Project: CrowdVerify (Research Prototype) Author: Jithin Mathews Status: Exploratory / Conceptual (Non-production)
This document describes the conceptual reward algorithm used in CrowdVerify to incentivize contributors to submit valuable evidence during real-world events.
The goal of the reward system is not to determine truth or guarantee fairness, but to:
- Encourage timely, high-signal contributions
- Discourage spam, redundancy, and low-effort submissions
- Adapt incentives dynamically as events evolve
- Make incentive behavior analyzable through simulation
This algorithm is designed as a research artifact and is intentionally conservative in scope.
The reward system follows these guiding principles:
-
Marginal Value Over Volume Contributors are rewarded for adding new information, not for submitting more data.
-
Time Sensitivity Early, high-quality evidence is more valuable than late or redundant evidence.
-
Demand-Driven Incentives Rewards scale with event uncertainty and civic impact, not popularity.
-
Diminishing Returns As supply increases, the reward for additional evidence decreases.
-
Reputation-Weighted Influence Historical reliability affects reward magnitude but does not fully exclude new contributors.
-
Auditability Over Optimization The algorithm favors transparency and replayability over maximal efficiency.
Event demand represents the current need for verification and is modeled as a function of:
- Event recency
- Conflicting claims or reports
- Geographic or civic impact
Demand is highest during early or highly uncertain phases and decays over time.
Supply represents the current volume and redundancy of submitted evidence:
- Number of submissions
- Similarity between submissions
- Geographic clustering
High supply reduces marginal reward value.
Novelty captures how much a submission contributes new information:
- Independent vantage point
- Distinct timing
- Unique metadata patterns
Novelty is evaluated heuristically and produces a normalized score.
Contributor reliability is an evolving signal based on:
- Historical accuracy
- Dispute resolution outcomes
- Consistency over time
Reliability influences rewards multiplicatively but is capped to avoid entrenched dominance.
Time weighting reflects that:
- Early submissions are more valuable
- Late submissions face diminishing impact
Temporal decay is smooth rather than abrupt to avoid cliff effects.
At a high level, rewards are modeled as:
Reward ∝ Demand × Novelty × Reliability × TimeWeight × SupplyAdjustment
Where:
- Each factor is normalized
- SupplyAdjustment decreases as redundancy increases
- The output represents relative reward magnitude, not currency
This formulation allows deterministic replay in simulation.
The algorithm adapts incentives across an event’s lifecycle:
- High demand
- Low supply
- Strong rewards for primary evidence
- Rising supply
- Conflicting claims
- Rewards favor independent confirmation
- Low marginal demand
- High redundancy
- Rewards shift toward synthesis and dispute resolution
This prevents late-stage spam and reward exploitation.
The design explicitly considers failure modes:
- Spam Flooding: Mitigated via diminishing returns
- Collusion: Reduced through diversity and novelty checks
- Sybil Attacks: Partially mitigated via capped reputation influence
- Early Capture: Temporal smoothing reduces winner-take-all effects
These mitigations are incomplete by design and are evaluated via simulation.
Because real-world validation is not feasible, the reward algorithm is tested through simulation:
- Synthetic events with varying demand curves
- Contributor behavior models (honest, opportunistic, malicious)
- Controlled attack scenarios
Metrics observed:
- Reward distribution skew
- Signal-to-noise ratio
- Abuse success rates
- Convergence time
- Identity and uniqueness assumptions are weak
- Reputation bootstrapping remains unresolved
- Novelty detection is heuristic
- Ethical implications of financial incentives require further study
These limitations are intentionally documented.
This reward system does not aim to:
- Guarantee correctness
- Replace editorial judgment
- Eliminate misinformation
- Optimize contributor earnings
The reward algorithm in CrowdVerify is treated as a policy-aware, time-sensitive incentive mechanism rather than a payout formula.
Its primary value lies in enabling experimentation, simulation, and critical evaluation of incentive behavior under uncertainty.
This document represents a personal research exploration and should not be interpreted as a production-ready economic model or policy recommendation.