Distribution
How projects reach judges: agent screening, the gate-failure inbox, load split, daily batches and the failed-gates rule.
/h/[slug]/setup/distributionWhoSigned-in organizersFind itSetup › DistributionDistribution decides how projects reach judges when each phase starts. Phases set how many judges review each project. This page sets the rest: whether the agent reviews every project, who handles the agent's gate failures, how the work is split, how fast queues open, and what happens when judges fail a gate.
Every judge gets a private link that shows only their own queue. The settings here apply to every phase.
Defaults#
| Setting | Default |
|---|---|
| Agent first | On |
| Agent-failed inbox | You, the hackathon owner |
| Strategy | Even split |
| Cadence | All at once (5 days if you switch to daily batches) |
| Show agent score | On |
| Failed gates | Rule it out after 2 failing reviews |
Change any setting and click Update in the page header to save.
Agent first#
When on, the agent reviews every project as soon as judging starts. It runs every criterion's mechanisms and checks the gates. Judges get their queues at the same time and don't wait for the agent, so a judge can open a project before its review is done.
Agent-failed inbox#
This setting only applies with Agent first on. A project the agent fails on a gate ranks last and can't win until someone looks at it. Pick who that is:
| Option | What happens |
|---|---|
| You (under Admins) | Gate failures wait in your Inbox for a call. |
| A judge | Gate failures wait in your Inbox, which names that judge as the one who reviews them. Judges have no inbox on their link, so the call is made from yours. |
| Nobody — drop them | Failures stand: those projects rank last and can't win. |
From the Inbox in the top bar, you either agree with the agent and keep the project out, or disagree and put it back in the pool. If you delete the judge who has the inbox, it goes back to you. See Failed inbox.
Strategy#
- Even split gives every judge the same number of projects.
- Mixed keeps the load the same and also rotates pairings, so no two judges share every project. That lets you compare scores across judges.
Cadence#
- All at once gives each judge their whole queue from the start.
- Daily batches opens a new batch for each judge every 24 hours. Set how many days with Spread over, from 1 to 30.
With daily batches, each judge gets an equal share of their queue each day, and can start the next batch early. Spreading over 1 day is the same as all at once. Short daily queues stop judges from rushing everything on the last day.
If email is set up on your server, each time a batch opens, judges with new projects in it get an email. When the phase's days are up, judges with projects left get a last call, and you get a note that the phase is due.
Show agent score#
When on, a judge sees the agent's score for a project only after they submit their own, so it can't anchor them. They can then adjust their score once.
Failed gates#
This rule covers gates that judges fail in their reviews. (The agent's gate failures go to the inbox above.)
- Flag it: a review that fails a gate flags the project, and you decide.
- Rule it out: once enough reviews fail the same gate, the project is ruled out. Set the number with Rule out after, from 1 to 20 failing reviews. Fewer failing reviews than that flag the project for you instead.
A ruled-out project ranks last, is eliminated when its phase closes, and can't win an award.
Preview#
The Preview panel under the settings shows what your phases and settings add up to:
- With Agent first on, how many projects the agent screens, and who gets gate failures.
- For each phase, the projects coming in, times reviews per project, divided by the judges in its pool, and the result: how many projects each judge gets.
- A rough time per judge, at about 4 minutes per project, and a per-day count with daily batches.
- no judges in pool when a phase's group is empty.
Click Edit phases to change the phases themselves. The numbers fill in once projects are added.
