Elena Petrova
Data Science
Elena writes data studies and benchmarks, and about the quality of collected data: how to measure it, clean it and build datasets that hold up.
Writes about
- Data studies and benchmarks
- Data quality
- Dataset building
Articles by Elena Petrova
-
November 10, 2026 · Knowledge
Web Archives as a Data Source: What the Wayback Machine Can and Cannot Tell You
The Wayback Machine holds over a trillion captures. How to query it, where its coverage and quality fall short, and why more sites now keep it out.
-
November 6, 2026 · Knowledge
Common Crawl: When a Free Crawl Is Enough, and When You Need Your Own
Common Crawl publishes billions of pages a month for free. What it holds, what it leaves out, how to check your sources are in it, and when to crawl yourself.
-
November 5, 2026 · Knowledge
Same Product, Different Price: Geographic Price Discrimination Across 20 Countries
We compared list prices for 10 games across 20 countries on one day. The same game cost up to 4.1 times more in one market. The data, method and traps.
-
November 3, 2026 · Knowledge
Wikipedia urges AI companies to use its paid API, and stop scraping
Wikipedia asked AI companies for attribution and to use Wikimedia Enterprise, then tightened its limits. What changed and the proper routes to the data.
-
October 31, 2026 · Knowledge
Most Scraped Websites in 2026: A 30-Day Web Scraping Snapshot
Which websites get scraped the most in 2026? We pulled 30 days of public usage data for 8,186 ready-made scrapers and 239.8 million runs, then ranked the targets by demand and run success.
-
October 29, 2026 · Knowledge
Collecting Multilingual & Regional Web Data for LLMs
How to collect multilingual and regional web data for LLM training: plan coverage by language, find local sources, fetch the right locale, identify languages, filter per language and balance the mix.
-
October 26, 2026 · Knowledge
AI Training Data Marketplaces & Licensing: When to Buy vs Collect
How AI training data marketplaces and licensing deals work, what licensed data really costs you, and a practical framework for deciding when to buy a dataset and when to collect it yourself.
-
October 24, 2026 · Knowledge
Provenance in the AI Training Pipeline: Proving Where Your Dataset Came From
What AI training data provenance means in practice: the record to keep for every document, what the EU AI Act summary asks for, and how to carry it through.
-
October 19, 2026 · Scraping
Retry Economics: The Attempt Number Where Retrying Stops Paying for Itself
Each retry is less likely to succeed than the last. How to measure where retrying stops paying off, using your own logs and a few lines of Python.
-
October 15, 2026 · Residential Proxies
ASN Diversity in Residential Proxy Pools: How to Measure It
A free Python script to measure ASN diversity in a proxy pool: top-network share, HHI and effective ASNs, plus what our 17-network benchmark found.
-
October 11, 2026 · Knowledge
Do CAPTCHAs Stop Bots or Just Slow Them Down? Measuring Deterrence Against Solve Rate
What two decades of research says about CAPTCHAs: how long they cost humans, how often bots and AI agents solve them, and why they now work as a cost.
-
October 8, 2026 · Scraping
Payload Efficiency: What Share of a Scraped Page Is Boilerplate You Paid For
We measured 278 content pages from the top 1,000 sites. The text you want is about 1% of the HTML and a tiny share of what a browser downloads.
-
October 7, 2026 · Residential Proxies
Shifter Now Has the Largest Live US Residential IP Pool
We re-ran Shifter through our residential proxy benchmark: 212,212 live US IPs in 250,000 requests, the largest live US pool of the 17 networks we measured.