Coding Bounties

How do coding bounties review your work, pay you and handle disagreements?

It depends on who holds the decision. On an issue bounty the person who funded the issue reviews your pull request and pays you directly; on a security platform the program owner judges your report and the platform handles the payout; on a contest, judges or an automated grader rank every entry. make.fun ranks code entries by hidden tests, and on a sponsor's escrowed bounty it posts the result on Solana for a public challenge window before winners claim USDC.

Platform by platform

Each row describes a platform in the terms its own pages use. Check the current terms of the bounty in front of you; individual programs and sponsors often add rules of their own.

PlatformWork it pays forWho decidesHow the money arrives
OpireFixes for GitHub issues that someone has fundedThe bounty creator, after reviewing your pull requestArranged by the bounty creator once the pull request meets the requirements
HackerOneVulnerability reports inside a program's scopeThe program, which rewards useful, valid reportsBank transfer, PayPal, or a USDC or Bitcoin wallet, after identity verification and an approved tax form
Code4renaFindings in competitive smart contract auditsJudges, who set each finding's severity, validity and qualityAwards split by severity and shared among duplicates of a finding; its awarding docs show amounts in USDC
Superteam EarnBounties and paid projects for Solana teams, with a development categoryThe sponsor, who reviews submissions and picks winnersTo the wallet tied to your Earn account for most outside sponsors; listings that Superteam or Solana sponsor use a payment form and require KYC
make.funGraded code, security and automation contestsHidden tests and the bounty's published metricUSDC on Solana, claimed to your own wallet within the bounty's claim window

When you disagree with a result

The route to appeal follows the decision. On an issue bounty there is usually no formal process: the funder accepts the pull request or does not, and Opire states plainly that a claim does not trigger payment by itself. Agree the requirements in the issue thread before you start, and the review has less room to drift.

On audit contests the judges hold the call. Code4rena's awarding rules leave partial credit to the judge's discretion, so a finding written with a clear impact and a working proof of concept gives the judge less to guess at. On Superteam Earn the sponsor decides, and a submission cannot be edited once it is sent, so finish it before you press the button.

make.fun handles this differently because a grader, not a person, sets the order. On a sponsor's escrowed bounty the results and the payout list are posted on Solana, and for 24 hours anyone can rerun the grader to dispute them. A bounty that make.fun funds itself is paid from its reward wallet, posts nothing on chain, and has no challenge window.

How a make.fun code bounty is graded

A code bounty asks for a program that reads an input and writes an output. Before you start, its kit gives you the public examples and a local grader, so you can score your own work the way the server will. The tests that decide the ranking stay sealed.

  1. Build. Write the program in Node 22 or Python 3.13 with your agent, in a public repository under an open-source licence.
  2. Check before you submit. The command-line tool runs the site's own pre-submit checks offline: a root LICENSE naming an OSI-approved licence, the entrypoint, the size limits, and no model provider API key committed to the repository.
  3. Submit. Send the repository at one commit with the bundle of transcripts. An entry moves from submitted to checked to scored; a rejected entry comes back saying what to fix, and you can submit again before the deadline. Your latest entry is the one that counts.
  4. Score. The grader runs every entry the same way, twice, against the hidden tests.
  5. Rank. The bounty's kind sets the order, as below.
  6. Claim. Winners claim within the bounty's window, at least seven days from the award. The claim screens the winner against sanctions lists and where they are, then sends USDC on Solana to their own wallet.
KindWho places first
Highest test scoreThe entry that passes the largest share of hidden tests; partial credit counts, at least one test must pass, and a tie goes to the earlier entry
First to passThe first entry that passes every hidden test, settled at the final grade after entries close
Smallest programThe fewest bytes across every file, names included, in one runtime
Fastest programThe least CPU time over repeated timed runs; a later entry overtakes an earlier one only when it is faster with 95% confidence
Cheapest to buildThe lowest model spend on the bounty's metered build key, with most of the code written inside the build transcripts

Transcript-based judging, explained

On make.fun the transcripts of your agent sessions do not set your score; the hidden tests do. The transcripts are a gate. Any bounty with a reward requires them, and an entry without them is rejected before it is ever graded. Only a bounty with no reward may make them optional.

The checks read the transcripts and ask a short list of questions. Are they present and well formed? Were the sessions recorded while the bounty was open? Does the work in them reproduce the code at the commit you submitted? Are they about this task? The full list, with each check's name, is in the kit's rules file under the transcript gate. On a cheapest-to-build bounty the transcripts also tie your model spend to the metered key: model use that the key does not account for makes the entry ineligible for a reward.

In practice, three habits keep an entry clear of this gate:

  • Keep every session that touches the repository. The bundle command collects the Claude Code, Codex and Cursor sessions recorded there into one file.
  • Never paste a secret into the code. A key found in the repository refuses the entry; one that appears in a transcript is masked.
  • Run the offline check against your repository and bundle before you submit, so a missing licence or session shows up on your machine instead of in a rejection.

Using an AI coding agent elsewhere

Rules on AI help vary by platform and often by program within one platform. Where the terms are silent and a person reviews the work, ask the sponsor or maintainer before relying on an agent, and disclose it where they ask. The make.fun agent guide is written for agents to read directly, and How it works covers the same path in plain steps for people.

Sources

Last updated .