The Price Was Never in the HTML: Font Obfuscation, and the Cracker That Recovers Every Glyph
Sites ship randomized fonts so your scraper stores garbage instead of prices. I built the obfuscator, cracked it: 11/11 glyphs recovered, code included.
Tag · 28 articles
Sites ship randomized fonts so your scraper stores garbage instead of prices. I built the obfuscator, cracked it: 11/11 glyphs recovered, code included.
Your scraper already collects searchable data. Ship it as a product: SQLite FTS5 with bm25 ranking, triggers, and zero infrastructure, tested at 50k rows.
Stop parsing HTML on Next.js sites. The data endpoints, embedded payloads, and flight streams are already there - tested tooling, real extraction, real numbers.
Reverse-engineer any protobuf or gRPC API without its .proto: decode the wire format in pure Python, infer a schema from traffic, and prove it by re-encoding.
Stock HTTP clients never send If-None-Match, so your monitors re-download unchanged pages at proxy rates. Tested validator store, A/B harness, and a challenge audit for lying ETags.
TV lags the venue by 5-10 seconds. I polled ESPN's scoreboard API through a live NFL touchdown and measured where the free feed is fast, and where it lies.
Build a $0 SERP API on self-hosted SearXNG: tiered engine pools, ban math, dedup + fusion ranking, and the honest benchmarks where it loses to paid APIs.
Anubis guards kernel archives and GNOME with SHA-256 puzzles. I built a solver, benchmarked every difficulty tier, and ran the crawler economics.
Leaked Playwright contexts cost ~93 MB of RSS each on Chromium 153. A tested fleet supervisor: crash recovery by liveness probing, deadline kills, and a zombie reaper.
Trap paths, seed fields and keyless decoys: a defender-side bot tripwire that catches scrapers and form spam at the app layer. Tested, 37 assertions green.
Scrapers die by 403, not 429. A tested client-side AIMD controller and circuit breaker that converges on the target's real rate limit instead of guessing.
Most scrapers bottleneck on the HTTP client, not anti-bot. Benchmarked aiohttp vs httpx vs requests, then built a tested async fetcher.