Glossary
AI Criterion Score. The AI baseline for one dimension, produced by combining the judge reads that cover it with their routing weights. Read-only.
AI judge panel. The set of independent AI reviewers that read every submission. Six lenses in Pitch Competition mode, five in Hackathon mode. They never see one another's scores.
AI Total Score. The advisory 0–10 number across all dimensions, produced by deterministic aggregation. It never ranks the batch.
Anchor. The description of what a given band looks like on a given dimension — what a 3 looks like against a 7. Anchors are what turn a ten-point scale into a scale judges use the same way.
Band. A range on the scoring scale (0–3, 4–6, 7–8, 9–10) with its own anchor description. A judge names the band before picking a number inside it.
Batch. All entries in a project, evaluated in one run rather than deck by deck.
Confidence. A per-dimension signal of how well the evidence supports a read. It can apply a downward adjustment of at most 15%, and never an upward one.
Deck completeness. A check across ten core sections — Problem, Solution, Market, Business Model, Traction, Team, Roadmap, Financials, Ask, Other — marking each present, thin or missing with a severity. Not a fact-check.
Deck triage. Screening a large field to decide what deserves a full human read. The job EvalLens's first read is built for.
Deterministic aggregation. Combining judge outputs into a score with fixed arithmetic and no model call, so the same inputs always produce the same total.
Dimension. One of the six fixed questions a deck is scored on (P1–P6).
Entry. One team's submission: the deck, the team and project details, an optional note for judges, and a status.
Entry Hub. The intake side of EvalLens — manual entry or a public submission page with access rules and a window.
Evidence-grounded scoring. Requiring a slide reference behind every claim a score relies on. No slide, no claim.
Final Score. Your criterion weights applied to the Jury Score. The number the leaderboard sorts on.
Judge contribution matrix. The report view showing which judges contributed to each dimension and where strong disagreements were flagged.
Jury Score. The human score, set per dimension from 0.0 to 10.0. The only input to the ranking.
Leaderboard. The ranked view of a batch, built from submitted Jury Scores and project weights.
LLM-as-a-judge. The general practice of using a language model to score outputs against criteria. EvalLens's version differs in three ways: several independent lenses rather than one verdict, evidence required before a number, and the arithmetic on top removed from the model entirely.
Mode. Chosen when a project is created; determines the panel and the rubric shape. Pitch Competition or Hackathon.
Project. One selection round, with its own criteria, weights, judges and intake settings.
Review Board. The decision workspace: the whole batch with statuses, scores, findings, comparison and the leaderboard.
Routing weight. How much a judge's read counts toward a dimension — primary 1.00, secondary 0.50, advisory 0.25, or not scored.
Score spread. Highest judge score minus lowest on a dimension. Under 1.5 consensus, 1.5–2.99 split, 3.0 or more conflict. It routes attention and never changes a score.
Security signal. A flag raised when an instruction aimed at the model is detected inside a deck. The instruction is excluded from scoring evidence and shown to the organizer.
SourceRef. The link from a finding back to the slide it came from — number, title and note.
Submission (billing). One deck that received one successfully generated evaluation report. Unprocessable decks do not count.
Next steps#
- Core concepts — the same terms in the order they appear in a run.
- How the score is built — where each of the score terms enters the arithmetic.