1 · Base layer
Most of NEAR's "activity" was one app, and it left.
In March 2026 one account, hotwallet.kaiching, signed 62% of all transactions. It stopped within 32 hours on 1 and 2 April, and monthly transactions fell 74%. About 560.1K of its accounts were then deleted in early October. Of 266,605 other signers in October 2026, 25,434 were engaged, down 77% since July 2025. That fall is mostly Telegram game accounts (down 84%). Engaged self-custody signers fell 29%, then rose 3.4x since August to 10,568. We do not know why yet.
2 · Is the volume real
Settled flow is real. How much of it is end-user demand is open.
The official dashboard and DefiLlama match to the dollar only since 2026-09-01. Before that they diverge, and revenue differs by month. So we rebuilt deposits and withdrawals from raw token events. One-way on-chain flow runs at 57% of reported volume, between 51% and 62% across 26 weeks. That flow includes solvers rebalancing through the same contract, so it is a ceiling, not end-user demand. Outside the top 100 addresses, withdrawals were 15% to 16% of reported volume in two weeks we priced, and 24% outside only the top 10. End-user demand sits between about 16% and 24% of reported volume.
3 · Who keeps the fee
Wallets own the customer, so wallets keep the toll.
The protocol's own fee is 1 pip, or 0.0001%. Everything else is a commercial fee set by whoever owns the customer. Swap-level data from the Intents Explorer shows swapkit carries 37% of 1Click volume and 70% of app fees, at about 77 bps. By NEAR's own count, NEAR kept 36% of gross fees in the last 30 days and 19% all time; partners, wallets and solvers kept the rest. Most of the gain in what NEAR keeps came from its own front end, not from partners. How the fee numbers fit: in October 2026, $100 swapped paid $0.160 in fees (16.0 bps); NEAR's 30-day split gives it 35.9%, so $0.057 (5.74 bps). NEAR's monthly revenue series puts calendar October 2026 at 6.28 bps ($2.5M of $6.4M in fees, 39.2%; the split's 30 days end at the 2026-11-10 pull), and the last three months at 4.65 bps, which is what kill row T3 grades. Same source, different windows; revenue a year is $30M at October's rate and $22.3M on three months (row A2).
4 · Token accrual
Buybacks are real. "Permanently removed" is not what the chain shows.
The official page says buybacks "permanently remove NEAR from circulation", and its FAQ says revenue "can be deployed" for buybacks, locking or staking. On-chain so far, the NEAR is held: 4.23M NEAR sits in two wallets, 99% of all official inflows, the largest daily burn since the buyback began was 3,851 NEAR, and a 2-confirmation multisig can move it. That is a discretionary policy, not a claim. In October 2026, burn plus captured revenue equalled 20% of issuance at best.
A larger flow we can see but not explain
treasury.near sent 30.17M NEAR out in April and May 2026, 7.1x what the revenue wallets hold and about $108.6M at today's price. It went to two multisigs and one hex account in single-day batches, and those three accounts are now close to empty. We can show the first hop. We cannot say who received it next or whether it was sold.
5 · Pricing power: the new test
Can NEAR raise its take without losing the flow? We could not tell.
A toll road is worth what it can charge, so we tried to measure that. Test one: the partner revenue share went live on 23 February 2026. The take did not step at that date: it went from 1.75 bps to 2.66 bps on three months either side, and most of that came in May. Test two: the take rose again from September, 3.00 bps to 5.39 bps, and rail share fell from 69% to 62%. Test three: across 19 lasting rises in a partner's posted fee, its share of swaps changed -12% on average, with a 95% interval of -52% to +43%. Too wide to read, and most of these partners are small.
What this does and does not show
Rail share was unchanged in one window and lower in the other, and neither interval rules out a 25% loss of volume against rivals. Most of the gain in capture is NEAR's own front end taking a larger share, which is mix, not price. So these tests do not show pricing power, and they do not show its absence. We keep the model's fee inputs as assumptions and say so.
6 · Staking and security
Stakers are paid in new tokens, and fees barely cover it.
Security spend is $117M a year at today's price, and fees burned cover about 1.1% of that. Issuance is mostly a transfer, not a cost to holders as a group: a staker earns about 4.8% after commission, or 2.3% after supply growth, and an unstaked holder is diluted about 2.5% a year. Only the 10% treasury share and commission leave the holder base, so the valuation treats issuance that way. Offsetting issuance is NEAR's stated goal. It is not what sets value in our model.
7 · Competitors on the same metrics
In line with chains. Far richer than tokens with real payouts.
The bull and bear cases are arguing about which comparison applies. Against gas fees on other chains, NEAR's market cap is about 148x what it keeps ($31.9M a year: L1 fees plus captured revenue), inside the peer range (median 394x). Against tokens with holder revenue, it trades at 157x captured revenue alone (October's $30M a year), or 102x after subtracting a peer-median chain value, against a median of 27x. Those comps are app tokens and NEAR is a whole chain, so neither comparison is clean.
8 · The hypothesis that volume leads usage
Does Intents volume lead network usage better than price does? Not in this data.
We ran six forward tests over 82 weeks. In none did volume beat price with an interval clear of zero. Out of sample, with overlapping windows purged, adding volume improved on a usage-only forecast in 2 of 6 tests, and adding price did in 3 of 6. Neither is a useful leading signal yet.
9 · AI narrative and developers
The agent story is mostly not on-chain yet.
Contracts with AI or agent names took 0.14% of October transactions, and 0.01% without the largest one. One real signal cuts the other way: the account escrow.ai.near signed 25% of direct calls into intents.near in October, though we cannot say how much value those calls move. Off-chain inference cannot be measured from here.
10 · Valuation
Change any assumption and see what NEAR is worth.
The model discounts holder earnings over five years, adds an exit value, and adds what NEAR is worth as a chain apart from Intents. At base-case inputs, $3.60 needs 63% volume growth a year for five years, 12x today. With a 40x exit it needs 7x, and with a $3.9B chain value 3x. Our weights give $2.91, but that leans on the bull weight: $2.16 at 10%, $3.67 at 30%, and about 29% gets to the market price. In the base case, chain value is 62% of the total ($1.66B of $2.69B, or $1.27 a NEAR over 1.31B NEAR), and it is an assumption.
11 · Kill list
What would prove this thesis wrong, with dates.
Set on 2026-11-01. Two rows were added after review and graded on the same snapshot as the rest: T6, an exploit rule, on 2026-11-12, and A5, which watches the chain premium, later the same day. The other nine thresholds are unchanged; on 2026-11-13 three row titles gained the window they measure (T2, T3, A2). Each row is graded by code at every build, with no human step (T6 reads a hand-kept incident list). If a toll row trips, Intents is not the durable toll road we say it is. If a token row trips, our caution on the token was wrong.
Criteria, graded by code at each build
Thresholds and dates are fixed. Values are from the build graded on 2026-11-13 (data through 2026-10-31).
Would kill the toll road
Would kill our caution on the token
12 · Limits
What we could not measure.
- Most reported volume sits outside the Explorer. It covers 1Click swaps, about 34% of reported volume on sample days. The definition behind the reported series is not published.
- Confidential intents are invisible by design. Official TVL there is about $143M. We cannot see who holds it or why.
- Solvers and market makers are not labelled. Large addresses may be solvers, exchanges or real whales, and small ones may be bots. That is why end-user demand is a wide range.
- Raw flow is not always economic. One week in October 2025 shows a 47M SOL deposit and same-day withdrawal, 26x reported volume. Medians keep it out of our ratio, and we do not know what it was.
- Security. A bug in the Omni deposit path cost about $3.8M on 1 October 2026, and the funds were returned (report). Kill rule T6, added on 2026-11-12, now covers an exploit: it trips if more than $10M of user funds stay unrecovered within a year. It reads a hand-kept incident list (thesis/incidents.json), the one manual input here.
- Off-chain AI usage and Foundation holdings are not measurable here. Sources also disagree on circulating supply: the NEAR public dataset gives 1.31B NEAR, NearBlocks 1.25B NEAR, and CoinGecko counts the whole 1.31B NEAR supply as circulating.
- Rail share is among native swap rails only. Relay, Across and Circle CCTP volumes sit behind a paid API. Relay's reported fees are about four times Intents'.
- The sample is short. Volume has about 22 months of history and fees about 18, in one market regime. Intervals here are wide on purpose.
- The kill list is ours. Thresholds were set by us on 2026-11-01 (T6 on 2026-11-12), so the raw values sit beside each one.
This is research, not investment advice. The authors make no claim about holdings.