NEAR Intents: real demand, shrinking visibility

An on-chain audit of the contract behind NEAR's headline volume: 44,994 transactions, three independent samples, six tests set in advance.

1. The thesis and how it was tested

NEAR's investment case leans on one number: the volume of NEAR Intents, published from NEAR's own data tables. We asked whether an outsider can confirm that number and what stands behind it. Six claims were written down with pass rules before the data was analysed. Each was then tested on a seeded random sample of the contract's transactions, repeated on a second sample, and checked out-of-sample on a third drawn after the rules were fixed.

T1–T5 show the pooled estimate of the first two samples with 95% intervals. A claim is supported or contradicted only when the whole interval is on one side of the rule and both samples' own estimates agree; otherwise it is inconclusive. T6 uses published figures.

2. Demand: real, dollar-denominated, and somewhat smaller than published

Value entering the contract in our sample is 0.79× (95% CI 0.57× to 1.06×) of the volume NEAR publishes for the same 28 days. The headline number is not an artefact of a dashboard: money of that order shows up on-chain. A second check agrees: the sample's transaction count is 0.97× the exact count. The independent replication sample measures 0.66× (95% CI 0.49× to 0.84×); pooling both samples gives 0.70× (0.56× to 0.84×). The verdict holds.

Stablecoins are 72.8% (95% CI 63.1% to 80.4%) of the value moved inside the ledger. This is a dollar-transfer product more than a trading venue. The independent replication sample measures 72.2% (95% CI 63.6% to 80.4%); pooling both samples gives 72.5% (66.4% to 78.5%). The verdict holds.

Exact counts for 180 days. The contract handled about 56,619 transactions a day in the first 30 days of this window (2026-05-18 onwards) and 113,293 a day in the last 30: 2.0× as many. Published volume over the same two spans went from $85M to $127M a day (1.5×). The busiest day was 2026-11-05 with 279,561 transactions.

What cuts against it: every one of the three samples puts deposits below the published figure, and the third sample alone is inconclusive. We can confirm roughly two thirds of the headline, not all of it. Whether one chain dominates deposits is not settled: the samples disagree.

3. The main finding: swaps moved out of public view

Up to 2026-10-30, 21.8% (95% CI 19.7% to 23.8%) of transactions carried a swap settled in the open. From 2026-10-31 it is 3.5% (95% CI 2.8% to 4.3%). Open swaps fell from about $38M to $16M a day while shielded value rose from $54M to $113M a day. The share has been under 3% on every one of the last 10 full days (since 2026-11-04), with open swaps averaging $2M a day. Re-drawing the sample 1,000 times puts the break between 2026-10-30 and 2026-11-03 in 95% of draws.

Published volume did not fall with it: $126M a day before, $132M after. Swaps we can see in the open were 30% of the published figure before the break and 12% after. The rest of the published number now has to be taken on trust.

But read it with this: that account is on one side or the other of 86.7% (95% CI 78.2% to 92.6%) of the value, and it is not an anonymous whale. NEAR's own documentation names it as the Confidential Intents Treasury: a transfer into it is the "shield" step that moves funds onto the confidential ledger, and a transfer out of it is the unshield. Every one of the 5,964 sampled transfers into it carries a long opaque memo (120 to 184 characters), and none of the 14,566 other transfers does. That matches the documentation, which says the shield memo carries the encrypted recipient. We stored only the memo's length; the one we re-fetched in full is base64 data.

For the thesis. Privacy is a product most cross-chain venues do not sell, and the flow adopted it within days. 2,770 different accounts shielded funds in the sampled windows. Two solver accounts, solver-priv-liq.near and solver-multichain-asset.near, supply 36% of the shielded value between them; the ten largest shielders together account for 61.2% (95% CI 44.7% to 72.8%). Accounts that each contribute under 1% make up 34%. So use is broad by count, with solver liquidity as the largest single block.

Against it. The swap itself executes on a separate private contract. From outside we can count what crosses the boundary, not who traded what or at what price. NEAR's take was 4.9 bps of published volume before the break and 4.0 bps after; what users paid went from 15.5 bps to 15.1 bps. The timing matches; we have no evidence of cause.

4. Structure: who may submit, and what NEAR keeps

Five accounts submit 97.5% (95% CI 97.2% to 97.7%) of every transaction that touches the contract. End users almost never sign a NEAR transaction themselves (1.6% of transactions are signed by an unnamed user account). The independent replication sample measures 97.8% (95% CI 97.4% to 98.1%); pooling both samples gives 97.6% (97.4% to 97.8%). The verdict holds.

NEAR's captured revenue is 4.9 bps of volume over the last 30 days: $1.8M on $3.8B. Users pay 16.0 bps; NEAR keeps 30.5% of that.

DefiLlama has both volume and revenue for 4 of these 9 venues. Among them NEAR Intents keeps 4.7 bps, ranking 2 of 4. The venues of comparable size keep less: Unit 2.4 bps and Maya Protocol 0.9 bps. By revenue in dollars it is first of the 9. This cuts against our own T6 objection: the 5 bps bar was ours, and the comparable venues we can measure do not clear it either.

5. Valuation: what the price asks of Intents

NEAR's market value is $4.09B, or 182× annualised Intents revenue of $22.4M. This is arithmetic on Intents revenue alone. The token is also valued for gas, staking and everything else built on NEAR, so these multiples overstate what Intents must deliver by itself. They are still the cleanest way to see how far revenue is from carrying the price.

The scenarios are illustrations with no probabilities attached. The point they make is structural: at today's take, even four times the volume leaves the multiple at 46×. The valuation case needs the take to rise, and the take is a policy choice that depends on who controls access.

6. Risks

7. What we will watch, and when we would change our mind

Five predictions were registered on 2026-11-14 and will be graded on 2026-12-14. A daily job re-samples the last 24 hours and scores them without anyone editing the result.

• Deposits fall wholly outside 0.5×–2× of published volume on a re-sample: the demand case is not verifiable and the thesis falls.
• NEAR publishes auditable aggregates for the confidential ledger: the visibility objection closes and the confidential share becomes a plain positive.
• The top five signers drop under 90% of transactions, or the take holds above 5 bps for three months without volume falling: the structural objections weaken.

8. Questions for the NEAR team

1. Who controls the keys of the Confidential Intents Treasury, and what stops a single party moving its funds?
2. Will the total shielded supply be published, so holdings can be compared with claims?
3. How is the published volume figure built, and does it include swaps executed on the confidential ledger?
4. Why did swaps settled on the public contract fall to almost nothing within about a week at the start of November?
5. Who may submit transactions to the contract today, and is there a plan to open relaying to others?
6. Why did the share of volume kept by NEAR fall in November, and what is the intended long-run take?
7. How many independent solvers fill confidential swaps, and how is their liquidity concentrated?
8. Which figures about the confidential ledger can an outside auditor verify today?

Appendix A. Method

• Sampling. The 30 days before block the end height are cut into equal strata; one 40-block window is drawn in each with a seeded generator, and every transaction touching intents.near in it is downloaded with its event logs. Main sample 720 windows; two further samples of 360 windows each with different seeds.
• Valuation. 99.7% of token legs are priced, most at the asset's price on the day of the transaction.
• Intervals. 95% window bootstrap, 2,000 draws. Pooled figures are inverse-variance averages.
• Check on the sampler. The sample's transaction count is 0.967× the exact count over the same days.
• Reproduction. python3 rebuild.py recomputes every figure in this note from the committed raw data; the test suite re-derives the sampling plan from the seeds.

29 of 39 re-checks give the same verdict as the main sample. No verdict flips to the opposite side. Differences: T1 is inconclusive in “Replication sample” (supported in the main sample); T2 is inconclusive in “Replication sample” (supported in the main sample); T4 is inconclusive in “Replication sample” (supported in the main sample); T1 is inconclusive in “Third sample” (supported in the main sample); T2 is inconclusive in “Third sample” (supported in the main sample); T4 is inconclusive in “Third sample” (supported in the main sample); T1 is inconclusive in “Older half only” (supported in the main sample); T3 is inconclusive in “Older half only” (supported in the main sample); T2 is inconclusive in “Recent half only” (supported in the main sample); T4 is inconclusive in “Recent half only” (supported in the main sample). T5 gives the same verdict in every run, and the treasury's share is between 86% and 92% in every run. T2 and T4 depend on a few very large deposits and transfers, so the headline marks them inconclusive. A third sample (360 windows, 11,387 transactions) was drawn after the verdict rule was fixed, so it is an out-of-sample check. On its own it gives T1 inconclusive, T2 inconclusive, T3 supported, T4 inconclusive and T5 contradicted: the same as the headline for 4 of 5 tests, and the opposite verdict for none. The headline verdicts were not changed after seeing it. Pooling all three samples gives T1 0.64× (95% CI 0.53× to 0.76×), against 0.70× (95% CI 0.56× to 0.84×) for the first two: still inside the 0.5×–2× band, but nearer its lower edge. All three samples put deposits below the published volume.

Appendix B. Limits

• It is a sample over 30 days. Totals are estimates; value shares depend on a few very large transfers.
• The confidential ledger is seen from outside only. What happens on the private contract is taken from NEAR's documentation, not observed. Shielded value includes solver liquidity and is an upper bound on confidential swap size.
• One account is identified from NEAR's public documentation. Every other account is described by behaviour; we do not claim to know who operates it.
• The rule for combining samples was written after the first two were drawn. The third sample was drawn after it.
• Peer figures come from different DefiLlama adapters and are a rough comparison. The BigQuery cross-checks were written but not run.

Appendix C. How NEAR Intents works, and a glossary

A user on another chain deposits; the public contract intents.near keeps the ledger; swaps settle either in the open or on a private contract behind one treasury account; relayers submit nearly every transaction.

Appendix D. Sources

The repository adds METHODOLOGY.md (estimators, event dictionary, a worked example), TUTORIAL.md (seven hands-on exercises) and SOURCES.md (every endpoint and reference, and what could not be used).