DAO Simulator

What Results Showed

• 15-Point Cliff, Not a Slope

  At 5% quorum, 99.9% of proposals pass. At 10%, 82%. At 20%, 25%. At 40%, zero. The drop is catastrophic, not gradual.

• 2Voters Agree — They Just Don’t Show Up

  Pass rate among proposals that reached quorum stayed 97–98%. The bottleneck is turnout, not disagreement.

• 3Bigger DAOs, Lower Turnout

  Scaling from 50 to 500 members dropped participation from 26% to 22%, while pass rate actually rose. Size dilutes engagement.

What Results Showed

• 1Fairness Improved Speed

  Quadratic voting cut whale influence from 0.449 to 0.256 (43% reduction) while pass rate rose from 92.7% to 98.5%. The expected tradeoff didn’t exist.

• 2Power Shape > Activity Throttles

  Velocity penalties alone had only small effects. Changing how power distributes mattered far more than limiting how often whales can act.

• 342% Capture Risk Drop

  Capture risk fell from 0.464 to 0.269 under the strongest settings, using a layered stack of quadratic voting + delegation caps + velocity controls.

What Results Showed

• 1No Tradeoff Found

  Raising temp-check pressure from 5% to 50% lifted pass rate from 96.4% to 98.5% while maintaining full throughput. Better filtering meant better outcomes, not slower ones.

• 212-Day Floor Held Quorum

  Fast-tracking with a 12-day minimum voting window kept quorum reach above 99%. Speed didn’t come at the cost of participation.

• 3Zero Proposals Abandoned

  47–50 proposals per cycle completed without a single abandonment, though very short expiry windows remain risky for complex work.

What Results Showed

• 1Negative Growth Is the Norm, Not a Bug

  Average growth stayed at –0.71 across all 12 configurations — matching the broader DAO industry trend where few treasuries have grown in real terms. Stabilization cut volatility from 0.45–0.50 to 0.24–0.27, buying time rather than generating returns.

• 2Buffers Beat Active Management

  Reserve buffers with spending caps ($10K–$13K final treasury) outperformed configurations without them, providing downside protection without complex rebalancing.

• 3No Set-and-Forget

  The negative growth floor reflects real market conditions. Treasury policy can’t create growth from nothing — it manages decline. Parameters need quarterly review as market conditions shift.

What Results Showed

• 1Structure > Intent

  Specialized partnerships generated 50% more inter-DAO proposals (75.8 vs 50.3) and 8% higher ecosystem treasury ($26,107 vs $24,071) than generic connections.

• 221% Success Is the Ceiling (So Far)

  Inter-DAO proposal success reached 21–23% with the best topology, versus 0% in isolation. Cooperation is fragile and needs explicit design.

• 3Moderate Alignment Is Normal

  Cross-DAO alignment stabilized at 0.53–0.56, not near 1.0. Expecting high alignment between autonomous organizations is unrealistic — design for disagreement.

What Results Showed

• 1Thinking Mode Is the Game Changer

  All-LLM+thinking achieved 79.6% pass rate — +6.7pt above the 72.9% rule-based baseline. Without thinking, all-LLM dropped to 57.7% (-15.2pt). Same model, same prompts, opposite outcomes.

• 2Information Parity Matters

  Enriched prompts (~2-3K tokens) give LLMs the same context rule-based agents use: treasury health, competing proposals, forum sentiment, vote history. The old 350-token prompts were asking the model to vote blind.

• 3Prior Finding Was Prompt Quality

  Experiment 12 (300 runs, 3 small models) found LLMs degrade governance. Experiment 17 (Gemma 4 E4B) shows this was caused by impoverished prompts and missing thinking mode, not a fundamental limitation.

What Results Showed

• 1Conviction Voting: 0% Across All 14 DAOs

  Not a single proposal passed under conviction voting for any of the 14 calibrated DAOs. The continuous-signal mechanism is fundamentally incompatible with discrete proposal governance.

• 2Quadratic Voting Consistently Best

  Across 14 DAOs, quadratic voting produced higher pass rates than majority for every DAO tested. Especially strong for high-concentration DAOs like Aave (+4.1pt) and Arbitrum (+2.4pt).

• 3Supermajority Paradox

  Supermajority yielded 100% pass rate for 6 DAOs (Aave, Compound, Arbitrum, Lido, Maker, Nouns). It acts as a tighter quality filter — the few proposals that clear the higher threshold are universally supported.

What Results Showed

• 160% Single Entity Control Reduction

  At 50 members, one entity can control 2.8% of votes. At 500, just 1.1%. Scale dilutes concentrated power more effectively than any voting mechanism.

• 2Pass Rates Scale With Size

  50 members: 92.8% pass rate. 500 members: 99.7%. Larger voter pools produce more decisive outcomes.

• 3Participation Holds Steady

  Voter retention stays at 91-93% across all scales. Bigger DAOs don't suffer engagement death.

What Results Showed

• 1All Four Mechanisms Identical

  Pass rates within 0.7pt, whale influence within 0.4pt, capture risk within 0.4pt. No statistically significant difference across 1,100 runs.

• 2Conviction Voting Breaks Everything

  0% pass rate across all 14 calibrated DAOs. Its continuous-signal design is fundamentally incompatible with discrete proposal governance.

• 3Quadratic Thresholds Are the Only Lever

  The only mechanism change that measurably improves governance is applying a quadratic voting threshold — and that's a parameter tweak, not a mechanism swap.

What Results Showed

• 1Conviction Stays Highest Under Stress

  At 5 black swans, conviction retains 81.3% pass rate vs majority 74.7% and supermajority 68.2%. Its gradual signal accumulation absorbs shocks.

• 2Supermajority Collapses Fastest

  Supermajority drops 12.8pt at highest shock frequency vs 8.0pt for majority. Higher thresholds amplify disruption.

• 3Growth Takes the Biggest Hit

  Black swans reduce member growth by 54% (0.333→0.153). The real damage is to recruitment, not governance quality directly.

What Results Showed

• 1+8.1pt From Basic Q-Learning

  Pass rate jumps from 87.5% to 95.6% with simple tabular Q-learning. Agents learn cooperative voting patterns within episodes.

• 2Tiers 2-4 Add Only +0.2pt

  Experience replay, shared experience, and DQN combined add just 0.2pt to pass rate. The complexity isn't worth it for governance quality alone.

• 3DQN Reduces Inequality

  Token Gini drops from 0.523 to 0.467 with the full DQN stack — a 5.6pt reduction. RL agents learn to redistribute, even if governance metrics plateau.