Scoring and judging
Status: Planned (programme)
Each Humanathon entrant is scored out of 100 from five components, and 60 of those points come straight from verified data. People judge only what data cannot. Judging runs in 3 layers, and the leaderboard shows scores and percentiles, never raw income.
A programme, not software
Humanathon is a programme that Merces runs, with rules and events. It is not part of the Merces software, and entering never obliges anyone to raise. Programme rules marked as planned may change before the season opens.
The five components
| Component | Points | How it is measured | Set by |
|---|---|---|---|
| Verified output | 35 | Income in the window from independent counterparties. Half on the amount, half on growth over the baseline. The baseline is floored, so a zero-income entrant cannot post infinite growth. Both as percentiles within the track. | The oracle (data) |
| Mandate delivery | 25 | Did you ship what your frozen mandate said? | Peer findings, then the jury (people) |
| Forecast accuracy | 15 | The distance between your actual income and the midpoint of your sealed forecast, penalised both ways. Aiming low costs as much as missing high. | The oracle (data) |
| Quality | 15 | The work itself, under adversarial review. | Peer findings, then the jury (people) |
| Reliability | 10 | All 6 weekly proofs on time, a public log kept, and challenges answered. | The oracle (data) |
Verified data sets 60 points; people set 40. An integrity breach is disqualification, not a deduction. The conviction book carries 0 weight in the score: the market and the judging stay separate.
Forecast accuracy is the component most specific to Merces. A backer in an income share is buying predictability as much as size, so the season rewards entrants who know their own numbers.
The three judging layers
- Automated. The oracle scores verified output, forecast accuracy and reliability for everyone. The top 30 go on to peer review.
- Peer adversarial review, which lasts 1 week. Each of the top 30 is assigned 3 others at random, who try to break their claims. Anyone can file a staked public challenge. Reviewers file findings, not scores. An upheld finding pays a bounty; a frivolous one loses its stake. The top 15 advance.
- A paid jury on Proof Day. 5 to 7 members: operators, accelerator partners, at most 1 sponsor seat, and 1 non-voting Merces seat. The jury rules on findings and scores delivery and quality, capped at 40 points. The top 5 become Fellows; places 6 to 15 are Finalists.
Proof Day is live and public. Jurors disclose conflicts and recuse, and a sponsor that stakes behind an entrant cannot hold a jury seat for that track.
What is public, and what never is
| Public | Never public |
|---|---|
| Scores and percentiles, on the leaderboard | Raw income |
| The frozen mandate | The sealed income forecast, which is revealed only as the accuracy score |
| Peer findings and jury rulings | What the jury sees in detail |
The jury sees detail. Backers see it only for Fellows who choose to raise. Scores are percentiles within each track, so tracks compare fairly.
What can go wrong
- An integrity breach disqualifies. It is not a deduction.
- Income from wallets linked to you, from yourself, or routed back through friends is excluded from verified output. Fair play lists every exclusion.
- A frivolous challenge loses its stake.
- Jury points are capped at 40, and the conviction book carries 0 weight, so popularity cannot lift a score; but the people-scored components are still judgement, and judgement can be challenged.
- Prize money from the season never counts as income.