Every lever, and the signatures that move it
One picture of the whole path: users' fees reach NEAR Intents, then the three revenue wallets, and stop
there; below it, new NEAR reaches the treasury and the stakers. The colour says how many signatures move each link
(data/control_map.json). It does not show why a link is set the way it is.
On a phone the picture scrolls sideways.
Who can change each link from the fee a user pays to the NEAR a holder keeps
The same links as the picture above, card by card: each one's value at the locked end block, who signs to change it, and how often they have. It does not show why a link is set the way it is.
Fees → holders
intents.near1 pipDAO 3 of 5 · used 6 timesIssuance → stakers
treasury.nearpaid out 30,171,200 NEAR in this yearmultisig 2 of 4Votes → rules
vote.dao)5 reviewers choose what is put to a voteowner DAO 3 of 5 · used 5 timesThe 5 DAOs have 22 member accounts; the number in more than one DAO is 0. “Check live” calls a public NEAR RPC from your browser and compares its answer with the answer saved at the locked block.
The buybacks wallets keep 10.8% of the NEAR issued
NEAR per day kept by the wallets, against NEAR issued per day, month by month. Full coverage needs what the buybacks wallets keep to grow ×9.2. It does not show why the wallets receive what they do.
Rating, latest window: —
The README's rules applied to a rolling window of thirty days, every day (data/rating.csv). Avoid needs the
latest window and the one ending thirty days before it to fail one of the two tests, or the bought NEAR to fall more than a tenth below its peak; Buy needs rule (b) on chain first: a lock no one can spend from.
A window with missing DefiLlama days is judged on the second test alone if every day has a price, and the two newest windows may still be revised.
- Buy only when (b) a published rule sends the buybacks to a declared lock the chain shows cannot send NEAR out (no keys and no code, or reviewed code, no full-access key and only function-call keys limited to the lock) and
treasury.nearopened no new multisig request in the window, and either the NEAR they keep, counted up to DefiLlama's revenue and fees, covers all new NEAR, or (a) it grows +75% a year, in dollars and in NEAR. - Avoid when this thirty-day window and the one ending thirty days before it each fail a test: the wallets keep less than three quarters of DefiLlama's revenue, or the buybacks wallets keep April's $6,270 a day or less (at this window's NEAR kept, a price of $0.655 or less), unless DefiLlama's fees fell too and the Confidential Intents Treasury account's stablecoins grew. Also Avoid, at once, while the buybacks stock is more than a tenth below its highest since the first of November. Avoid overrides Buy; Hold has no upper price.
Also measured: NEAR Intents moved $4.02B in DefiLlama's thirty days to 7 November (not the window) · a staked holder's share of all NEAR grows 2.44% a year before the validator's commission, an unstaked holder's changes -2.40%; to 2031, after an assumed validator commission and with HSP-029, a staked holder's share grows 6.3% · since the flow went private the three revenue wallets kept 13,800 NEAR a day, against 14,261 before, while DefiLlama's revenue changed -6% in NEAR · the Confidential Intents Treasury account held $123.5M at the end of that comparison.
Peers: market cap at each token's mean price over the same month ÷ DefiLlama's yearly revenue. The other layer-one tokens' rows are their chains' own revenue, without their apps, so NEAR is also shown on gas alone.
The chain is used less: daily transactions not relayed -58% in a year
From NearBlocks' daily statistics (scripts/usage.py). All gas paid in those twelve months was
$1.2M, 0.8% of a year of new NEAR. Neither NEAR
Intents' volume nor the price leads this use by up to a month that we can detect (smallest p 0.44):
the track's own example hypothesis, tested. NEAR's public BigQuery dataset gives the same daily transactions on
365 of 365 days (scripts/bq_usage.py).
Both read NEAR's own block data: this checks the counting, not the chain.
NEAR Intents volume that NearBlocks sees on mainnet, as a share of DefiLlama's: 66% in the month before swaps began to settle privately, 28% in the thirteen days after, 12% in the first week of November, as swaps moved to the private path.
Who calls NEAR Intents' contract (BigQuery, October)
3,515,141 calls to intents.near, by path.
| Path | Share of calls | Failed |
|---|---|---|
Relay calls: execute_intents signed by intents.near's own function-call keys, carrying users' and solvers' signed intents | 40% | 28,731 |
| The contract's own callbacks (withdrawals, storage deposits) | 23% | 4 |
Other contracts, mostly token deposits (bridged *.omft.near tokens, native USDC, wrap.near and others) | 21% | 48 |
escrow.ai.near, the NEAR AI Agent Market's escrow, calling execute_intents itself | 14% | 496,805 of 496,807 |
| Transactions other accounts sign to it | 2% | 212 |
Successful relay calls, the path swaps settle on (one can carry several intents): 45,918 a day in October,
103,267 in the first eleven days of November. The escrow's own calls fail almost every time; whether the Agent Market's
payments settle inside relayed intents is not separated here. With this on-chain activity in place of DefiLlama's volume, still no lead
(smallest p 0.17). Calls, not dollars. Queries: sql/.
The same rules on four other protocols
Are the rules built so that everything comes out Hold? scripts/peers_rating.py runs the same gates on
Hyperliquid, Pump.fun, Aave and Jupiter: a lock first, then tokens bought a year against new tokens a year.
From their docs and DefiLlama, not read on their chains: no drain test and no growth leg, so none can be Avoid.
The flow went private; capture did not stop
NEAR the three revenue wallets (the buybacks wallets and the front-end fund) kept each day on mainnet. The dashed line marks the day from which Tideline, another public repository, reports private settlement growing on NEAR Intents; the flat lines are the mean before (14,261 NEAR a day) and after (13,800). DefiLlama's revenue, which counts the same wallets, changed -6% in NEAR over the same days; in dollars, both fell about a fifth. Of five dates tried for the change, Tideline's gives the smallest fall. It does not show which fees paid for what the wallets received.
python scripts/private_shift.py --verify recomputes every bar from the saved balances.
Even every fee would buy 47.9% of the NEAR issued
What all fees recorded by DefiLlama would buy in NEAR per day, against what the wallets kept and what was issued. Fees must grow ×2.1 to cover issuance. It does not include confidential volume, solver spreads or gas.
The lever model: how much each lever adds, and who moves it
Coverage if each lever were pulled in turn, from today to full coverage. Only issuance is a rule; every other step is a choice of the signers named beside it. It does not predict that any lever will be pulled.
Re-run it
The same formula with your own assumptions: coverage = fees in dollars × growth × share kept ÷ NEAR price ÷ (NEAR issued × issuance factor). At today's values it reproduces the figures above exactly.
To cover issuance by 2031, what the buybacks wallets keep must grow +75% a year (+56% once validators run HSP-029)
NEAR issued per day on the first day of each year, under today's rule (at most 1/40 of supply a year) and under HSP-029 (from 2027-02-14 16:48 UTC), against what the buybacks wallets keep a day today. Keeping every fee instead of today's share can add ×2.9, once; from April to October what the three wallets kept rose ×2.9 while all fees counted in NEAR changed ×0.5. It forecasts no price, volume or fee; it holds the window's ratio of actual to maximum issuance and the window's level of what is kept.
One account holds 73.3% of the NEAR AI farm
Active accounts and the largest account's share, on the first block of the first and the fifteenth of each month. The credits rate changed once, ×2.0 on 2026-09-29 14:21 UTC: credits now pay more than a validator only while NEAR is below $4.84 (before: $2.42). It does not show who holds the largest position, or whether credits are used.
treasury.near paid out 30,171,200 NEAR this year, all in April and May
Balance at the end of each month. Its reward rule is reproduced at 101 of 101 epoch starts. The purpose of each payment is not on chain.
| Receiver | NEAR |
|---|---|
ecotreasury5.multisignature.near | 4,900,000 |
ecotreasury4.multisignature.near | 4,900,000 |
one implicit account (tag sha256:e630d735…) | 20,371,200 |
foundation.near (came back seconds later) | 3 |
Passed is not in force
A House of Stake vote changes nothing on chain by itself. Left: the share of stake whose block producers signal the protocol version being watched (HSP-027's, then the next release once it took effect), against the threshold. Right: the veNEAR of each HSP-029 voter, largest first; the first 6 reach the quorum. It does not show which release each validator runs before it signals.
To watch. A House of Stake draft of 8 November would put a share of NEAR Intents' fees in US Treasury bills, and holders would vote on buybacks or dividends only after venture exits. It is a draft: no number, no vote, nothing on chain. It is a reason not to add to a position, not a reason to sell one.
Eight predictions, anchored on chain before they are judged
The file's SHA-256 is stored in SocialDB by a NEAR transaction; anyone can read it back.
PREREG.mdWhat this does not show, and how to check it
- Confidential NEAR Intents volume: not visible on chain (NearStats reports a figure, not checked here).
- Volume rebuilt from the chain: DefiLlama's volume comes from a NEAR dataset on Dune.
- Credits in dollars: an off-chain NEAR AI Cloud setting, read only with a signed-in session; whether credits are used.
- Who holds the largest farm position, and where its NEAR came from.
- What
treasury.near's payments were for.