What it is
On any rubric-driven score sheet, judges see an AI suggest scores button. One click returns Claude's score for each criterion with a one-sentence justification grounded in the nomination's actual content (free-text answers, supporting docs, attached writing samples). The judge can accept the whole proposal, edit any number, or ignore it entirely.
The judge always submits the final scores. The AI suggestion is a draft; nothing is recorded until the judge clicks Submit review.
When to use it
- Calibration on round 1. New judges burn ~30 minutes per nomination calibrating their internal scale. AI suggestions give them a starting point so they can spend that time questioning the suggestion rather than building a baseline from scratch.
- High-volume rounds. Public-vote shortlists where the same five rubric items are scored across 200 nominations. The judge reviews each suggestion in seconds and adjusts only the outliers.
- Plagiarism / over-reach checks. Judges often spot AI-written submissions faster when comparing against an AI's own read of the content — the AI's confidence score on the rubric drops sharply when the writing is inconsistent.
When NOT to use it
- Final-round judging. The whole point of the final round is human deliberation. The button is disabled by default on rounds marked FINAL.
- Sensitive categories (e.g. Lifetime Achievement, Best Adaptation) where the judge's lived perspective is what you're paying for. Producers can disable the button per category.
How it's built
- Claude reads the nomination's free-text fields, the rubric definition, and the show's tone (set by AI Show Director).
- Each criterion gets a score in the rubric's range (typically 1–10) and a
reasoningstring. - The proposal is stored as a
ReviewSuggestionrow keyed to the judge + nomination. If the judge submits, the suggestion is preserved alongside the actual review for analytics. - Judges' edits are tracked: we can show producers a "% of AI suggestions accepted unchanged" metric per judge, useful for calibration training.
Cost
One credit per nomination per judge, billed against your AI credit pool. Roughly $0.02/judge/nomination at current Claude rates. Producers can cap the per-show spend in Settings → AI.