Pairwise allocation leverages relative preferences to streamline the cognitively taxing process of allocating resources among many projects. Participants compare pairs of projects (e.g., “A is better than B”) rather than evaluating all projects simultaneously or assigning absolute scores. These pairwise comparisons are aggregated into a graph structure where preferences “flow” through the network until reaching a steady-state global ranking. This global ranking determines how funds are distributed among the projects. Unlike traditional allocation mechanisms, pairwise preference simplifies decision-making and enables more granular and nuanced evaluations, especially for large communities.

First proposed in 2018 by Daniel Kronovet, Aron Fischer, and Jack du Rose from Colony. Pairwise was initially conceptualized as a variation of Google’s PageRank algorithm adapted for capital allocation. In 2023-2024, General Magic implemented the model for Optimism’s Retroactive Public Goods Funding (RetroPGF) rounds, iterating its design over several cycles.

Advantages

• Cognitive Simplicity: Reduces mental load by breaking down complex allocation decisions into simple binary choices.
• Preference Discovery: Helps surface high-quality but lesser-known projects by enabling direct comparisons regardless of reputation.
• Signal Quality: Produces more nuanced distributions by focusing on relative value rather than absolute scores

Limitations & Risks

• Conversion Challenge: Translating relative preferences into absolute allocations can feel opaque and reduce participant trust in outcomes.
• Decision Volume: The minimum number of required comparisons for the outcome to represent global preference is not always clear.

Design Considerations

• Evaluators Assignment: Decide whether evaluators must make a fixed amount or unlimited evaluations. More pairwise comparisons per evaluator will increase the quality of the overall distribution but at the cost of more effort needed per evaluator. Consider using random sample assignment where evaluators are assigned a specific subset of all choices to reduce the potential for bias and popularity contests.
• Tiered Sorting: Implement initial quality-based grouping to reduce total comparisons and focus participant attention on meaningful choices.
• Information-Maximizing Pairs: Design adaptive pairing algorithms that prioritize showing project combinations that will yield the most valuable preference data.
• Interface Clarity: Create intuitive visualizations showing how pairwise preferences flow through the network to build allocation outcomes.