NEAR's long-term case is a bet that NEAR Intents becomes the settlement layer for cross-chain money moved by software, including autonomous agents. We test the three halves of that bet separately, from public data only.
Usage moved to NEAR Intents: it settled 82% of NEAR's DEX volume over the last 6 full months and $11.71B in the 90 days to 2026-11-04, 1.80× the 90 days before (T1, T2).
Some of that value reaches NEAR holders, through buybacks rather than protocol fees. Since the fee switch of 2026-02-23, captured Intents revenue buys NEAR on the open market. In the 90 days to 2026-11-04 that was $4.8M, 27% of Intents' fees; NEAR's gas burn added $254.1K (C1).
It is small next to the token. Annualized, captured value is $20.6M, 0.42% a year of NEAR's $4.92B market cap. Over the same months NEAR's total supply grew 2.5% a year net of burns, about $120.8M a year at $3.75; the buyback is 16% of that. The bought-back NEAR is held (2.19M NEAR in buybacks.multisignature.near), not burned. If NEAR governance's HSP-029 cut (max issuance 2.5% → 1.6% over 24 months) runs its full ramp, the same buyback would cover 25% of net issuance.
NEAR's weekly price does not move with Intents volume (r = 0.00 over 95 weeks, T4; with BTC's moves taken out, r = -0.03, p = 0.79, T4m), and agent commerce is tiny next to Intents (T7). Small tickets are not an edge case: 48% of sampled swap legs and 47% of withdrawals were under $50 (P1). Agent-size tickets into NEAR get a 1Click quote: 0–0% under $50 (T9, quotes of 2026-11-06 01:09Z).
Read as an investment case: the usage half is supported by the record so far (T1 and T2 describe the past; the forward tests F1 and F2 check it on new data), value capture exists but covers about 16% of issuance, and the agent half is still an option. The case for the token is that the captured share grows faster than supply; we would call it made when the buyback reaches half of net issuance over a 90-day window.
Model: when does the buyback reach half of new supply?
Model (`node model.mjs`, every assumption is a flag): starting from today's 16%, at the fade the data implies (Intents' growth halving about every 1.0 quarters, see below) the buyback reaches half of net issuance not within 10 years; with a slower fade, halving every 2 quarters (the default flag), in 2027-Q4 (4 quarters); with no further growth it stays at 16%; if the measured 1.80× per 90 days never fades, in 2027-Q2 (2 quarters). A scenario, not a forecast. The measured quarter-on-quarter growth went 6.3× → 5.7× → 2.3× → 0.7× → 1.1× → 1.8×; fading from the oldest quarter to the latest four implies a half-life of about 1.0 quarters, so the 2-quarter default is on the generous side. The bought-back NEAR sits in a multisig (held, not burned); `--resold` models some of it returning to market. If HSP-029 (max issuance 2.5% → 1.6% over 24 months; House of Stake vote 99.97% for, timelock until 2026-11-16; no start date set) began its ramp now: at the implied fade, in 2028-Q4 (8 quarters); at the 2-quarter default, in 2027-Q3 (3 quarters).
Coverage = NEAR bought back per year ÷ net new NEAR per year. Now 16.2%. At the fade the data implies (growth halving every 1.01 quarters): 31.8% in 4 quarters, 33.4% in 8. With the slower default flag (halving every 2 quarters): 55.1% in 4 quarters, 77.9% in 8.
growth fade
buyback share 27.1%
buyback share 50.0%
buyback share 100.0%
no growth
– / –
– / –
0 / –
half-life 1 q
– / –
2 / –
0 / 2
half-life 2 q
4 / –
2 / 4
0 / 2
half-life 4 q
3 / 5
1 / 3
0 / 1
never fades
2 / 4
1 / 3
0 / 1
Quarters until the buyback reaches half / all of net new supply (– = not within 10 years), NEAR price flat. Re-run with your own numbers: node model.mjs --half-life 4 --capture 0.5 --price-q 0.1 (formulas in lib/model.mjs).
Tests
T1PASSUsage
Did NEAR usage shift from counts to settled value?
Over the last 6 full months (2026-05 → 2026-10), NEAR Intents settled 82% of NEAR's DEX volume. From 2025-01 to 2026-10, NEAR's daily transactions fell 90% and its active accounts 92%, while monthly Intents volume grew 622×.
Rule: share >= 0.70 AND txns and active accounts down >= 50% AND Intents monthly volume up >= 50x (2025-01 -> last full month)
Intents settled $11.71B in the 90 days to 2026-11-04, 1.80× the $6.49B of the 90 days before. Cross-check: DefiLlama's daily volume moves with NearBlocks' independent daily count of intents.near transactions (r = 0.64 for daily changes over 99 days).
In the last 90 days the NEAR protocol earned $363.1K in fees, 2.0% of the $17.8M that NEAR Intents earned. This test counts protocol fees only; Intents' value now reaches holders through buybacks instead (C1).
Does NEAR's price move with Intents volume once the crypto market (BTC) is taken out?
Over 95 weeks, BTC explains 34% of NEAR's weekly price moves (beta 1.42). What is left after that does not track Intents volume (r = -0.03, p = 0.79; before removing BTC, r = -0.00, p = 0.97). Added after the lock; T4 itself is unchanged.
Rule: post-lock REPORT (2026-11-05), no threshold: OLS beta on BTC, residual vs Intents volume, two-sided permutation p (10,000 shuffles, seed 1)
How much of NEAR's on-chain traffic is Intents right now?
Intents received 24% of NEAR's transactions on 2026-10-30 → 2026-11-05 by NearBlocks' daily stats, and 20% of the 11234 transactions in our 135 RPC block samples. NearBlocks' hourly counter, an estimate, says 4%; it is kept for the record only.
Rule: reported, no threshold (daily stats and RPC sample; the hourly counter reads an estimated count, kept for the record)
What share of a live agent's money flow is cross-chain?
The agent has sent 753 transactions (robinhood 697, bsc 4, hyperevm 1, near 51). 3 of them (0.4%; 0.4% of its 702 EVM transactions) moved money across chains, at a cost of 5.42%, 4.92%, 0.38%. Its 51 NEAR transactions are the companion project anchoring its records, not demand.
data/agent-ground-truth.json (every bridge the agent sent, each with its tx hash)
{
"sent_txs": 753,
"chains_with_sent_txs": [
"robinhood:697",
"bsc:4",
"hyperevm:1",
"near:51"
],
"nonces_read_at": "2026-11-06T01:09:55.803Z",
"cross_chain_ops": 3,
"cross_chain_share": 0.00398406374501992,
"evm_sent_txs": 702,
"cross_chain_share_evm_only": 0.004273504273504274,
"cross_chain_cost_pct": [
5.42,
4.92,
0.38
],
"cross_chain_ticket_usd": [
0.466068,
2.019158,
1
],
"near_account_txs": {
"sent": 51,
"nearblocks_txns_only": 51,
"what": "SocialDB set calls (receiver social.near) that anchor the companion project's records on NEAR mainnet; single-chain, not bridges"
},
"decisions": 2
}
T9REPORTAgents
What does a cross-chain move cost at agent ticket sizes, through NEAR Intents vs the route the agent used?
Robinhood Chain ETH → BSC BNB: NEAR Intents refused $0.47, $2, $30 and accepted the rest at 0.31%. Robinhood Chain USDG → NEAR (native): NEAR Intents accepted every ticket at 0.00%. (quotes of 2026-11-06 01:09Z)
Rule: reported, not scored (added after the first probe on 2026-10-31; see THESIS.md)
How much of Intents' usage reaches NEAR holders, next to NEAR's issuance?
In the 90 days to 2026-11-04, $4.8M of Intents revenue bought NEAR back (27.1% of Intents' fees) and gas burn added $254.1K: $20.6M a year, 0.42% of NEAR's market cap. NEAR's supply grew 2.5% a year net of burns ($120.8M); the buyback is 16% of that. Cross-check: the buyback wallet holds 69% of the NEAR the series would have bought at daily prices (outflows not decoded).
Rule: reported, not scored (added 2026-10-31 after the lock: the 2026-02-23 fee switch, missed before the lock; see THESIS.md)
From 300 random blocks over 7 days (163 transactions to intents.near): transfer (mostly fee payouts, see THESIS.md) n=204, 85% under $50, 79% under $10, median $0.29; token_diff n=98, 48% under $50, 26% under $10, median $65; withdraw n=55, 47% under $50, 21% under $10, median $68. Agent-size describes the ticket, not who sent it.
Rule: reported, no threshold (cutoffs < $50 and < $10 registered before sampling; THESIS.md)
Value captured for NEAR holders vs new NEAR supply, per year
Annualized from the 90 days to 2026-11-04: Intents fees $72.4M, of which $19.6M bought NEAR back; gas burn $1.0M; net new NEAR supply $120.8M at $3.75 (C1).
Built by claude-10x, an autonomous AI agent operated by gh, for NEAR HACKATHON. Raw inputs: data/raw/*.json, data/nearblocks-charts.json, data/snapshots.jsonl, data/agent-ground-truth.json.