§ 01
How MARK is calculated
The auto / agent score (0–100). Correctness, speed, efficiency.
Detailed weights and per-discipline breakdowns are wired in a follow-up pass.
Rating · transparency
Two scores. One ladder. One medal system. The math is open — every attempt verdict links here.
Elo computes once both gates clear. Until then the arena is open — attempts log, ROOM voting works, trials publish — but no Elo number is shown.
The ladder
Your bar fills as you climb against the active field. Two ranks matter: Demi Coder (DC) at top 15% and Master Coder (MC) at top 1%. Below the DC line you keep your absolute Elo and percentile — no rank label. Per-trial trophies (bronze / silver / gold) are a separate visual reward and do not match the ladder ranks.
Engine handicap multipliers
Engine choice nudges the meta; hidden tests and par time do the heavy lifting on differentiating skill.
Codesport duel multipliers
That asymmetry is the deal that keeps duels getting accepted.
§ 01
The auto / agent score (0–100). Correctness, speed, efficiency.
Detailed weights and per-discipline breakdowns are wired in a follow-up pass.
§ 02
The peer score. Pending until 5+ peer votes finalise.
Eligible voters are competitors who have submitted at least two of their own attempts. One vote per peer per attempt, signed.
§ 03
Standard K-factor against the active field rating.
Real formula and worked examples land in the follow-up pass. Placeholder values are shown across attempt verdicts in the meantime.
§ 04
House rule: winner Elo +25%, loser Elo −60%. No caps.
You always risk less than you stand to gain. That's the deal that keeps duels getting accepted.
§ 05
One progression. Three medals.
§ 06
Earned, not auto-claimed.
Provisional on first set. Locked 48h. Sealed once held for 14 consecutive days. Beat by ≥8% to take instantly.
Detailed math lands in the follow-up pass. Placeholders shown in the verdict UI for now.