Elo System
Chronomancy has two reputation systems that coexist and reinforce each other:
| System | Axis | Purpose |
|---|---|---|
| Chrono Score | Forecaster vs. market | Financial signal — determines FF prices, Rewind discounts |
| Elo | Forecaster vs. forecaster | Social/competitive signal — determines seasonal ranking, Wizard Duel outcomes |
They are not the same thing, and intentionally so.
CS vs. Elo: The Distinction
Section titled “CS vs. Elo: The Distinction”CS measures absolute accuracy. Did your 70% prediction resolve YES 70% of the time? CS is a calibration score — you are competing against the market probability, not other forecasters.
Elo measures relative performance. Did you beat the other forecaster in the same market? Elo is a zero-sum tournament score — you gain when others lose.
A forecaster can have high CS (excellently calibrated) but mediocre Elo (they make correct predictions that everyone else also makes — they don’t beat the crowd). A different forecaster could have high Elo (they consistently beat specific opponents) but moderate CS (they’re contrarian, not perfectly calibrated).
Both signals matter. CS gates financial products. Elo gates competitive prestige.
Elo Mechanics
Section titled “Elo Mechanics”Rating System
Section titled “Rating System”Standard Elo with K-factor calibrated for prediction market dynamics:
- New forecasters start at 1200
- K=32 for forecasters with <30 predictions; K=16 thereafter
- Seasonal reset: all Elo scores partially decay toward 1200 at season end (σ=200 regression)
Sealed Predictions
Section titled “Sealed Predictions”The key design decision: Elo matchups use sealed predictions.
Both forecasters submit their probability for the same market. Predictions are locked (sealed) before either is revealed. After market resolution, both predictions are revealed simultaneously, and Elo updates based on who was more accurate.
This solves the information-hiding problem: if Elo rewarded beating opponents in the live market, forecasters would have incentive to hide their true beliefs (to prevent opponents from trading against them). Sealed predictions mean there’s no information advantage from knowing your opponent’s position — you submit your best honest probability.
Adversarial Collaboration
Section titled “Adversarial Collaboration”Beyond pure competitive Elo, Chronomancy supports adversarial collaboration duels: two forecasters publicly commit to opposing positions, share their reasoning, and the community rates the quality of the reasoning. The Elo update in this mode is influenced by reasoning quality scores, not just who was right.
This rewards how you think, not just that you were right — a meaningful distinction when both forecasters make good probabilistic arguments and one just happened to be correct.
Seasonal Resets
Section titled “Seasonal Resets”At the end of each LOOP season:
- All Elo ratings regress 50% toward 1200
- This prevents permanent stratification — new forecasters can challenge established ones without impossible Elo gaps
- Glicko-2 RD resets independently (uncertainty increases for all)
The seasonal structure means every season is a fresh competitive landscape, while the persistent Chrono Score means accumulated skill carries forward financially.
Elo vs. CS on the Same Leaderboard
Section titled “Elo vs. CS on the Same Leaderboard”The public leaderboard displays both CS (calibration) and Elo (competitive rank) side-by-side. Users can sort by either. This creates two distinct paths for reputation:
- The precision path: optimize for calibration (CS), access better financial terms
- The competitive path: optimize for head-to-head wins (Elo), access competitive prestige and seasonal bonuses
Many users will value both. Prediction market legends like Samotsvety (Metaculus) have both high absolute accuracy and strong competitive records — the two axes are correlated but not identical.
The Teams Problem
Section titled “The Teams Problem”Elo in team formats creates tribalism — “my team won” replaces “I was right.” Chronomancy deliberately avoids team Elo in the core protocol, keeping competition at the individual level.
Wizard Duels are the team-adjacent mechanic — but they are 1v1, not faction-vs-faction. The adversarial collaboration format (share reasoning, community rates quality) further discourages tribal dynamics by making individual reasoning quality the valued output.
Related:
- Wizard Duels — the 1v1 prediction duel spec
- Harm Reduction — Elo as a path from gambling to forecasting
- Loop — seasonal epochs that Elo resets within