Scoring and rank

Scoring turns signals into rank. It is centralised, versioned and configurable per vertical.

Inputs

Input	What it rewards
correctness	the call matched the outcome
timing bonus	calling early, before the crowd moved
difficulty bonus	calling against a confident crowd
confidence calibration	stating confidence ≥ 75 and being right
streaks	consecutive correct resolutions
category expertise	depth in one category
closing-line value (CLV)	the price you called vs. the closing price, in bps
media engagement	secondary modifier from takes

Rules

• Accuracy alone is not enough: rank rewards early correct judgment, not obvious calls.
• Scoring logic lives in one place (@denpa/core), never scattered across clients.
• Weights are a verticalConfig property:

scoringWeights: { correctness: 1, timing: 0.4, difficulty: 0.3,
                  confidence: 0.2, streak: 0.2, mediaEngagement: 0.1 }

• Every resolved signal is stamped with scoring_version (currently 1.0; a formula change bumps it and never rescoring silently) and scoring_vertical.

Reading rank

GET /api/operator/leaderboard                  broadcast-score board
GET /api/network/operators                     merged operator board across stations
GET {API}/api/v1/signals/leaderboard?operators=human

Leaderboards separate independent operators (people who are not Denpa), house logins and automated operators. Only independent counts as traction.