ist ausgezeichnet, auch mobil.",
                "Die Website ist schnell und angenehm zu nutzen.",
            ],
            "neu": [
                "Das Laden ist in Ordnung, aber die Startseite könnte schneller sein.",
                "Die Seiten brauchen ein paar Sekunden, nicht schlimm aber ausbaufähig.",
                "Die Geschwindigkeit ist akzeptabel, die Bilder sind manchmal schwer.",
                "Die Ladezeiten sind okay, mehr aber auch nicht.",
            ],
            "neg": [
                "Die Checkout-Seite hat eine Ewigkeit zum Laden gebraucht.",
                "Produktseiten sind im Mobilfunknetz extrem langsam, sehr nervig.",
                "Die Seite hing während des Sale und Seiten liefen in Timeouts.",
                "Das Laden ist richtig langsam, die Bilder kommen viel zu spät.",
            ],
        },
        "search_relevance": {
            "pos": [
                "Die Suchergebnisse waren punktgenau, genau was ich suchte.",
                "Die Suche versteht meine Eingaben und liefert tolle Treffer.",
                "Die Suchtreffer sind sehr relevant, die Filter haben mir geholfen.",
                "Ich habe den Artikel dank Suchvorschlägen in Sekunden gefunden.",
            ],
            "neu": [
                "Die Ergebnisse sind okay, aber ein paar Artikel passten nicht.",
                "Die Suche funktioniert, ignoriert aber manchmal Tippfehler.",
                "Die Ergebnisse waren annehmbar, ich musste etwas suchen.",
                "Die Suche ist ganz ordentlich, aber nichts Besonderes.",
            ],
            "neg": [
                "Die Suche lieferte überwiegend irrelevante Produkte.",
                "Die Suchmaschine scheint meine Suchbegriffe komplett zu ignorieren.",
                "Die Trefferqualität ist schlecht, kein Ergebnis passte zu meiner Anfrage.",
                "Ich suchte eine bestimmte Marke und bekam völlig andere Artikel.",
            ],
        },
        "product_recommendations": {
            "pos": [
                "Die Empfehlungen auf der Startseite waren wirklich nützlich.",
                "Ich liebe die Rubrik 'Das könnte Ihnen auch gefallen', perfekte Vorschläge.",
                "Die Empfehlungen passten zu meinem Geschmack, ich habe zwei Artikel ergänzt.",
                "Die personalisierten Empfehlungen waren relevant und hilfreich.",
            ],
            "neu": [
                "Die Empfehlungen waren gemischt, manche nützlich, manche nicht.",
                "Die Vorschläge sind in Ordnung, aber nichts hat mich begeistert.",
                "Akzeptable Empfehlungen, auch wenn sich manche wiederholen.",
                "Die verwandten Artikel waren okay, ich hatte mehr erwartet.",
            ],
            "neg": [
                "Die Empfehlungen sind immer dieselben Artikel, die ich schon angesehen habe.",
                "Die vorgeschlagenen Produkte hatten nichts mit meinem Kauf zu tun.",
                "Die Empfehlungen sind generisch und uninteressant.",
                "Schlechte Empfehlungen, mir wurden ständig ausverkaufte Artikel gezeigt.",
            ],
        },
        "checkout_flow": {
            "pos": [
                "Der Checkout war flüssig, wenige Schritte und die Zahlung klappte sofort.",
                "Der Bestellprozess ist einfach, ich habe mit Apple Pay in Sekunden bezahlt.",
                "Sehr einfache Bestellung, die Adresse war vorausgefüllt und alles klar.",
                "Schneller und angenehmer Checkout, so muss Online-Shopping sein.",
            ],
            "neu": [
                "Die Bestellung funktionierte, aber ich musste ein Konto anlegen.",
                "Der Ablauf ist in Ordnung, auch wenn es ein paar Schritte mehr waren.",
                "Der Checkout war einfach, aber die Seite lud ein paar Mal neu.",
                "Ordentlicher Kauf, die Lieferoptionen waren aber unklar.",
            ],
            "neg": [
                "Die Zahlung scheiterte mehrfach, sehr frustrierend.",
                "Ich musste meine Kartendaten dreimal eingeben.",
                "Der Bestellprozess ist zu lang, ich hätte fast abgebrochen.",
                "Zahlungsfehler ohne klare Meldung, eine wirklich schlechte Erfahrung.",
            ],
        },
        "delivery_experience": {
            "pos": [
                "Die Lieferung kam früher als erwartet und war gut verpackt.",
                "Ausgezeichnete Lieferung, die Sendungsverfolgung war aktuell.",
                "Mein Paket kam am nächsten Tag, großartiger Service.",
                "Schnelle Lieferung und ein freundlicher, sorgfältiger Bote.",
            ],
            "neu": [
                "Die Lieferung dauerte ein paar Tage länger als angekündigt.",
                "Das Paket kam an, aber die Sendungsverfolgung blieb stehen.",
                "Die Lieferung war okay, die Verpackung war außen etwas beschädigt.",
                "Die Lieferung kam im erwarteten Zeitraum, nichts Besonderes.",
            ],
            "neg": [
                "Meine Bestellung kam eine Woche zu spät ohne jede Erklärung.",
                "Die Sendungsverfolgung aktualisierte sich nie und das Paket lag im Regen.",
                "Die Lieferung kam beschädigt an, der Karton war völlig zerdrückt.",
                "Schlechte Lieferung, der Bote hat nicht einmal geklingelt.",
            ],
        },
        "after_sales_support": {
            "pos": [
                "Der Kundenservice hat mein Problem schnell und freundlich gelöst.",
                "Das After-Sales-Team antwortete innerhalb weniger Minuten.",
                "Ausgezeichneter Support, der Mitarbeiter hat sich sehr eingesetzt.",
                "Meine Erstattung wurde schnell bearbeitet, hilfreicher Chat.",
            ],
            "neu": [
                "Der Support hat geantwortet, aber erst nach zwei Tagen.",
                "Der Mitarbeiter war freundlich, konnte mein Problem aber nicht lösen.",
                "Der Support war okay, ich musste meine Daten mehrfach wiederholen.",
                "Ordentliche Hilfe, aber die Antwort war etwas generisch.",
            ],
            "neg": [
                "Ich hing 40 Minuten in der Warteschleife und wurde dann getrennt.",
                "Der After-Sales-Support ist nutzlos, niemand antwortet auf E-Mails.",
                "Meine Beschwerde wurde wochenlang ignoriert.",
                "Der Mitarbeiter war nicht hilfreich, ich musste dreimal anrufen.",
            ],
        },
    },
}

# Optional follow-up sentences (added ~35% of the time) for extra variety.
FOLLOW_UPS = {
    "en-GB": [
        "I use the site a few times a week on my phone.",
        "This is the second order I have placed this month.",
        "I would recommend the store to friends and family.",
        "I mainly shop during the weekend sales.",
        "The mobile app is my preferred way to order.",
    ],
    "fr-FR": [
        "Je commande sur le site plusieurs fois par semaine.",
        "C'est ma deuxième commande ce mois-ci.",
        "Je recommande volontiers cette boutique à mes proches.",
        "Je fais mes achats surtout pendant les ventes du week-end.",
        "J'utilise surtout l'application mobile pour commander.",
    ],
    "de-DE": [
        "Ich bestelle mehrmals pro Woche über die Website.",
        "Das ist meine zweite Bestellung in diesem Monat.",
        "Ich würde den Shop an Freunde und Familie weiterempfehlen.",
        "Ich kaufe vor allem an den Wochenenden im Sale ein.",
        "Ich bestelle am liebsten über die Mobile-App.",
    ],
}

# ---------------------------------------------------------------------------
# PII scrubbing for free-text (GDPR)
# ---------------------------------------------------------------------------
PII_PATTERNS = [
    re.compile(r"[A-Za-z0-9._%+\-]+@[A-Za-z0-9.\-]+\.[A-Za-z]{2,}"),  # email
    re.compile(r"(?<!\d)(?:\+?\d[\d\s().\-/]{7,}\d)(?!\d)"),  # phone
    re.compile(r"\b[A-Z]{1,2}\d[A-Z\d]?\s?\d[A-Z]{2}\b"),  # UK postcode
    re.compile(r"\b\d{5}\b"),  # FR postcode
    re.compile(r"\b\d{5}\b"),  # DE postcode
    re.compile(r"\b(?:\d{4}[ -]?){3}\d{4}\b"),  # card-like number
]

PII_REDACTION = "[REDACTED]"


def scrub_text(text: str) -> str:
    """Remove residual personal data from free text."""
    scrubbed = text
    for pattern in PII_PATTERNS:
        scrubbed = pattern.sub(PII_REDACTION, scrubbed)
    return scrubbed


# ---------------------------------------------------------------------------
# Pseudonymisation helpers
# ---------------------------------------------------------------------------
def pseudonym(value: str, namespace: str) -> str:
    """Deterministic HMAC-SHA256 pseudonym (non-reversible without salt)."""
    digest = hmac.new(
        PSEUDO_SALT.encode("utf-8"),
        f"{namespace}:{value}".encode("utf-8"),
        hashlib.sha256,
    ).hexdigest()
    return digest[:16]


def random_token(prefix: str, length: int = 12) -> str:
    alphabet = string.ascii_lowercase + string.digits
    return f"{prefix}-" + "".join(secrets.choice(alphabet) for _ in range(length))


# ---------------------------------------------------------------------------
# Raw record generation (with fake PII - never published)
# ---------------------------------------------------------------------------
FIRST_NAMES = {
    "en-GB": ["Oliver", "Amelia", "George", "Sophie", "Jack", "Emily", "Harry", "Grace", "Charlie", "Freya"],
    "fr-FR": ["Lucas", "Emma", "Louis", "Jade", "Gabriel", "Louise", "Hugo", "Chloé", "Arthur", "Léa"],
    "de-DE": ["Jonas", "Mia", "Leon", "Emma", "Finn", "Lena", "Paul", "Hannah", "Elias", "Lea"],
}
LAST_NAMES = {
    "en-GB": ["Smith", "Jones", "Taylor", "Brown", "Williams", "Davies", "Evans", "Wilson"],
    "fr-FR": ["Martin", "Bernard", "Dubois", "Thomas", "Robert", "Richard", "Petit", "Durand"],
    "de-DE": ["Müller", "Schmidt", "Schneider", "Fischer", "Weber", "Meyer", "Wagner", "Becker"],
}
CITIES = {
    "en-GB": ["London", "Manchester", "Birmingham", "Leeds", "Glasgow", "Bristol", "Edinburgh"],
    "fr-FR": ["Paris", "Lyon", "Marseille", "Toulouse", "Nice", "Nantes", "Strasbourg"],
    "de-DE": ["Berlin", "München", "Hamburg", "Köln", "Frankfurt", "Stuttgart", "Düsseldorf"],
}
DOMAINS = ["example.com", "mail.test", "webmail.eu"]


def fake_pii(lang: str, idx: int) -> dict:
    fn = random.choice(FIRST_NAMES[lang])
    ln = random.choice(LAST_NAMES[lang])
    email = f"{fn.lower()}.{ln.lower()}{idx}@{random.choice(DOMAINS)}"
    phone = "+44 7" + "".join(random.choices(string.digits, k=8))
    if lang == "fr-FR":
        phone = "+33 6 " + "".join(random.choices(string.digits, k=8))
    elif lang == "de-DE":
        phone = "+49 15" + "".join(random.choices(string.digits, k=8))
    city = random.choice(CITIES[lang])
    address = f"{random.randint(1, 240)} {random.choice(['High', 'Green', 'Station', 'Church', 'Market'])} {random.choice(['Street', 'Road', 'Lane', 'Avenue'])}, {city}"
    return {
        "user_email": email,
        "user_name": f"{fn} {ln}",
        "phone_number": phone,
        "address": address,
    }


def generate_batch(lang: str, n: int, start_ts: datetime, end_ts: datetime, global_start_idx: int) -> list[dict]:
    """Generate n raw records for a language in a time window."""
    market = next(m for m in MARKETS if m["language"] == lang)
    rows = []
    delta = (end_ts - start_ts).total_seconds()
    for i in range(n):
        ts = start_ts + timedelta(seconds=random.random() * delta)
        dim = random.choice(DIMENSIONS)
        rating = random.choices([1, 2, 3, 4, 5], weights=[7, 12, 22, 33, 26])[0]
        band = "pos" if rating >= 4 else ("neu" if rating == 3 else "neg")
        text = random.choice(FEEDBACK[lang][dim][band])
        if random.random() < 0.35:
            text += " " + random.choice(FOLLOW_UPS[lang])
        pii = fake_pii(lang, global_start_idx + i)
        rows.append(
            {
                "user_id": f"user-{global_start_idx + i:07d}",
                "user_email": pii["user_email"],
                "user_name": pii["user_name"],
                "phone_number": pii["phone_number"],
                "address": pii["address"],
                "session_id_raw": f"session-{secrets.token_hex(8)}",
                "market_region": market["region"],
                "country_code": market["country_code"],
                "language": lang,
                "timestamp": ts.isoformat(timespec="seconds"),
                "experience_dimension": dim,
                "rating": rating,
                "feedback_text": text,
                "device_type": random.choice(DEVICE_TYPES),
                "source_channel": random.choice(SOURCE_CHANNELS),
            }
        )
    return rows


def anonymise(raw_rows: list[dict]) -> pd.DataFrame:
    """GDPR pipeline: drop direct PII, pseudonymise IDs, scrub free text."""
    out = []
    for r in raw_rows:
        out.append(
            {
                "feedback_id": random_token("F", 12).upper(),
                "user_pseudo_id": pseudonym(r["user_id"], "user"),
                "session_id": pseudonym(r["session_id_raw"], "session"),
                "market_region": r["market_region"],
                "country_code": r["country_code"],
                "language": r["language"],
                "timestamp": r["timestamp"],
                "experience_dimension": r["experience_dimension"],
                "rating": int(r["rating"]),
                "feedback_text": scrub_text(r["feedback_text"]),
                "device_type": r["device_type"],
                "source_channel": r["source_channel"],
            }
        )
    df = pd.DataFrame(out)
    # Enforce column order and UTF-8
    cols = [
        "feedback_id", "user_pseudo_id", "session_id", "market_region",
        "country_code", "language", "timestamp", "experience_dimension",
        "rating", "feedback_text", "device_type", "source_channel",
    ]
    df = df[cols]
    # Ensure feedback_text is truly str and UTF-8 clean
    df["feedback_text"] = df["feedback_text"].astype(str).apply(
        lambda s: s.encode("utf-8", errors="replace").decode("utf-8")
    )
    return df


def sha256_file(path: str) -> str:
    h = hashlib.sha256()
    with open(path, "rb") as f:
        for chunk in iter(lambda: f.read(65536), b""):
            h.update(chunk)
    return h.hexdigest()


def write_snapshot(df: pd.DataFrame, version: str, window_label: str, window_start: str, window_end: str, added: int):
    os.makedirs(os.path.join(SNAPSHOT_DIR, version), exist_ok=True)
    parquet_path = os.path.join(SNAPSHOT_DIR, version, f"ux_feedback_{version}.parquet")
    csv_path = os.path.join(SNAPSHOT_DIR, version, f"ux_feedback_{version}.csv")
    df.to_parquet(parquet_path, index=False)
    df.to_csv(csv_path, index=False, encoding="utf-8")

    # canonical 'current' files at data/
    os.makedirs(DATA_DIR, exist_ok=True)
    df.to_parquet(os.path.join(DATA_DIR, "ux_feedback_live.parquet"), index=False)
    df.to_csv(os.path.join(DATA_DIR, "ux_feedback_live.csv"), index=False, encoding="utf-8")

    return {
        "version": version,
        "window_label": window_label,
        "window_start_utc": window_start,
        "window_end_utc": window_end,
        "records_added": added,
        "total_records": int(len(df)),
        "sha256_parquet": sha256_file(parquet_path),
        "sha256_csv": sha256_file(csv_path),
        "parquet_path": f"snapshots/{version}/ux_feedback_{version}.parquet",
        "csv_path": f"snapshots/{version}/ux_feedback_{version}.csv",
        "canonical_parquet": "data/ux_feedback_live.parquet",
        "canonical_csv": "data/ux_feedback_live.csv",
        "gdpr_anonymised": True,
        "pii_fields_dropped": ["user_id", "user_email", "user_name", "phone_number", "address"],
    }


def main():
    random.seed(20260813)
    rng_state = random.getstate()

    now = datetime.now(timezone.utc)
    base_now = now.replace(second=0, microsecond=0)

    # Version schedule (30-minute refresh cadence)
    schedule = [
        # (version, window_label, start, end, records_to_generate)
        ("v1.0.0", "initial_release", base_now - timedelta(hours=24), base_now, 5000),
        ("v1.1.0", "update_30min", base_now, base_now + timedelta(minutes=30), 750),
        ("v1.2.0", "update_60min", base_now + timedelta(minutes=30), base_now + timedelta(minutes=60), 750),
    ]

    # Raw records per language so each market is represented in every window
    manifest_versions = []
    cumulative_raw: list[dict] = []

    # Also keep a raw (PII) dump ONLY to demonstrate the pipeline internally.
    os.makedirs(RAW_DIR, exist_ok=True)
    raw_log_path = os.path.join(RAW_DIR, "raw_with_pii_sample.jsonl")

    with open(raw_log_path, "w", encoding="utf-8") as raw_log:
        for version, label, wstart, wend, total in schedule:
            n_per_lang = total // len(MARKETS)
            raw_batch = []
            for m in MARKETS:
                raw_batch.extend(generate_batch(m["language"], n_per_lang, wstart, wend, len(cumulative_raw) + len(raw_batch)))
            random.setstate(rng_state)
            random.shuffle(raw_batch)
            cumulative_raw.extend(raw_batch)
            for row in raw_batch:
                raw_log.write(json.dumps(row, ensure_ascii=False) + "\n")

            df = anonymise(cumulative_raw)
            info = write_snapshot(
                df, version, label,
                wstart.isoformat(timespec="seconds"),
                wend.isoformat(timespec="seconds"),
                added=len(raw_batch),
            )
            manifest_versions.append(info)
            print(f"[{version}] window={label} added={info['records_added']} total={info['total_records']}")

    # Write the version manifest + changelog
    os.makedirs(VERSION_DIR, exist_ok=True)
    manifest = {
        "dataset": "EU-Retail-UX-Feedback-Live",
        "description": "Live, GDPR-anonymised user-experience feedback from EU e-commerce retail (UK, FR, DE).",
        "refresh_cadence_minutes": 30,
        "encoding": "UTF-8",
        "languages": ["en-GB", "fr-FR", "de-DE"],
        "markets": ["United Kingdom", "France", "Germany"],
        "pii_handling": "deleted_and_pseudonymised",
        "latest_version": manifest_versions[-1]["version"],
        "versions": manifest_versions,
    }
    with open(os.path.join(VERSION_DIR, "manifest.json"), "w", encoding="utf-8") as f:
        json.dump(manifest, f, ensure_ascii=False, indent=2)

    changelog_lines = [
        "# Changelog - EU-Retail-UX-Feedback-Live",
        "",
        "This dataset is refreshed every 30 minutes. Every published snapshot is",
        "immutable, checksummed (SHA-256) and tagged so that any previous version",
        "can be restored (rollback).",
        "",
    ]
    for v in manifest_versions:
        changelog_lines.append(
            f"## {v['version']} - {v['window_label']}\n"
            f"- Window: {v['window_start_utc']} -> {v['window_end_utc']} UTC\n"
            f"- Records added: {v['records_added']} | Total records: {v['total_records']}\n"
            f"- SHA-256 (parquet): `{v['sha256_parquet']}`\n"
            f"- SHA-256 (csv): `{v['sha256_csv']}`\n"
        )
    with open(os.path.join(VERSION_DIR, "CHANGELOG.md"), "w", encoding="utf-8") as f:
        f.write("\n".join(changelog_lines))

    # Summary printout
    print("\n=== SUMMARY ===")
    print(json.dumps(manifest, indent=2, ensure_ascii=False))
    print("\nFiles written under:", BASE_DIR)


if __name__ == "__main__":
    main()