Key concepts
The words the app uses — schema blocks, criteria, phases, gates, assignments — and how they fit together.
RoundOne uses a small set of terms across its pages. This page defines each one and shows how they connect.
How the pieces fit#
- A hackathon has a schema: the blocks every project fills in.
- Criteria say what projects are scored on. Each criterion reads some of the schema blocks, and the agent checks it with one or more mechanisms.
- Judging runs in phases. When a phase starts, its judges get assignments. The agent writes an agent review of each project, by default starting when judging starts. Judges don't wait for it.
- A gate is a criterion every project must pass. Projects the agent fails on a gate wait in the failed inbox.
- Closing a phase advances its top projects. Closing the last one gives every remaining project a final rank.
- Reward tiers turn ranks and awards into winners, and publishing the results puts them on a public page.
Hackathons#
Hackathon#
One event you run, with a name, tagline, dates, color, logo and banner. Its slug, made from the name when you create it, appears in its URLs: /h/[slug] for you and /w/[slug] for the public winners page. A hackathon belongs to the organizer who created it. Other organizers on the same instance don't see it.
Stage#
Where a hackathon is in its life: Setup, Judging or Results. A new hackathon starts in Setup. Start judging moves it to Judging, and closing the last phase moves it to Results. The home screen shows the stage as a status such as "Setup — draft", "Judging — group review" or "Results published".
The Setup, Judging and Results switch in a hackathon's header opens any area at any stage. Switching areas doesn't change the stage.
Submissions#
Schema block#
One field of a project submission, set up in Setup › Schema. A block has a title, a type, a description and a note on what's expected. The types are Text, Long text, Number, URL, Video, Repo, File, Select, Image and Team.
The first Text block is the project's name, and the next Text or Long text block is its one-line pitch.
Project#
One team's submission: a value for each schema block, plus a contact email where results emails go. Each project gets a number, shown as #012, and a status: in the running, eliminated or disqualified. Projects come in by hand, through the public submission form, or through the Intake API. The project's audit trail records how it came in and what happened to it since: edits, moves between phases, scores and status changes.
Scoring#
Criterion#
One thing projects are scored on, set up in Setup › Criteria.
- Scale: Score 1–10 or Pass / fail.
- Weight: its share of the total, in percent. Weights should add up to 100%.
- Description: the rubric that judges and the agent score against.
- Inputs: the schema blocks it reads, or All inputs.
- If inputs are missing: Judge what's there or Score 0.
A review's total is the weighted score out of 10.
Mechanism#
How the agent gathers evidence for a criterion. A criterion can use several.
| Mechanism | What it does |
|---|---|
| Agent judge | LLM reads inputs, scores against the rubric |
| Sandbox run | Clones and builds the repo, runs tests |
| Video reviewer | Transcript, scene check, length check |
| Code scraper | Greps the repo for imports and call sites |
| Web scraper | Visits the live demo, captures pages |
| Human only | No agent score — judges only |
Model tier#
How much model the agent spends on a criterion, set on the criterion's Agent tab.
| Tier | What it means |
|---|---|
| Quick | Small, fast model. One pass over what the mechanisms found. |
| Balanced | Checks a few things itself with tools before it scores. |
| In-depth | Largest model, with room to dig: more tool calls, slower. |
The models behind each tier are set in the environment. See Configuration.
Double check#
Whether a second pass re-checks the agent's verdict on a criterion: Off, When unsure (low confidence, flags or a failed gate), or Always. The double check runs one tier up from the criterion's tier, so Quick is re-checked on Balanced and Balanced on In-depth. In-depth is already the top tier, so it's re-checked on In-depth again.
Gate#
A Pass / fail criterion with Gate eligibility turned on. Failing it makes a project ineligible, whatever its other scores.
- When the agent fails a project on a gate, the project goes to the failed inbox.
- When judges fail a project on a gate, the Failed gates setting in Setup › Distribution decides what happens: Flag it for you, or Rule it out once enough reviews agree.
A project that's ruled out ranks last, is eliminated when its phase closes, and can't win an award.
Judging#
Phase#
One round of judging, set up in Setup › Phases. A phase has:
- Judges: All judges, or one saved judge group.
- Reviews per project: a set number of judges per project, or Everyone reviews all.
- Advance: how many top projects go on to the next phase. The last phase picks the winners, so nothing advances from it.
One phase runs at a time. Closing it ranks its projects: gate passes first, then average judge score, then agent score, then submission order. The top projects move to the next phase and the rest are eliminated.
Judge#
A person who scores projects. Judges don't have accounts. Each one has a private link, /j/[token], that shows only their own queue. You add judges in Setup › Judges with a name, title, email, photo and any custom fields you define.
Judge group#
A saved set of judges, such as a final panel. A phase can go to one group instead of every judge.
Assignment#
One judge reviewing one project in one phase. Assignments are made when a phase starts, following Setup › Distribution:
- Strategy: Even split gives every judge the same number of projects. Mixed also rotates who is paired with whom, so no two judges share every project.
- Cadence: All at once, or Daily batches spread over a number of days. With daily batches, each judge's queue is split into one batch per day.
An organizer can also add their own review of a project on top of the judges' reviews.
Agent review#
The agent's review of one project. It runs each criterion as a step: it gathers evidence with the criterion's mechanisms, investigates with tools, scores, and double-checks when the criterion asks for it.
With Agent first on, an agent review is queued for every project when judging starts. Run agent on Judging › Progress reviews any project the agent hasn't. On a project's page you can rerun, override or flag a single step.
With Show agent score on, judges see the agent's score only after they submit their own, and can adjust theirs once.
Failed inbox#
Projects the agent failed on a gate, waiting for a person's call. Open it from Inbox in the hackathon's header once judging has started.
- Agree — keep it out: the failure stands. The project ranks below every project that passed and can't win an award. You can eliminate it at the same time.
- Disagree — back in the pool: the failure is cleared and the project ranks on its scores.
The Agent-failed inbox setting in Setup › Distribution decides who makes the call: you, one judge, or nobody (failures stand).
Results#
Reward tier#
A prize, set up in Judging › Rewards. A tier is either:
- Final rank: every project whose final rank falls in the range wins it, for example ranks 2 to 3.
- Award: a set number of winners, picked By hand or suggested by the Top on a criterion.
Each tier holds one or more items: cash, credits, a link, a code, text, an image, a file or swag. The One prize per team setting stops several entries from the same team from winning more than one reward.
Winner#
A project that wins at least one reward tier. Rank tiers follow the final ranking; award winners are picked on Results › Winners.
Published results#
Publish results on Results › Winners makes the winners page at /w/[slug] public. You can publish once the last phase has closed. Before that, only you can preview the page. Unpublish makes it private again. Judges' scores, notes and redeemable prize details such as codes stay private either way.