Every Telegram OSINT collection set starts as a pile of channels someone pasted from a Twitter thread. Most of that pile is junk, and the junk is expensive: every junk channel in your collection set adds review load, dedup noise, and false-alarm alerts forever.
I classify candidate channels on five axes before they enter the set. Score each 0-2; keep at 7+, review at 5-6, reject below.
1. Origin (0-2). Does this channel create reporting, or only republish? Check: original photos/videos with consistent watermarks, embedded metadata visible in preview thumbnails, posts that other channels quote-link. Pure republishers score 0. They're not useless - they're just not sources, they're distribution.
2. Language consistency (0-2). A channel that reports in one language with a stated target audience is a product; a channel that switches languages to chase engagement is a farm. Two lines of code: language-detect each post, flag the distribution.
3. Cadence regularity (0-2). Plot posts per hour. Real operations channels (military admins, emergency services, shipping feeds) post in bursty but structured rhythms with quiet nights. Farms post flat 24/7 at machine cadence, including 06:00 UTC spam. The autocorrelation of the posting series separates these cleanly.
4. Audience quality proxy (0-2). For channels above ~1000 subscribers the public preview exposes view counts. Compute view/subscriber ratio and the variance of per-post views. Real audiences have high-variance views (some posts matter, most don't). Purchased audiences are suspiciously uniform.
5. Link hygiene (0-2). Scan the last 20 posts' outbound links. Crypto giveaway domains, URL shortener services, gambling affiliates = 0. Links to primary sources (official statements, maps, registries) = 2.
The meta-point: each axis is computable from public preview HTML - subscriber series, view counts, post cadence, language, outbound links. No API keys, no logins, no paid data. The classification is boring, reproducible, and the reason a 12-channel curated set beats a 200-channel dump on alert quality.
The scoring sheet with all eleven rules (the six not shown here cover admin provenance, cross-post graph position, content-farm fingerprints, and more) ships in the Telegram & Web OSINT Bundle ($5). Free sample brief: here.
Collection runs free on GitHub Actions - no server, no paid APIs.
Top comments (1)
Some comments may only be visible to logged-in visitors. Sign in to view all comments.