# Thesis: NEAR Intents' buybacks are a policy, not a rule

*Investment Thesis track, NEAR HACKATHON. Dated 13 November 2026, rating rules as of 12 November (the drain test was added on 12 November, after the predictions were anchored, so it is not pre-registered); horizon 1 January 2031, checked every 12 months. Window 2 October → 1 November 2026 (00:00 UTC), blocks 218,126,921 → 222,491,356 (`data/window_2026-10-02_2026-11-01.json`). Every other figure measured here is copied from [`data/numbers.json`](data/numbers.json) with a hidden marker, and `python scripts/build_numbers.py --check THESIS.md` fails if one differs. Thresholds are the author's choice. The longer version of this file, as of 8 November, is in [DETAILS](docs/DETAILS.md#thesis-in-full-version-3-8-november).*

## Claim

Users pay real fees on NEAR Intents, but only a small part reaches NEAR holders: the buybacks ([Terms](README.md#terms)) hold NEAR equal to <!--n:n1.coverage_buybacks-->10.8%<!--/n--> of new issuance and burn none. Every link from the fee a user pays to the NEAR a holder keeps can be changed by a few signers (<!--n:c.buybacks_confirmations-->2 of 4<!--/n--> for the buybacks wallet) or by a company off-chain; none is fixed in code. Issuance, on the other side of the ledger, is a protocol rule.

**Rating: Hold, not Buy, to 1 January 2031.** Hold means keeping NEAR already owned, staked: staking pays about <!--n:iv.staking_yield_gross-->5.0%<!--/n--> a year before the validator's commission while supply grows <!--n:iv.issuance_rate_per_year-->2.46%<!--/n-->, so a staked holder's share of all NEAR rises. It is not a Buy because what holders get from fees covers a small part of issuance and nothing obliges anyone to keep it. Growth is not what is missing: from April to October what the buybacks kept grew <!--n:bt.a_pace.near_multiple-->×2.1<!--/n--> in NEAR and <!--n:bt.a_pace.usd_multiple-->×7.5<!--/n--> in dollars, more than Buy's growth test asks (<!--n:bt.a_pace.required_multiple-->×1.32<!--/n-->), though from a low April and mostly through the share of fees NEAR keeps (<!--n:lr.history.2026-04_2026-10.take_rate_multiple-->×4.2<!--/n-->), which has a ceiling. What is missing is a lock. On the whole token: at Hyperliquid's, Jupiter's and Aave's multiples, NEAR Intents' revenue would be worth <!--n:sc.split.intents_share_of_value.low-->10%<!--/n-->–<!--n:sc.split.intents_share_of_value.high-->16%<!--/n--> of NEAR's value if the policy holds (<!--n:sc.split.fdv.intents_share_of_value.low-->16%<!--/n-->–<!--n:sc.split.fdv.intents_share_of_value.high-->39%<!--/n--> at their fully diluted value ÷ revenue); the other <!--n:sc.split.premium_share.low-->84%<!--/n-->–<!--n:sc.split.premium_share.high-->90%<!--/n-->, mostly a layer-one premium, is not rated here. If you hold no NEAR, Hold means do not buy yet; Avoid would mean not holding it for this thesis.

The rating turns to **Buy** only when a published rule sends the buybacks to a lock, declared in `daily.LOCKS`, that the chain shows cannot send NEAR out (no keys and no code, or code we reviewed, no full-access key and only function-call keys limited to the lock), and either the buybacks cover issuance in a window in which `treasury.near` opens no new multisig request, or, at a yearly check whose window is quiet in the same way, what they keep has grown <!--n:lr.required_growth.buybacks.A.2031-->+75%<!--/n--> a year in dollars and in NEAR. It turns to **Avoid** at once if the bought NEAR the wallets hold falls more than 10% below its peak since 1 November (one transfer of more than a tenth of it is enough), or if two back-to-back 30-day windows each fail a test: the wallets keep less than 75% of DefiLlama's revenue, or the buybacks keep April's dollars or less. The rules as code, every edge case included, are in the [README](README.md#the-rating-as-rules).

## Evidence for

**Real.** Users pay <!--n:n2.fees_usd_per_day-->$206,049<!--/n--> a day in fees, <!--n:n2.fees_bps-->16.0<!--/n--> bps of volume (DefiLlama, from a dataset on Dune that is not rebuilt here). On chain, the wallets' balances agree with DefiLlama's revenue (<!--n:n1.coverage_with_front_end-->16.5%<!--/n--> and <!--n:n2.defillama_revenue_coverage-->16.5%<!--/n--> of issuance): almost nothing flows out.

**It did not stop when the flow went private.** Tideline, another public repository, reports private settlement growing from 27 October. Read on mainnet, the three revenue wallets kept <!--n:ps.after.kept_near_per_day-->13,800<!--/n--> NEAR a day from then to 8 November, against <!--n:ps.before.kept_near_per_day-->14,261<!--/n--> from 2 to 26 October (<!--n:ps.change.kept_near_per_day-->-3%<!--/n-->, inside the range of any <!--n:ps.after.days-->13<!--/n--> days before; a block bootstrap gives <!--n:ci.shift.kept_near.low-->-28%<!--/n--> to <!--n:ci.shift.kept_near.high-->+28%<!--/n-->, so the 13 days rule out a stop, not a small fall), while DefiLlama's revenue, which counts the same wallets, changed <!--n:ps.change.revenue_near_per_day-->-6%<!--/n--> in NEAR. With the change on other days from 20 October to 2 November, the wallets' change runs from <!--n:ps.cut.kept_near_lowest-->-13%<!--/n--> to <!--n:ps.cut.kept_near_highest-->-3%<!--/n-->; Tideline's date gives the smallest fall. In dollars both fell about a fifth (<!--n:ps.change.kept_usd_per_day-->-19%<!--/n--> and <!--n:ps.change.revenue_usd_per_day-->-21%<!--/n-->).

**Small.** The buybacks wallets keep <!--n:n1.buybacks_near_per_day-->9,570<!--/n--> of the <!--n:n1.new_near_per_day-->88,392<!--/n--> NEAR issued a day: coverage <!--n:n1.coverage_buybacks-->10.8%<!--/n--> (<!--n:n1.coverage_with_front_end-->16.5%<!--/n--> with the front-end fund). It has been below a quarter of issuance in every month since April (highest <!--n:f1.coverage_buybacks.2026-09-->15.3%<!--/n-->). If NEAR kept every fee, coverage would be <!--n:n2.max_coverage-->47.9%<!--/n-->.

**The chain's own use is falling, and it pays holders almost nothing.** From NearBlocks' daily statistics, NEAR's transactions fell from <!--n:u.txns.first-->4.27M<!--/n--> a day in November 2025 to <!--n:u.txns.last-->716k<!--/n--> in October 2026 (<!--n:u.txns.change-->-83%<!--/n-->; <!--n:u.txns_not_meta.change-->-58%<!--/n--> without relayed transactions; <!--n:bq.signers.without_top10_change-->-45%<!--/n--> without each day's 10 busiest signers, from BigQuery), most of it in one step in April, and all gas paid in those twelve months was <!--n:u.fee_usd_year-->$1.2M<!--/n-->. Over <!--n:u.lead.weeks-->52<!--/n--> weeks neither Intents' volume nor the price detectably leads that use (`scripts/usage.py`), the hypothesis the track's prompt example proposes. NearBlocks sees <!--n:u.visible.before_private-->66%<!--/n--> of DefiLlama's Intents volume before 27 October and <!--n:u.visible.week_85-->12%<!--/n--> on 2–8 November. NEAR's public BigQuery dataset gives the same daily transactions on <!--n:bq.agree.txns.days_equal-->365<!--/n--> of <!--n:bq.agree.days-->365<!--/n--> days, and shows who calls NEAR Intents' contract: in October <!--n:bq.paths.october.relay.share-->40%<!--/n--> of calls were relayed intents (users' and solvers' signed intents, which an off-chain relay submits, signed with intents.near's function-call keys), and successful relay calls (one can carry several intents) went from <!--n:bq.paths.october.relay.ok_a_day-->45,918<!--/n--> a day in October to <!--n:bq.paths.november.relay.ok_a_day-->103,267<!--/n--> on 1–11 November (`scripts/bq_usage.py`).

**Discretionary.** Each link can be changed by people (`scripts/control_map.py`, `scripts/control_history.py`):

- The bought NEAR, <!--n:n1.buybacks_held_near-->2,145,248<!--/n--> at the end block, sits in wallets that <!--n:c.buybacks_confirmations-->2 of 4<!--/n--> signers can move, and none is burned; with the front-end fund (<!--n:c.front_end_transfer-->2 of 5<!--/n-->), <!--n:n1.wallets_held_near-->4,109,089<!--/n--> NEAR.
- The fee in the contract is <!--n:c.intents_fee_pips-->1 pip<!--/n--> (0.0001%); what users pay is set off-chain, and the terms of 1Click, NEAR Intents' swap service, let its operator change fees "at any time on a prospective basis".
- `treasury.near` paid out <!--n:ts.year_out_near-->30,171,200<!--/n--> NEAR in April and May 2026, each payment a <!--n:c.treasury_confirmations-->2 of 4<!--/n--> multisig confirmation.
- Not a buybacks link, but the same way of setting things: one account raised the NEAR AI credits rate <!--n:f6.change1.ratio-->×2.0<!--/n--> on <!--n:f6.change1.utc-->2026-09-29 14:21 UTC<!--/n-->, with no proposal of the owner's DAO.
- HSP-029, the issuance cut, passed: <!--n:f8.approval_share-->100.0%<!--/n--> of the veNEAR (locked NEAR's voting weight) cast for or against was in favour (<!--n:f8.abstain_votes-->3<!--/n--> voters abstained), on a turnout of <!--n:f8.turnout_of_snapshot-->62.3%<!--/n-->; as stake-weighted votes allow, <!--n:f8.fewest_voters_reaching_quorum-->6<!--/n--> of its <!--n:f8.voters-->36<!--/n--> voters held enough veNEAR to reach the quorum alone. The vote executes nothing on chain; it takes effect when validators run a release that includes it, the same step that protects issuance.

Issuance, by contrast, is code: at most <!--n:c.issuance_max_inflation_rate-->1/40<!--/n--> a year, changed only by a release run by block producers with more than <!--n:c.issuance_upgrade_stake_threshold-->4/5<!--/n--> of their stake, as the halving upgrade was.

## Evidence against

1. **The policy has been kept.** From April to October the wallets kept <!--n:f2.wallets_kept_share_of_revenue.2026-04_2026-10-->96.9%<!--/n--> of DefiLlama's revenue, and the buybacks wallet's keys have not changed since <!--n:h.keys.buybacks.last_change-->2026-03-18<!--/n-->. "Discretionary" describes what can change, not a change that has happened.
2. **Capture is rising, in dollars.** From April to October what the buybacks keep grew <!--n:lr.history.2026-04_2026-10.buybacks_kept_usd_multiple-->×7.5<!--/n--> in dollars (<!--n:lr.history.2026-04_2026-10.buybacks_kept_near_multiple-->×2.1<!--/n--> in NEAR). "Small" may not last.
3. **Part of "small" is the price.** Revenue is earned in dollars and issuance paid in NEAR: at <!--n:n2.equilibrium_price_usd-->$2.33<!--/n--> or less every fee would buy as many NEAR as are issued.
4. **Governance is moving toward rules.** HSP-027 passed and has been in effect since <!--n:pr.v87.first_utc-->2026-11-04 11:46 UTC<!--/n-->, <!--n:pr.v87.days_from_hsp027_passed-->124<!--/n--> days after the vote. HSP-029 passed; at its target rate the ceiling would be <!--n:n2.hsp029_max_coverage-->74.8%<!--/n-->.
5. **Counting Intents, NEAR is the cheapest L1 here.** Its market cap is <!--n:peer.near_all.mcap_to_revenue-->234×<!--/n--> a year of revenue with Intents, against <!--n:peer.solana.mcap_to_revenue-->1,995×<!--/n--> for Solana and <!--n:peer.ethereum.mcap_to_revenue-->3,394×<!--/n--> for Ethereum; on gas alone it is <!--n:peer.near_l1.mcap_to_revenue-->4,091×<!--/n--> ([peers](docs/DETAILS.md#peers-on-one-basis)).

## What would change our mind

| Part of the claim | We are wrong if | Today | Check |
|---|---|---|---|
| Small | the buybacks wallets keep 25% or more of issuance in three consecutive calendar months | highest month <!--n:f1.coverage_buybacks.2026-09-->15.3%<!--/n--> | `scripts/coverage.py` |
| Discretionary | bought NEAR moves by a published rule to an account no one can spend from, or the contract's fee is raised with its collector set to the buybacks or a locked account | <!--n:c.buybacks_confirmations-->2 of 4<!--/n--> multisig; fee <!--n:c.intents_fee_pips-->1 pip<!--/n--> | `scripts/control_map.py` |
| Small, to 2031 | at a yearly check, what the buybacks keep has grown <!--n:lr.required_growth.buybacks.A.2031-->+75%<!--/n--> a year in dollars and in NEAR: at least <!--n:lr.checkpoint.A.2027.near_per_day-->16,716<!--/n--> NEAR a day on 1 November 2027, <!--n:lr.checkpoint.A.2028.near_per_day-->29,244<!--/n--> in 2028, <!--n:lr.checkpoint.A.2029.near_per_day-->51,083<!--/n--> in 2029 and <!--n:lr.checkpoint.A.2030.near_per_day-->89,231<!--/n--> in 2030 ([all floors](README.md#what-would-change-our-mind)) | <!--n:n1.buybacks_near_per_day-->9,570<!--/n--> NEAR and <!--n:iv.buybacks_kept_usd_per_day-->$46,692<!--/n--> a day | `scripts/rating.py` |
| Did not stop when private | PREREG-3's P6 fails: from 9 November to 9 December the buybacks keep less than half the dollars a day of the 2 October window | <!--n:iv.buybacks_kept_usd_per_day-->$46,692<!--/n--> a day in the window; the three wallets kept <!--n:ps.after.kept_near_per_day-->13,800<!--/n--> NEAR a day, 27 October to 8 November | `scripts/private_shift.py`; P6 judged 9 December |
| Issuance is the only rule | HSP-029 is still not in force on 1 November 2027 | in timelock until <!--n:f8.timelock_end_utc-->2026-11-16 16:48 UTC<!--/n--> | `scripts/upgrade_watch.py` |

A row can prove the claim wrong while the rating stays Hold: Buy also needs a lock that cannot send NEAR out, which we put at 0.25 by 2031 ([how likely a rule is](docs/DETAILS.md#how-likely-a-rule-is)). The thresholds are the author's; with other floors and growth rates, and even with a published rule assumed, all <!--n:sens.month_ratings-->378<!--/n--> monthly ratings from April to October are Hold (`scripts/sensitivity.py`), because a monthly rating reaches Buy only through coverage of 100%. The growth test is judged once a year, and with a published rule the April–October history would already pass it (<!--n:bt.a_pace.near_multiple-->×2.1<!--/n--> in NEAR and <!--n:bt.a_pace.usd_multiple-->×7.5<!--/n--> in dollars, against <!--n:bt.a_pace.required_multiple-->×1.32<!--/n-->). Before 11 December 2026 only a drain of the bought NEAR can give Avoid, because the two-window tests need a failing window 30 days earlier and those ending 1–10 November passed and are final. Eight predictions are anchored on NEAR mainnet before they are judged, and none has been amended ([`PREREG.md`](PREREG.md), [`PREREG-2.md`](PREREG-2.md), [`PREREG-3.md`](PREREG-3.md)).

## Metrics we cannot compute

- **Fees and volume from the chain.** They come from NEAR's datasets on Dune, through DefiLlama; the NEAR Intents Explorer API needs a partner key. Solver spreads and gas are not in the fees.
- **The private path.** Volume and fees settled on FAR are not visible; only what the Treasury account holds on mainnet is read.
- **Which fees fund the buybacks.** What arrives in the wallets is measured; no swap from fee income to those inflows is traced.
- **Credits in dollars, and their use.** An off-chain NEAR AI Cloud setting, read only when signed in.
- **Purposes.** Who holds an account and why NEAR moved are not on chain; amounts, blocks and transactions are.

## Reproduce

Python standard library only, no network, no keys. One command rebuilds every number and every chart's
input from the committed raw files, checks every marked figure in this file, the README and the details, and
ends with `REBUILD: PASS`:

```
python scripts/rebuild.py
```

The charts, each with its query: [dashboard](https://nikolaskahris-png.github.io/near-thesis-data/dashboard/).