Papers

Qais Alassa · Osama Alashqar

Research Paper

Calibration-Gated Reputation

Strictly Proper Scoring Rules as Trustless Merit Filters in Decentralized Prediction Systems

We introduce calibration-gated reputation: a mechanism in which agents commit cryptographic hashes of probabilistic predictions, reveal them after a mandatory delay, and accumulate reputation through a strictly proper scoring rule. We prove that strict propriety is both sufficient and necessary for this construction. Five theorems validated through 10,000 Monte Carlo trials.

16 pages · 5 theorems · 30 references · Targeting AFT 2026

Download PDF Read Summary Simulations (GitHub)

Technical Specification

Notch Protocol

A Cryptographic Framework for Verifiable Prediction Scoring and Alpha Commoditization

The companion technical specification covering the commitment scheme, Brier score calibration, difficulty adjustment, the instrument layer (Alpha Passes, Futures, Indices), dual oracle architecture (Chainlink + Pyth), and $NOTCH token economics.

DOI: 10.5281/zenodo.19118356

Download PDF View on Zenodo

Key Results

Theorem 3.1

Strict propriety is necessary

Under any proper but not strictly proper scoring rule, Sybil adversaries can produce wallets indistinguishable from skilled predictors.

Theorem 3.3

Accuracy-only scoring is insufficient

Reputation based solely on directional accuracy admits polynomial-cost Sybil attacks regardless of prediction count.

Theorem 4.1

Merit-gating

Under the Brier Score with commit-reveal, no Sybil strategy produces high-calibration wallets without genuine skill. Validated: 0/10,000 trials.

Theorem 4.3

Sybil cost amplification

The cost of faking reputation scales at least linearly in Sybil identities. Empirical: 264x cost ratio at K=1,000.

Theorem 4.5

Stationary convergence

Under temporal decay, each predictor's score converges to a unique value monotonically increasing in skill.