Every paid task has the same hard question: who decides the work is good enough to pay for? I've been earning small amounts from three different answers, so here is a side-by-side. I'm a student and open source contributor (Rust, Node.js), not a Verdikta employee, and one of these systems paid me, so read with that in mind.
The three approaches
1. Verdikta: a rubric plus an AI jury, paid from on-chain escrow
On Verdikta Bounties the creator writes a rubric with weights, locks the reward in an escrow contract, and two AI models (one from OpenAI, one from Anthropic, 50% weight each) score every submission. If the weighted score passes the threshold, the payout is released. The page for each bounty states that the escrow can't change terms or return the funds before the deadline.
A real example: bounty #139 paid 0.01 ETH on a 50% threshold; my submission scored 91%, and the jury's reasoning is public on IPFS.
Strengths
- The rules are visible up front: criteria, weights and threshold are all on the page.
- Money is locked before you start, so "the client never paid" isn't a risk.
- Fast, and no negotiation or bidding.
- The reasoning is stored publicly, so you can see why you scored what you scored.
Weaknesses
- Models can be inconsistent, and "authenticity" style criteria are judged by feel.
- Thresholds can be strict (I'm seeing 90% on some bounties), and a near miss pays nothing.
- Payouts are small (0.001 to 0.01 ETH on the bounties I've seen), so it suits short, well-defined tasks, not open-ended work.
- You need a crypto wallet.
2. Traditional freelance platform (Upwork-style): a human client decides
A client posts a job, freelancers send proposals, the client picks one and judges the result.
Strengths
- Humans can judge messy, subjective work and negotiate scope.
- Large market with long-term relationships, reviews and repeat clients.
- Dispute process with a platform behind it.
Weaknesses
- You pay before you earn: proposals cost "Connects" (15 for one job I applied to), and the platform keeps about 10% of what you bill.
- A new account has no reviews, so it competes against 20 to 50 or more proposals per job.
- Some listings are noise. I saw dozens of near-identical Web3 posts from clients that turned out to be suspended.
- Slow: days between proposal, interview and payment.
3. Maintainer review on open source bounties: one person decides
A maintainer reads your pull request, merges it, and (hopefully) pays.
Strengths
- The reviewer knows the codebase better than any outside judge.
- Your work lands in a real project, which is worth something on its own.
Weaknesses
- Payment depends on one person being active and honest. Several "bounty" issues I looked at never said who pays.
- No locked escrow in most cases, so a merged PR can still go unpaid.
- Waiting time is unpredictable.
Side by side
| Verdikta | Freelance platform | Maintainer review | |
|---|---|---|---|
| Who judges | 2 AI models + rubric | The client | One maintainer |
| Money locked first | Yes (escrow) | Sometimes | Rarely |
| Cost to try | Gas (small) | Connects + fees | Your time |
| Speed | Minutes to days | Days to weeks | Days to never |
| Best for | Short, well-defined tasks | Long, fuzzy work | Real code in real projects |
| Main risk | Model inconsistency | Competition and cost | Unpaid work |
My take
None of these wins everywhere. The AI jury is the most transparent (you can read the rubric before you start and the reasoning after) and the only one where the money is locked up front. It's weaker where judgment is fuzzy or the task is large. A freelance platform is better for long relationships, and maintainer review is better when the goal is to ship real code.
If you only try one thing, try a small, well-defined bounty and read the rubric first. The rubric tells you exactly how you'll be judged.
Links: Verdikta Bounties and the bounty used as the example, #139.
