Posterior
α = 4 + weighted successes; β = 1 + weighted failures. The posterior mean is α ÷ (α + β).
Prediction target
What the estimate means
For participant u and proposed trade t, the model estimates the probability of acceptable completion given the participant's role, trade class, stake, duration, complexity, verification method, and assurance structure.
Acceptable completion means the material obligations were completed under the registered terms or a mutually agreed amendment, required evidence passed review, settlement obligations were met, and no attributable unresolved dispute or material deception remains.
Additionality and the no-trade baseline remain separate deal properties. A reliable participant can still offer a non-additional action, and a genuinely additional offer can come from a new participant with limited history.
Model formula and uncertainty
Bayesian calculation
Evidence accumulates with uncertainty preserved
Model v1 uses a Beta prior with 4 virtual successes and 1 virtual failure. Verified outcomes update that prior after evidence, recency, independence, context, and capped stake weights are applied.
Public score
100 × the posterior's lower 10th percentile. Sparse records therefore remain visibly uncertain.
Context transfer
Exact role and category evidence receives full weight. Same-role evidence receives 70%, same-category evidence 60%, and unrelated history only 25%.
Dimensions
Objective performance, separated by function
Free-form ratings can provide context, but they do not directly determine the core estimate. Platform or independently reviewed transaction events carry the score.
Fulfilment
45%Was the agreed commitment completed under the agreed deadline or amendment?
Evidence integrity
25%Was required evidence authentic, sufficient, and consistent with the registered terms?
Settlement compliance
15%Were payments, authorizations, or other settlement obligations completed correctly?
Dispute conduct
10%Did the participant cooperate with review and comply with the final resolution?
Responsiveness
5%Did the participant meet objective, agreed response requirements?
Observation weighting details
Event weighting
What makes one observation more informative than another
Proof quality
Platform-verified, independent, or adjudicated evidence receives full weight; bilateral confirmation receives partial weight; unilateral self-report receives little.
Recency
Evidence decays exponentially with a 365-day half-life.
Independence
Repeated transactions with one counterparty have sharply diminishing evidentiary value, reducing wash-trade and reciprocal-rating incentives.
Stake
Stake can increase evidentiary weight only logarithmically and is capped at 2×. Wealth therefore cannot purchase credibility linearly.
Safety separation
Severe integrity events are non-compensatory
Fraud, forged evidence, coercion, threats, identity duplication, and account compromise enter a separate eligibility system. They can require manual review or restrict an account; successful micro-transactions cannot wash them away.
- No cause, worldview, political position, follower count, or popularity input.
- No donation total or wealth proxy for virtue.
- No score reduction from an unadjudicated complaint alone.
- Raw event evidence and restriction reasons remain private by default.
- Every material event is designed to carry a reason code and appeal path.
Transaction safeguards
The estimate changes protection, not social status
Unproven
Small pilot, staged performance, independent evidence, and low unsecured exposure.
Developing
Moderate limits with standard or enhanced evidence.
Established
Higher limits for familiar trade classes, while preserving challenge windows.
Strong
Lower friction for low-risk contexts; irreversible or high-stake trades still require proportionate safeguards.
Validation state