Quadratic funding under constrained budget is suboptimal

This paper examines the optimality of Quadratic Funding (QF) mechanisms under individual budget constraints through both theoretical analysis and large-scale empirical data. Building on Weyl et al. (2019), we show that QF’s efficiency depends critically on the assumption that participants possess sufficiently large budgets to equalize marginal utilities across multiple public goods. When individual budgets are binding, this first-order condition may not be satisfiable, leading to systematically suboptimal allocations. We formalize this limitation via utility function analysis and a proof by contradiction. Empirically, we analyze 5.2 million donations from 23 Gitcoin Grants rounds and multiple Octant Epochs, revealing a marked shift in donor behavior over time. While early participants supported multiple projects with larger contributions, recent cohorts—driven by outreach and marketing strategies—tend to make small ($1–10) contributions to a single project. This concentration of support produces highly peaked preference distributions, challenging QF’s core assumption of distributed positive utilities across projects. The resulting funding patterns exhibit a power law in the number of projects supported and an exponential decay in contribution amounts, with the rate parameter λ serving as a concise descriptor of community giving behavior. These findings suggest that under realistic budget constraints and concentrated preferences, QF may fail to achieve its theoretical efficiency, and that evolving donor demographics can further exacerbate this divergence. We argue that mechanism design must adapt to these behavioral realities—incorporating donor education, information structure, and incentive alignment—to sustain QF’s effectiveness in public goods provision.

Simocracy: Generative Agentic Impact Evaluators

Decentralized funding ecosystems face a critical bottleneck: as funding scales grow, human evaluators struggle with burnout, inconsistency, and the cognitive demands of assessing diverse impact claims. We introduce Generative Agentic Impact Evaluators—LLM-based digital twins that simulate individual evaluators' reasoning and value judgments. Through structured interviews with five participants at the Impact Evaluator Research Retreat in Iceland, we captured each person's evaluation philosophy and constructed their digital counterparts. When tested on value-laden funding scenarios, these digital twins demonstrated remarkable variation in fidelity: the most successful achieved 95.0% agreement with their human counterpart, while overall system accuracy reached 76.7%—more than double random chance. Our findings reveal that generative models can meaningfully replicate evaluative reasoning when participants articulate consistent values, offering a path toward scalable, pluralistic evaluation systems that preserve individual perspectives while reducing human cognitive load. However, model performance depends critically on interview quality and participant consistency, highlighting both the promise and current limitations of AI-mediated governance. A working demonstration is available at simocracy.org.

Simocracy: Generative Agentic Impact Evaluators

Decentralized funding ecosystems face a critical bottleneck: as funding scales grow, human evaluators struggle with burnout, inconsistency, and the cognitive demands of assessing diverse impact claims. We introduce Generative Agentic Impact Evaluators—LLM-based digital twins that simulate individual evaluators' reasoning and value judgments. Through structured interviews with five participants at the Impact Evaluator Research Retreat in Iceland, we captured each person's evaluation philosophy and constructed their digital counterparts. When tested on value-laden funding scenarios, these digital twins demonstrated remarkable variation in fidelity: the most successful achieved 95.0% agreement with their human counterpart, while overall system accuracy reached 76.7%—more than double random chance. Our findings reveal that generative models can meaningfully replicate evaluative reasoning when participants articulate consistent values, offering a path toward scalable, pluralistic evaluation systems that preserve individual perspectives while reducing human cognitive load. However, model performance depends critically on interview quality and participant consistency, highlighting both the promise and current limitations of AI-mediated governance. A working demonstration is available at simocracy.org.

Quadratic funding under constrained budget is suboptimal

This paper examines the optimality of Quadratic Funding (QF) mechanisms under individual budget constraints through both theoretical analysis and large-scale empirical data. Building on Weyl et al. (2019), we show that QF’s efficiency depends critically on the assumption that participants possess sufficiently large budgets to equalize marginal utilities across multiple public goods. When individual budgets are binding, this first-order condition may not be satisfiable, leading to systematically suboptimal allocations. We formalize this limitation via utility function analysis and a proof by contradiction. Empirically, we analyze 5.2 million donations from 23 Gitcoin Grants rounds and multiple Octant Epochs, revealing a marked shift in donor behavior over time. While early participants supported multiple projects with larger contributions, recent cohorts—driven by outreach and marketing strategies—tend to make small ($1–10) contributions to a single project. This concentration of support produces highly peaked preference distributions, challenging QF’s core assumption of distributed positive utilities across projects. The resulting funding patterns exhibit a power law in the number of projects supported and an exponential decay in contribution amounts, with the rate parameter λ serving as a concise descriptor of community giving behavior. These findings suggest that under realistic budget constraints and concentrated preferences, QF may fail to achieve its theoretical efficiency, and that evolving donor demographics can further exacerbate this divergence. We argue that mechanism design must adapt to these behavioral realities—incorporating donor education, information structure, and incentive alignment—to sustain QF’s effectiveness in public goods provision.

Quadratic funding under constrained budget is suboptimal

This paper examines the optimality of Quadratic Funding (QF) mechanisms under individual budget constraints through both theoretical analysis and large-scale empirical data. Building on Weyl et al. (2019), we show that QF’s efficiency depends critically on the assumption that participants possess sufficiently large budgets to equalize marginal utilities across multiple public goods. When individual budgets are binding, this first-order condition may not be satisfiable, leading to systematically suboptimal allocations. We formalize this limitation via utility function analysis and a proof by contradiction. Empirically, we analyze 5.2 million donations from 23 Gitcoin Grants rounds and multiple Octant Epochs, revealing a marked shift in donor behavior over time. While early participants supported multiple projects with larger contributions, recent cohorts—driven by outreach and marketing strategies—tend to make small ($1–10) contributions to a single project. This concentration of support produces highly peaked preference distributions, challenging QF’s core assumption of distributed positive utilities across projects. The resulting funding patterns exhibit a power law in the number of projects supported and an exponential decay in contribution amounts, with the rate parameter λ serving as a concise descriptor of community giving behavior. These findings suggest that under realistic budget constraints and concentrated preferences, QF may fail to achieve its theoretical efficiency, and that evolving donor demographics can further exacerbate this divergence. We argue that mechanism design must adapt to these behavioral realities—incorporating donor education, information structure, and incentive alignment—to sustain QF’s effectiveness in public goods provision.

Quadratic funding under constrained budget is suboptimal

This paper examines the optimality of Quadratic Funding (QF) mechanisms under individual budget constraints through both theoretical analysis and large-scale empirical data. Building on Weyl et al. (2019), we show that QF’s efficiency depends critically on the assumption that participants possess sufficiently large budgets to equalize marginal utilities across multiple public goods. When individual budgets are binding, this first-order condition may not be satisfiable, leading to systematically suboptimal allocations. We formalize this limitation via utility function analysis and a proof by contradiction. Empirically, we analyze 5.2 million donations from 23 Gitcoin Grants rounds and multiple Octant Epochs, revealing a marked shift in donor behavior over time. While early participants supported multiple projects with larger contributions, recent cohorts—driven by outreach and marketing strategies—tend to make small ($1–10) contributions to a single project. This concentration of support produces highly peaked preference distributions, challenging QF’s core assumption of distributed positive utilities across projects. The resulting funding patterns exhibit a power law in the number of projects supported and an exponential decay in contribution amounts, with the rate parameter λ serving as a concise descriptor of community giving behavior. These findings suggest that under realistic budget constraints and concentrated preferences, QF may fail to achieve its theoretical efficiency, and that evolving donor demographics can further exacerbate this divergence. We argue that mechanism design must adapt to these behavioral realities—incorporating donor education, information structure, and incentive alignment—to sustain QF’s effectiveness in public goods provision.

Quadratic funding under constrained budget is suboptimal

This paper examines the optimality of Quadratic Funding (QF) mechanisms under individual budget constraints through both theoretical analysis and large-scale empirical data. Building on Weyl et al. (2019), we show that QF’s efficiency depends critically on the assumption that participants possess sufficiently large budgets to equalize marginal utilities across multiple public goods. When individual budgets are binding, this first-order condition may not be satisfiable, leading to systematically suboptimal allocations. We formalize this limitation via utility function analysis and a proof by contradiction. Empirically, we analyze 5.2 million donations from 23 Gitcoin Grants rounds and multiple Octant Epochs, revealing a marked shift in donor behavior over time. While early participants supported multiple projects with larger contributions, recent cohorts—driven by outreach and marketing strategies—tend to make small ($1–10) contributions to a single project. This concentration of support produces highly peaked preference distributions, challenging QF’s core assumption of distributed positive utilities across projects. The resulting funding patterns exhibit a power law in the number of projects supported and an exponential decay in contribution amounts, with the rate parameter λ serving as a concise descriptor of community giving behavior. These findings suggest that under realistic budget constraints and concentrated preferences, QF may fail to achieve its theoretical efficiency, and that evolving donor demographics can further exacerbate this divergence. We argue that mechanism design must adapt to these behavioral realities—incorporating donor education, information structure, and incentive alignment—to sustain QF’s effectiveness in public goods provision.

Quadratic funding under constrained budget is suboptimal

This paper examines the optimality of Quadratic Funding (QF) mechanisms under individual budget constraints through both theoretical analysis and large-scale empirical data. Building on Weyl et al. (2019), we show that QF’s efficiency depends critically on the assumption that participants possess sufficiently large budgets to equalize marginal utilities across multiple public goods. When individual budgets are binding, this first-order condition may not be satisfiable, leading to systematically suboptimal allocations. We formalize this limitation via utility function analysis and a proof by contradiction. Empirically, we analyze 5.2 million donations from 23 Gitcoin Grants rounds and multiple Octant Epochs, revealing a marked shift in donor behavior over time. While early participants supported multiple projects with larger contributions, recent cohorts—driven by outreach and marketing strategies—tend to make small ($1–10) contributions to a single project. This concentration of support produces highly peaked preference distributions, challenging QF’s core assumption of distributed positive utilities across projects. The resulting funding patterns exhibit a power law in the number of projects supported and an exponential decay in contribution amounts, with the rate parameter λ serving as a concise descriptor of community giving behavior. These findings suggest that under realistic budget constraints and concentrated preferences, QF may fail to achieve its theoretical efficiency, and that evolving donor demographics can further exacerbate this divergence. We argue that mechanism design must adapt to these behavioral realities—incorporating donor education, information structure, and incentive alignment—to sustain QF’s effectiveness in public goods provision.

Simocracy: Generative Agentic Impact Evaluators

Decentralized funding ecosystems face a critical bottleneck: as funding scales grow, human evaluators struggle with burnout, inconsistency, and the cognitive demands of assessing diverse impact claims. We introduce Generative Agentic Impact Evaluators—LLM-based digital twins that simulate individual evaluators' reasoning and value judgments. Through structured interviews with five participants at the Impact Evaluator Research Retreat in Iceland, we captured each person's evaluation philosophy and constructed their digital counterparts. When tested on value-laden funding scenarios, these digital twins demonstrated remarkable variation in fidelity: the most successful achieved 95.0% agreement with their human counterpart, while overall system accuracy reached 76.7%—more than double random chance. Our findings reveal that generative models can meaningfully replicate evaluative reasoning when participants articulate consistent values, offering a path toward scalable, pluralistic evaluation systems that preserve individual perspectives while reducing human cognitive load. However, model performance depends critically on interview quality and participant consistency, highlighting both the promise and current limitations of AI-mediated governance. A working demonstration is available at simocracy.org.

Simocracy: Generative Agentic Impact Evaluator

Decentralized funding ecosystems face a critical bottleneck: as funding scales grow, human evaluators struggle with burnout, inconsistency, and the cognitive demands of assessing diverse impact claims. We introduce Generative Agentic Impact Evaluators—LLM-based digital twins that simulate individual evaluators' reasoning and value judgments. Through structured interviews with five participants at the Impact Evaluator Research Retreat in Iceland, we captured each person's evaluation philosophy and constructed their digital counterparts. When tested on value-laden funding scenarios, these digital twins demonstrated remarkable variation in fidelity: the most successful achieved 95.0% agreement with their human counterpart, while overall system accuracy reached 76.7%—more than double random chance. Our findings reveal that generative models can meaningfully replicate evaluative reasoning when participants articulate consistent values, offering a path toward scalable, pluralistic evaluation systems that preserve individual perspectives while reducing human cognitive load. However, model performance depends critically on interview quality and participant consistency, highlighting both the promise and current limitations of AI-mediated governance. A working demonstration is available at simocracy.org.

Simocracy: Generative Agentic Impact Evaluators

Decentralized funding ecosystems face a critical bottleneck: as funding scales grow, human evaluators struggle with burnout, inconsistency, and the cognitive demands of assessing diverse impact claims. We introduce Generative Agentic Impact Evaluators—LLM-based digital twins that simulate individual evaluators' reasoning and value judgments. Through structured interviews with five participants at the Impact Evaluator Research Retreat in Iceland, we captured each person's evaluation philosophy and constructed their digital counterparts. When tested on value-laden funding scenarios, these digital twins demonstrated remarkable variation in fidelity: the most successful achieved 95.0% agreement with their human counterpart, while overall system accuracy reached 76.7%—more than double random chance. Our findings reveal that generative models can meaningfully replicate evaluative reasoning when participants articulate consistent values, offering a path toward scalable, pluralistic evaluation systems that preserve individual perspectives while reducing human cognitive load. However, model performance depends critically on interview quality and participant consistency, highlighting both the promise and current limitations of AI-mediated governance. A working demonstration is available at simocracy.org.

Simocracy: Generative Agentic Impact Evaluators

Decentralized funding ecosystems face a critical bottleneck: as funding scales grow, human evaluators struggle with burnout, inconsistency, and the cognitive demands of assessing diverse impact claims. We introduce Generative Agentic Impact Evaluators—LLM-based digital twins that simulate individual evaluators' reasoning and value judgments. Through structured interviews with five participants at the Impact Evaluator Research Retreat in Iceland, we captured each person's evaluation philosophy and constructed their digital counterparts. When tested on value-laden funding scenarios, these digital twins demonstrated remarkable variation in fidelity: the most successful achieved 95.0% agreement with their human counterpart, while overall system accuracy reached 76.7%—more than double random chance. Our findings reveal that generative models can meaningfully replicate evaluative reasoning when participants articulate consistent values, offering a path toward scalable, pluralistic evaluation systems that preserve individual perspectives while reducing human cognitive load. However, model performance depends critically on interview quality and participant consistency, highlighting both the promise and current limitations of AI-mediated governance. A working demonstration is available at simocracy.org.