I run a small radar that lists paid GitHub issues (Algora, Opire, IssueHunt, BountyHub, TaskBounty). Recently I audited my own feed and found it was wrong in an instructive way. Here is what I got wrong, and the check that fixes it.

The mistake: trusting text on the issue

My scanner treated "/bounty $700" in:body as proof of an Algora bounty. That query matches roughly 67,000 open issues on GitHub. Of the 97 issues I had flagged as verified Algora bounties, 83 (86%) had no comment from the real Algora app. That was about $18,900 of money that didn't exist.

The three patterns of fake

  1. Amounts typed by hand on throwaway repos. Anyone can write /bounty $700 in an issue; no money is behind it.
  2. Look-alike bots. One bot posts "💎 $9400 bounty" in exactly Algora's format, but it isn't Algora's app.
  3. "AI only" farms. Dozens of issues with labels like "AI only allowed - no humans", created by the same author, some even using the real bot.

The signal that works

The real Algora app comments "💎 $X bounty" on the issue, from the account algora-pbc. So this search returns the real set:

commenter:algora-pbc state:open type:issue 

Today that is about 33 open issues, not thousands. Parse the amount from that comment, not from the issue body.

What else to check before you spend hours

  • The issue is still open and unassigned.
  • No merged PR already references it.
  • The repo is active and not brand new.
  • The thread isn't a swarm (dozens of competing attempts).

The result

My public numbers went from "104 bounties / $47k" to a handful of workable ones. It looks worse, but it's true, and it saves hours: a big number is not a verified number.

I turned these checks into a free live board (numbers deliberately coarse): https://bountyos.rovidev.com/en/github-bounty-board/?utm_source=devto&utm_medium=article&utm_campaign=antifraud

More on spotting fakes: https://bountyos.rovidev.com/en/blog/how-to-spot-fake-github-bounties/?utm_source=devto&utm_medium=article&utm_campaign=antifraud

What other cheap ground-truth signals do you use to avoid wasted effort? I'd like to hear them.