# Jury Room

> Set up hackathon judging before anyone scores, then write the judging plan or the judge brief. Enter the teams
> expected, the confirmed judges with their types and conflicts, the judging window, the demo format, the brief's
> categories with weights and the prize tracks. The page works out the allocation and the rules; after scoring it ranks
> a pasted score sheet. A metered run writes the judging plan or the judge brief with one line per judge. Built on the
> SkillSafe skill @samber/hackathon-judging (samber/dev-event-organizer-skills, MIT). Prize amounts are out of scope.

https://jury-room.skillsafe.ai/

## What is free
- The allocation band: full coverage while N teams <= window W / minutes per team t (no judge count in that cap); a sampled panel with k = floor(J x W / (N x t)) judges per team; past J x W / t a pass/fail pre-screen then finals. Minutes per team default from the format (Expo 5, Stage 10, MLH's science-fair round 4).
- A named allocation map: judge, panel, start and end time, teams in visiting order, judge types mixed within panels, a contract-required sponsor judge confined to its own track, recusals routed to a replacement.
- The rubric (weighted criteria, integer weights summing to 100, at most four), the combination rule tied to the band (raw average, per-judge normalization, or synchronous consensus), the conflict rule (self-declared recusal, recusal plus replacement, or a written registry), the four-rung tie-break ladder and the six-item readiness gate.
- A score combiner: paste judge, team and one column per criterion; weighted totals, z-scores within each judge's slice under a sampled panel, ties walked down the ladder, and a sponsor track checked with its sponsor judges' scores dropped.
- Allocation-map CSV, a judge calendar (.ics, one event per judge with their teams in order), Markdown and JSON exports, and a per-judge copy of the brief with that judge's slot.

## The metered lanes
Two tasks on gpt-terra, metered per run, routed by `task`. Input string fields: `task` (plan or brief), `event`, `allocation` (the band, the arithmetic and the map) and `facts` (the page's picks, the readiness gate and every finding).
- `plan` replies with `summary`, `allocation`, `rubric`, `combination`, `conflicts`, `ties`, `announcement`, `signals` and `caveats`; the page holds the band, judges per team, both rules, every criterion and weight and the ladder to its own.
- `brief` replies with `brief` (the text sent with the calendar invite), `per_judge` (one line per judge on the map) and `note`; the page requires every criterion and weight, the abstain rule, and each judge's real start time and first team.

## Links
- API tutorial: https://jury-room.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@samber/hackathon-judging