keywordbarrage/ozon-scraper

Python Code:

import sqlite3
from datasets import load_dataset
import pandas as pd

# Method 1: Load via Hugging Face datasets library
# This repository contains an open 1,000-row verified benchmark sample.
# For full, high-frequency production datasets, please refer to the
# "Zero-Code Enterprise Deliverables Matrix" section in the dataset card.
dataset = load_dataset("daas-intelligence/enterprise-sku-depletion-and-waf-telemetry")
print("Hub Sample Record:", dataset["train"][0])

# Method 2: Load zero-latency local pre-indexed SQLite deliverable
# This requires a local 'enterprise_feed.sqlite' file, typically provided
# as part of the full enterprise data deliverables.
try:
    conn = sqlite3.connect("enterprise_feed.sqlite")
    df = pd.read_sql_query("SELECT sku_id, price_usd, stock_level, depletion_velocity FROM sku_stream ORDER BY timestamp DESC LIMIT 10", conn)
    print(df)
    conn.close()
except sqlite3.OperationalError:
    print("\n'enterprise_feed.sqlite' not found. This is part of the full enterprise deliverable.")
    print("The Hugging Face repository provides a sample via 'load_dataset'.")

Dataset Card:

---

pretty_name: "Enterprise High-Frequency SKU Depletion & WAF Telemetry Feed: Ozon" annotations_creators: - machine-generated language: - en license: apache-2.0 task_categories: - tabular-classification - time-series-forecasting tags: - ecommerce - time-series-forecasting - anti-bot-bypass - sku-depletion - pre-indexed-sqlite - parquet-benchmark size_categories: - 50K<n<250K configs: - config_name: default data_files: - split: train path: "data/train-.parquet" - split: validation path: "data/val-.parquet"

Dataset Summary & Research Motivation

This repository contains an open 1,000-row verified benchmark sample of the Ozon SKU depletion feed, structured with pre-indexed SQLite and Snappy Parquet partitions for quantitative market intelligence research and data engineering evaluation.

Quantitative analysts, hedge funds, and data engineering teams face a critical challenge: brittle scraping pipelines that continuously break due to sophisticated anti-bot countermeasures. Traditional residential proxies and headless browser emulators fail against modern Web Application Firewalls (WAFs) that inspect client-side runtime consistency, TLS/JA4 signatures, WebGL/Canvas entropy, and state-level IP filtering. These failures lead to broken data feeds, delayed market intelligence, and significant engineering overhead.

This dataset exemplifies a production-grade data extraction service that bypasses these cutting-edge edge challenge loops with 99.9% verified pipeline availability. Our protocol-level data extraction methodology ensures resilient, high-frequency data delivery, even on Tier-1 hardened enterprise targets like Ozon, which employs a Universal WAF Defense Difficulty Score: 9.4 / 10 (Tier-1 Hardened Enterprise). The underlying ingestion engine intercepts internal GraphQL/REST catalog JSON endpoints and Nuxt.js SSR initial state payloads directly, eliminating the need for computationally expensive and unreliable "AI Vision" or headless browser abstractions.

Zero-Code Enterprise Deliverables Matrix

This benchmark sample is representative of the four enterprise-grade formats provided for full-scale deployments, designed for seamless integration into quantitative research and data engineering workflows:

1. Pre-indexed SQLite (.sqlite): Instant local querying with pre-built indices on critical fields such as sku_id, price_delta, and timestamp. Ideal for rapid prototyping, local analytics, and integration with BI tools without complex setup.
2. Snappy Parquet: Optimized for direct ingestion into cloud data warehouses like Snowflake, Google BigQuery, and AWS Athena. Parquet offers columnar storage and Snappy compression for efficient query performance and reduced storage costs.
3. Zero-Dependency Offline DataGrid (.html): A self-contained, interactive HTML table for quantitative stakeholders to visually inspect, filter, and analyze data without requiring SQL knowledge, API keys, or specialized software. This provides immediate data verification.
4. CDC JSONL Stream: Change Data Capture (CDC) formatted JSON Lines (JSONL) for efficient updates to Enterprise RAG pipelines, vector databases, or real-time streaming analytics platforms. This format ensures that only changed or new records are delivered, minimizing data transfer and processing load.

Dataset Schema & Data Dictionary

The Ozon SKU depletion feed provides comprehensive retail intelligence, including product details, pricing, stock levels, fulfillment options, and review metrics. The schema is designed for granular analysis of market trends and competitive positioning.

Column Name	Type	Nullability	Description
timestamp_utc	TIMESTAMP	NOT NULL	UTC timestamp of when the data record was extracted.
sku_id	TEXT	NOT NULL	Unique Ozon Product ID / SKU.
ean_barcode	TEXT	NULLABLE	Universal EAN-13 Barcode for the product.
product_title	TEXT	NOT NULL	Full product title as listed on Ozon.
brand_name	TEXT	NULLABLE	Brand name of the product (e.g., Xiaomi, Samsung, Tefal).
manufacturer	TEXT	NULLABLE	Manufacturer of the product.
category_hierarchy	TEXT	NOT NULL	Full category path (e.g., Elektronika > Smartfony).
price_ozon_card_rub	REAL	NULLABLE	Discounted price in Russian Rubles (₽) when paying with Ozon Card (Tsena s Ozon Kartoy).
price_standard_rub	REAL	NOT NULL	Standard marketplace price in Russian Rubles (₽).
price_rrp_rub	REAL	NULLABLE	Original Recommended Retail Price (RRP) in Russian Rubles (₽).
stock_level	INTEGER	NULLABLE	Current estimated stock level for the SKU.
buybox_seller_type	TEXT	NOT NULL	Classification of the current Buy-Box seller (Ozon / Prodavets Ozon vs. Partner Merchant).
ozon_premium_eligible	BOOLEAN	NOT NULL	Indicates if the product is eligible for Ozon Premium loyalty benefits (e.g., free express shipping, cashback).
fulfillment_mode	TEXT	NOT NULL	Primary fulfillment method (e.g., Ozon Rocket Courier, Pick-up Point / PVZ, Ozon Box Parcel Locker, Ozon Fresh).
customer_rating_stars	REAL	NULLABLE	Average customer review rating out of 5 stars (Ocena tovara iz 5 zvezd).
total_reviews_count	INTEGER	NULLABLE	Total number of verified customer reviews.
technical_specs_json	JSON	NULLABLE	JSON object containing detailed technical specifications (Kharakteristiki tovara: Model, Ves, Razmery, Strana proizvodstva).
velocity_depletion_rate_24h	REAL	NULLABLE	Estimated SKU depletion rate over the last 24 hours, calculated from observed stock changes.
target_domain	TEXT	NOT NULL	The target domain from which the data was extracted (e.g., ozon.ru).
detected_waf	TEXT	NOT NULL	Name of the detected WAF/anti-bot technology (e.g., Custom In-House Anti-Bot / State Firewall).
tls_ja4_signature	TEXT	NULLABLE	TLS JA4 fingerprint signature observed during connection.
delivery_status	TEXT	NOT NULL	Status of the data delivery for this record (e.g., SUCCESS, PARTIAL, FAILURE).

Sample Record (JSON)

{
  "timestamp_utc": "2026-09-24T14:35:01Z",
  "sku_id": "1234567890",
  "ean_barcode": "4601234567890",
  "product_title": "Смартфон Xiaomi Redmi Note 12 Pro 8/256 ГБ, черный",
  "brand_name": "Xiaomi",
  "manufacturer": "Xiaomi Inc.",
  "category_hierarchy": "Электроника > Смартфоны и гаджеты > Смартфоны",
  "price_ozon_card_rub": 22990.00,
  "price_standard_rub": 25990.00,
  "price_rrp_rub": 29990.00,
  "stock_level": 150,
  "buybox_seller_type": "Prodavets Ozon",
  "ozon_premium_eligible": true,
  "fulfillment_mode": "Ozon Rocket Courier",
  "customer_rating_stars": 4.8,
  "total_reviews_count": 1245,
  "technical_specs_json": {
    "Model": "Redmi Note 12 Pro",
    "Вес": "187 г",
    "Размеры": "162.9 x 76 x 7.9 мм",
    "Страна производства": "Китай",
    "RAM": "8 ГБ",
    "Storage": "256 ГБ"
  },
  "velocity_depletion_rate_24h": 5.2,
  "target_domain": "ozon.ru",
  "detected_waf": "Custom In-House Anti-Bot / State Firewall",
  "tls_ja4_signature": "t13d160201080005070007070006060