How DAOs allocate resources: Lido, Aave, Sky and Uniswap, 2023–2025

This cp0x case study examines how four of the most transparent DAOs, Lido, Aave, Sky and Uniswap, allocated resources in 2023–2025 and what happened to protocol metrics after their decisions. It covers 112 decisions selected by pre-set materiality thresholds, including 58 spending and incentive decisions. Of the 58 spending and incentive decisions, 54 state a goal, but only 14 state it in terms of metrics published for all protocols. The pre-set rule for a cross-DAO pattern was met in only 2 of 20 readings, with opposite signs, so the study finds no robust pattern. Spending was 1.07 and 1.19 times DAO revenue at Lido in 2024 and 2025, fell from 0.92 to 0.44 of revenue at Aave between 2023 and 2025, and was about 0.5 at Sky, while the Uniswap DAO had almost no revenue. All results are descriptive associations on a small, non-random sample, with outcomes measured to 2026-10-05.

Key findings

• In this study, 54 of 58 major spending and incentive decisions taken by Lido, Aave, Sky and Uniswap in 2023–2025 stated a goal, but only 14 stated it in terms of public metrics such as fees, revenue, category share, total value locked or token price, and 40 named goals that no public source measures systematically.
• In this study, the pre-set rule for a cross-DAO pattern in 2023–2025 decisions at Lido, Aave, Sky and Uniswap was met in 2 of 20 readings, one negative (7 of 10 class A decisions) and one positive (3 of 4 DAOs for incentives with fixed category coverage), so no robust pattern was found.
• In this study, user incentives aimed at growth were the most common measurable decision type (5 decisions in 4 DAOs), and 12 months later the protocol's share of category fees had risen after Lido's Liquidity Observation Lab and Aave's Merit and fallen after Sky Token Rewards and two Uniswap incentive waves.
• In this study, Lido's spending was 1.07 times DAO revenue in 2024 and 1.19 times in 2025, or 0.88 and 1.04 excluding Token Reward Plan compensation paid in LDO, with revenue as defined by DefiLlama.
• In this study, Aave's spending fell from 0.92 to 0.54 to 0.44 of revenue in 2023, 2024 and 2025 as revenue grew, Sky spent about 0.5 of revenue, and the Uniswap DAO had almost no revenue until its fee switch at the end of December 2025.

Study at a glance

Question and motivation

A separate cp0x validation study found that DAO spending cannot be recovered from treasury outflows, so it must come from reports. This study examines four DAOs that report actual spending, representing four governance models. Lido works through working groups and committees, Aave through contracted service providers, Sky through an operating structure funded by the DAO, and Uniswap through a split between the DAO and the Uniswap Foundation (UF).

The study traces each decision to its spending and a measurable outcome, then asks whether outcomes repeat across DAOs. It is descriptive and makes no causal claims.

Data

A decision enters the study if it was executed in 2023–2025 and is an approved spending or budget decision of at least 10% of the materiality denominator, an incentive decision of at least 10%, a change in annual spending on a line item of more than 25% and at least 5% of the denominator, or any value distribution decision. The denominator is spending by the DAO and the foundations and committees it funds in the previous full year, or the current approved budget where no such year exists (Lido 2024, Uniswap 2023, both flagged).

Class A is investment in the protocol, from contributors and grants to incentives and marketing. Class B is buybacks, burns, holder payouts and fee switches. Class C (treasury management) is described but not evaluated. A transfer to a foundation is funding, and spending is what the foundation spent. Token-only amounts were not converted to USD.

Method

1. Freeze the method with a sha256 hash before data collection.
2. Executors without access to outcome series build a decision table and case card per DAO and log incidental contact with outcome figures.
3. A separate executor, seeing only the decision tables, maps each stated goal to a category and metric.
4. Freeze decision tables, case cards and goal map with sha256 hashes.
5. Join outcomes by script from the execution date (t0). Usage goals use the change in the protocol's share of category fees, monetisation goals the change in take rate (revenue divided by fees), capital goals the change in total value locked (TVL), and token price goals the token's return against ETH. Fees and revenue are 90-day sums ending on the date, TVL is the value on the date, and price is the 7-day median.
6. Use 12 months as the main window. A window counts only if it ended by the cutoff of 2026-10-05, and unfinished windows are marked "not yet observable".
7. Flag a pattern candidate if decisions of one type with the same goal occur in at least 3 DAOs and at least 2/3 of observed cases, and at least 3, share one 12-month sign. The sign rule has no dead zone.
8. After review, test the rule under all 20 readings of type and counting unit, with the frozen share measure and with a fixed-coverage category (change log entry D7).

Results

Decisions and goals

Class A decisions number 11 at Lido, 12 at Aave, 16 at Sky and 19 at Uniswap. Sky supplies 45 of 112 rows, 26 of them changes to Smart Burn Engine parameters (buyback rate and split), so per-decision shares lean towards Sky.

This table shows the goal categories in the frozen goal map.

Of 58 class A decisions, 54 state a goal and 14 map to a public metric. Line-item rows are year-on-year aggregates with no decision text, so they are counted separately (D8), and most goals without a metric state a purpose with no metric at all.

Outcomes of the 17 measurable decisions

This table shows the 12-month log change in each primary metric, and for category share also the fixed-coverage version. Percent equals e^x − 1, a restatement added for this page.

Results are mixed on a very small sample. Merit and Ahab aimed at DAO revenue, which rose 44.7% in the 12 months after the 2024 extension and fell 12.1% after the 2025 continuation. DefiLlama added decentralised exchange (DEX) adapters without history during these windows (Orca, Meteora DLMM), and fixed coverage flips two signs and halves the Uniswap falls, though it also drops genuine new entrants.

Pattern rule under 20 readings

This table shows the 8 readings with at least 3 DAOs. The other 12 cover types found in at most 2 DAOs.

A result that changes sign with the counting method is not a pattern. In the negative reading, Uniswap supplies 4 of the 7 negatives and Lido's −0.0008 is effectively zero.

Scale of spending

This table shows spending by each DAO and the bodies it funds against DefiLlama calendar-year fees and revenue, in USD millions. Revenue follows DefiLlama, which for Lido is the treasury's share after node operators.

Lido's ratios fall to 0.88 and 1.04 without Token Reward Plan (TRP) compensation in LDO ($9.7M in 2024, $6.4M in 2025), and its 2023 spending cannot be determined. Aave's 2022 and 2023 figures come from different reporters, 2022 includes a one-off $16.3M V3 retroactive payment, and Safety Module and Umbrella rewards are excluded. Sky's 2025 figure excludes $5.8M spent by the Sky Frontier Foundation itself, and Sky's liquid treasury fell to $0.3M because it was moved to that foundation. Runway (liquid treasury divided by spending) is therefore shown only for Lido and Aave. Uniswap Labs spending is not published.

Robustness and replication

Limitations

• Four cases, chosen as the most transparent DAOs, not a random sample, and all decisions fall in 2023–2025.
• Several rows belong to one programme or one vote (two Uniswap incentive waves, three Merit rows), yet per-decision counts treat them as independent.
• Spending reports use different methods, and some years have only budgets or partial periods.
• Token-only decisions were included provisionally when a reported aggregate cost passed the threshold and excluded otherwise.
• Goal mapping is a judgement call in several cases, share of category TVL could not be computed, and category fee share depends on when DefiLlama began tracking competitors.
• Outcomes were known to the authors in advance, so the main protection is the freeze before the join. The pattern rule did not define decision type or counting unit in advance.
• 16 Sky decisions before the SKY price series (late September 2024) have no token return against ETH.
• Rocket Pool, Maple, SSV, Nexus Mutual and Yearn were not analysed.

What this means for DAOs

These DAOs explain why they spend but rarely tie the goal to a public metric, so their disclosures cannot show that one type of spending works better than another. This does not mean incentives fail.

The recommendation goes beyond the study but follows from it. Every material budget or incentive programme should state, before the vote, one primary public metric, its baseline value, the expected direction or target, the review date, and who will publish the review. Without this, a year later nobody can tell a good expense from a bad one. This is a governance observation, not investment advice.

Frequently asked questions

Do DAOs state goals for their spending?

DAOs usually state a goal, and in this study, 54 of 58 major spending and incentive decisions at Lido, Aave, Sky and Uniswap in 2023–2025 stated a goal. Only 14 of them map to metrics published for all protocols.

Do token incentives increase market share?

The data do not answer this. After 5 growth incentive decisions in 4 DAOs, category fee share rose within 12 months for Lido and Aave and fell for Sky and two Uniswap waves. With fixed category coverage 3 of 4 DAOs show a rise, so even the sign depends on measurement.

Is there a pattern across DAOs?

No robust one. The pre-set rule was met in 2 of 20 readings, one negative (7 of 10 class A decisions) and one positive (3 of 4 DAOs for incentives with fixed coverage).

How much does Lido spend relative to revenue?

Lido's spending was $55.2M in 2024 and $49.6M in 2025, which is 1.07 and 1.19 times DAO revenue as DefiLlama defines it. Excluding Token Reward Plan compensation paid in LDO, the ratios are 0.88 and 1.04.

Why might Uniswap's market share falls be overstated?

DefiLlama added DEX adapters without history during the windows, such as Orca from November 2024 and Meteora DLMM from January 2025. Holding the category to products tracked 90 days before the decision shrinks the fall after the second incentive wave from 56.9% to 23.9%, and after the AlphaGrowth programme the share rises instead.

Was the study pre-registered?

The study was only partly pre-registered. The method was frozen with sha256 b1143c78…b551a on 2026-10-07 before data collection, and decision tables and the goal map were frozen before outcomes were joined. The pattern rule left decision type and counting unit undefined, so it was checked under all 20 readings after results.

Data and code

All data, code, the frozen method and the review for this study are on GitHub.

To reproduce, run code/join_outcomes.py, then code/sensitivity.py and code/make_scale.py, as described in code/README.md.

Version history