Zeat · Research note · NEAR HACKATHON 2026 · Investment Thesis track
Is NEAR Intents' revenue defensible?
Three pre-registered causal tests of NEAR Intents' moat, each against a control built from competitors, checked against 3,346 NEAR blocks read directly from the chain.
Abstract
Most NEAR theses measure how much NEAR Intents earns. An investor's next question is whether it can keep earning it. This note tests that three ways, each against a synthetic control built from competing venues, so a market-wide move is never mistaken for a NEAR-specific one. The questions, data rules and pass/fail thresholds were committed to git before any result was computed.
Table 1Results under the pre-registered rules · Weak moat evidence, 1 of 3 passed
| Test | Question | Estimate | Placebo p | Verdict |
|---|---|---|---|---|
| T1 Pricing power PASS | When NEAR started taking a cut on Feb 23, 2026, did volume leave? | +16%volume vs control, 8 weeks | 0.00* | PASS |
| T2 Trust shock FAIL | After the Oct 1, 2026 exploit, did volume come back? | -21%volume vs control, weeks 4–5 | 0.05 | FAIL |
| T3 Price premium FAIL | Does Intents charge more than rivals and still hold share? | 14.0 vs 18.4bps, Intents vs rivals; dearer in 4% of weeks | n/a | FAIL |
T1 and T2 fail only on a loss worse than −10% with placebo p < 0.10. T3 needs Intents to be dearer than rivals in at least 75% of weeks with a share that is not falling. T2: Not robust to the choice of controls.
* T1's p-value is not informative: 100% of its placebo windows fall in Intents' 2025 hypergrowth and show larger gains than the real effect (see Section 2).
Key findings for an investor
- The fee switch did not cost volume. NEAR raised its share of fees and volume kept pace with competitors. Taking a cut has not driven integrators away, so far.
- Intents is the cheaper venue, not the premium one. Its users pay less than on native cross-chain rivals. That is a growth strategy, not evidence of a moat: a cheaper rival could compete the same way.
- Trust is the open risk. Volume trailed its control by weeks 4–5 after the exploit. The signal passes the pre-registered test but is fragile under robustness checks; it is the number to watch.
1Question and design
Revenue is only worth a high multiple if it is defensible. NEAR Intents now earns real fees, so the question is whether it has a moat: can it raise its take, survive a loss of trust, and charge more than rivals, without losing volume?
Figure 1NEAR Intents weekly volume, log scale
Intents grew from almost nothing to a large venue in 2025; growth flattened in 2026. Both events tested here sit on the mature part of the curve.
Source: DefiLlama near-intents daily volume, summed to Monday–Sunday weeks; weeks with
fewer than 4 reported days are left blank.
Each test compares Intents with a synthetic control: a weighted mix of competitors (THORChain, Chainflip, Maya, SOCKET, 0x and the all-DEX market) fitted to Intents' log weekly volume over the 26 weeks before an event. The effect is the gap that opens after it. To judge whether a gap is unusual, the same estimator is re-run at every earlier fake event date (a placebo test). LI.FI, Rango and Bitget route part of their flow through Intents, so they are kept out of the control and used only as a sensitivity check. Full rules: METHODS.md.
2Pricing power: the fee switch PASS
On Feb 23, 2026 NEAR Intents switched on a protocol cut of the fees integrators charge. If integrators or their users had cheaper alternatives, volume should have fallen relative to competitors.
Figure 2Share of fees NEAR keeps, weekly
The switch was real. NEAR's share of fees rose from 7.8% in the 26 weeks before to 13.8% in the 8 weeks after, while the fee rate users paid stayed level (18.0 bps before, 17.6 bps after).
Source: DefiLlama near-intents daily revenue ÷ daily fees, weekly.
Because the price users paid did not move, T1 measures whether integrators and their flow left, not users' price sensitivity.
Figure 3Fee switch: NEAR Intents vs its synthetic control
In the 8 weeks after the switch, Intents ran +16% relative to its synthetic control. Verdict: PASS
Control weights, fitted on the 26 weeks before the event: Maya Protocol 63%, SOCKET 37%. Pre-period fit error (RMSPE) is 0.57 log points, which is loose: Intents' 2025 hypergrowth is hard for any mix of mature competitors to match.
Read this with care. 100% of the 29 placebo windows show a larger gain than the real one (median +430%), because they sit in 2025, when Intents was outgrowing everything. So the fee switch coincided with Intents' growth lead shrinking. The T2 placebo series (Section 3) shows that lead had already fallen towards zero by January 2026, weeks before the switch, so the slowdown is not evidence that the switch caused it.
Placebo tests for T1 Figures 4–5
Figure 4Gap to control, real vs 29 placebo windows
Grey: the same estimator run at fake event dates before the switch. Blue: the real gap.
Figure 5T1 placebo effects by fake event date
Each dot is the effect estimated at a fake event date; the line is the real effect, +16%.
Robustness checks not used for the verdict
| Variant | Effect |
|---|---|
| Leave one control out (range) | +16% to +31% |
| Difference-in-differences | +60% |
| 13-week pre-window | +25% |
| Adding LI.FI, Rango, Bitget (contaminated) | +18% |
| without THORChain | +16% |
| without Chainflip | +16% |
| without Maya Protocol | +24% |
| without SOCKET | +31% |
| without All DEX volume | +16% |
3Trust shock: the exploit FAIL
On Oct 1, 2026 NEAR Intents was exploited for about $3.8M through a deposit-notification bug. The contract was paused within about 30 minutes, deposits resumed on Oct 2, and the funds were returned. A venue with loyal users should see volume come back within weeks.
Figure 6Exploit: NEAR Intents vs its synthetic control
No immediate collapse (weeks 0–1: -1%), then Intents trailed its control by -21% in weeks 4–5. Week by week: wk 0: -25%, wk 1: +29%, wk 2: -14%, wk 3: -3%, wk 4: -24%, wk 5: -18%.
Control fitted on 26 pre-event weeks (Chainflip 1%, 0x Bridge Aggregator 18%, All DEX volume 81%); pre-period RMSPE 0.13. Recovery statistic: mean gap in weeks 4–5.
Figure 7How unusual is -21%? Placebo effects at 2026 fake event dates
Against all 63 placebo windows, p = 0.05. Using only the 34 windows dated 2026, after the hypergrowth phase, p = 0.09. Both are below 0.10, so the 2026-only check agrees with the pre-registered one. Verdict: FAIL
Each dot re-runs the T2 estimator at a fake event date; the line is the real effect.
Why this is fragile. Dropping a single control moves the estimate between -63% and +12%. Adding the contaminated aggregators gives +27%. In the placebo-in-space test, Intents' post-event deviation ranks 5 of 5: less unusual than its own controls'. And the shape is not a typical trust shock: there was no drop in the first two weeks. Treat this as a warning signal to re-test, not a settled loss.
All placebo windows for T2 Figures 8–9
Figure 8Gap to control, real vs 63 placebo windows
Figure 9T2 placebo effects by fake event date, all windows
In 2025 Intents outgrew every control, so those placebos sit far above zero; that is why Figure 7 uses 2026 windows only.
Robustness checks not used for the verdict
| Variant | Effect |
|---|---|
| Leave one control out (range) | -63% to +12% |
| Difference-in-differences | -49% |
| 13-week pre-window | -32% |
| Adding LI.FI, Rango, Bitget (contaminated) | +27% |
| without Chainflip | -20% |
| without SOCKET | -21% |
| without 0x Bridge Aggregator | +12% |
| without All DEX volume | -63% |
4Price premium FAIL
A business with a moat can charge more than rivals and keep its customers. This test compares what users pay per dollar swapped on Intents with what they pay on native cross-chain rivals.
Figure 10Fee rate paid by users, basis points
Intents' median fee rate was 14.0 bps against 18.4 bps for competitors; Intents was dearer in only 4% of 26 weeks.
Source: DefiLlama daily fees ÷ daily volume, weekly, per venue. The competitor rate is volume-weighted across THORChain, Chainflip and Maya for the venues reporting that week (grey lines).
Figure 11NEAR Intents' volume share against Chainflip
Share went from 74% to 83%; trend -0.08% a week (90% CI -0.45% to +0.34%), so share is holding. Against THORChain and Chainflip over the 20 weeks both report, the trend is -1.94% a week. That broader comparison points to share slipping against the larger rival, so “holding” applies to Chainflip only.
Share uses only competitors that report every week of the window, because a denominator that gains and loses venues would move on its own.
T3 needs both a premium in at least 75% of weeks and a share that is not falling. Verdict: FAIL
5On-chain check: reading NEAR blocks directly
Every test above uses DefiLlama, an aggregator. To check it against the chain itself, the build reads a
deterministic sample of NEAR blocks and counts what intents.near executed in each one: signed
intents (intents_executed events), swaps (token_diff events) and receipts. A weekly
total is the mean per sampled block times the number of blocks in that week.
5.1 Does the aggregator match the chain?
Figure 12On-chain executed intents vs DefiLlama volume, indexed
DefiLlama's volume loosely tracks on-chain activity. Across 32 weeks the correlation is 0.67 in levels and 0.42 in week-to-week changes; the second is the stricter test.
Both series indexed to their mean over the first 8 weeks, log scale. Volume is dollars and on-chain counts are actions, so they need not move one-for-one.
5.2 The two events, on-chain
Fee switch. Executed intents in the 8 weeks after Feb 23 were -3% against the 8 weeks before; DefiLlama volume was -17%. In this window the week-to-week correlation of the two is 0.22.
Exploit. On-chain activity moved the same way, far more strongly: executed intents in weeks 4–5 after Oct 1 were +88% against the 8 weeks before, while DefiLlama volume was +7% over the same weeks.
These are before/after comparisons with no competitor control, because competitors' activity lives on other chains, so they are not comparable with T2's synthetic-control gap: both raw measures can rise while Intents still loses ground to rivals. Counts of signed intents are not dollars, and one strong week (Nov 5–11) drives most of the on-chain rise.
5.3 The Oct 1 pause
Figure 13intents.near receipts, one block sampled every 30 minutes, Sep 30 to Oct 2
A typical block before Oct 1 carried 1.9 intents receipts. The longest run of sampled blocks with none was 5 (about 2.5 hours), from 2026-10-01 08:48 UTC to 2026-10-01 11:00 UTC. Before Oct 1, 28% of sampled blocks already had none, so a run this long would appear somewhere in 144 samples 15% of the time by chance. The sample is consistent with a pause of a few hours but cannot pin one down.
144 blocks; successful receipts only.
5.4 Who fills the swaps?
Each swap settles as a token_diff between a user and a solver. Users rarely appear in more
than one sampled block; solvers do. Over the last 8 weeks, 13 accounts appeared in swaps in at least
3 different sampled blocks.
Table 2Repeat counterparties in swaps, last 8 weeks
| Account | Share of appearances |
|---|---|
solver-priv-liq.near | 35.5% |
blockzen-1-main.near | 17.0% |
solver-multichain-asset.near | 7.5% |
crux-solver.near | 7.0% |
kipseli.near | 6.0% |
0x8c60fa34790d4661ae4b5302f2a6… | 5.0% |
0x94845333028b1204fbe14e1278fd… | 4.5% |
brlabs-solver.near | 4.5% |
The top account carries 36% and the top 3 carry 60% (HHI 0.18): moderately concentrated. A concentrated solver set is a single point of failure for price and uptime, which matters for the trust result in T2.
Descriptive. An account that routes on behalf of many users can look like a solver, so this measures repeat counterparties, not confirmed solvers.
6The long-term case for NEAR
NEAR holders are diluted by new issuance; Intents revenue is what can offset it. Revenue is volume × fee rate × NEAR's cut, and the three tests say which of those levers NEAR can pull.
Table 3Intents revenue against the value of NEAR issuance
| NEAR issued in the last year, from block headers | 31.7M NEAR (2.48%) |
| Value of that issuance at the 30-day average price ($4.45) | $141.3M a year |
| Intents protocol revenue, trailing 12 weeks annualised | $21.6M a year |
| Revenue as a share of issuance value | 15% |
| NEAR's cut of Intents fees, trailing 12 weeks | 27% |
| Cut needed to cover issuance at today's volume and fee rate | 177% |
| Volume growth needed at today's cut and fee rate | 6.5× |
Issuance is the exact change in total_supply between two block headers a
year apart (net of burns). Price is CoinGecko's 30-day average before the snapshot. Revenue is DefiLlama's.
Covering issuance is a yardstick for value accrual, not a claim about how buybacks are executed.
Table 4Which revenue lever is safe to pull
| Lever | What the moat tests say |
|---|---|
| NEAR's cut27% now | T1: raising it did not cost volume. This is the proven lever, but it cannot cover issuance on its own, even at 100% capture. |
| Fee rate15.8 bps | T3: Intents is already cheaper than rivals and has no premium to spend; raising it is untested. |
| Volume | T2: volume is the lever exposed to trust. Growth of 6.5× has to come from users who stay through incidents. |
Figure 14What it takes: a scenario calculator
Starts from today's values in Table 3. Move the levers to see how much of NEAR's issuance Intents revenue would cover.
$21.6M a year of revenue against $141.3M a year of issuance.
Revenue = annual volume × fee rate × NEAR's cut; issuance value = NEAR minted per year (31.8M NEAR, from block headers) × price. A scenario tool, not a result or a forecast: it treats the levers as independent, though a higher fee rate could cost volume.
7Limits
- All volume and fee data comes from DefiLlama, an aggregator. Its history can be revised; the snapshot hashes in Section 8 pin the exact version analysed.
- The control pool is small (6 clean series), so placebo-in-space p-values cannot go below about 1/6.
- The on-chain figures come from a sample of blocks, not a full census, so each weekly estimate carries
sampling error (bootstrap intervals are in
results.json). The sample is pinned indata/onchain/and can be checked against the chain withpython3 -m zeat verify-onchain. - Six weeks of post-exploit data is the most the snapshot allows. T2 should be re-run as data arrives;
python3 -m zeat refreshthenbuilddoes exactly that, and a daily job republishes this page on fresh data.
8Reproduce
Python 3.10+, standard library only, no API keys. The build is offline and byte-deterministic: it must
reproduce this page, results.json and THESIS.md exactly.
git clone https://github.com/Dydex/zeat && cd zeat
python3 -m zeat verify # data files match their sha256 hashes
python3 -m zeat build # rebuilds every number, this page and THESIS.md (~40 s)
git status --short # empty: the rebuild matches the committed files
python3 -m unittest # known-answer and byte-exact rebuild testsData sources and hashes snapshot fetched 2026-11-12T22:52:48Z
| Series | Role | Source | sha256 |
|---|---|---|---|
| NEAR Intents | treated | intents_volume | 2fc4816f19c4 |
| NEAR Intents | fees | intents_fees | a44b4af41cc8 |
| NEAR Intents | revenue | intents_revenue | f924b3d7c9b0 |
| THORChain | donor | thorchain_volume | a67f24e75217 |
| THORChain | fees | thorchain_fees | 7dbd56e56167 |
| Chainflip | donor | chainflip_volume | dae294068da1 |
| Chainflip | fees | chainflip_fees | 4b9273f78088 |
| Maya Protocol | donor | maya_volume | ec37981613b1 |
| Maya Protocol | fees | maya_fees | 901486292806 |
| SOCKET | donor | socket_volume | 5172a339cb6c |
| 0x Bridge Aggregator | donor | zerox_volume | 8e9094885f95 |
| All DEX volume | donor | alldex_volume | 68e8632acc30 |
| LI.FI | contaminated | lifi_volume | 966ac3fbdfd5 |
| Rango | contaminated | rango_volume | d0bb77b1db7a |
| Bitget Wallet X | contaminated | bitget_volume | b4e72e899170 |
| NEAR price (CoinGecko) | price | near_price | 353165b52511 |