# NEAR to 2031: the buybacks are a policy, not a rule [![verify](https://github.com/nikolaskahris-png/near-thesis-data/actions/workflows/verify.yml/badge.svg)](https://github.com/nikolaskahris-png/near-thesis-data/actions/workflows/verify.yml) **[Demo video, 96 seconds](https://x.com/nikmaker4t/status/2121285434391044317)** · [The thread on X](https://x.com/nikmaker4t/status/2121259594257485850) **NEAR Intents' buybacks are a policy, not a rule, and this repository checks them every day.** | | | |---|---| | **Rating** | **Hold** to 2031. If you hold NEAR, stay staked: stakers receive most of the new NEAR; unstaked NEAR receives none. If you don't, don't buy yet. | | **The number** | Users paid $206,049 a day in fees in October (DefiLlama), and part of that buys NEAR. The buybacks wallets kept 10.8% as much NEAR as was issued (11 cents per $1 of new NEAR); all three revenue wallets, 16.5%. None of it is burned or locked. | | **Why not Buy** | Nothing makes them keep it: 2 of 4 signers can move the buybacks' NEAR, and every other money lever we mapped on chain moves with 1 to 3 signers; only issuance is a protocol rule (the picture below). Until a lock exists, no price gives Buy; the same rules give Hyperliquid, whose bought tokens sit at a keyless address, Buy on one of four counts of its new tokens. | | **Why not Avoid** | The policy has held: from April to October the three revenue wallets kept 96.9% of DefiLlama's revenue for NEAR Intents, and after swaps began to settle privately on 27 October the NEAR they kept a day changed by -3% (to 8 November, against 2–26 October; [claim D](#d-it-did-not-stop-when-the-flow-went-private)). | | **What changes it** | Code re-reads mainnet every day ([every condition](#the-rating-as-rules)). **Avoid** at once if the buybacks' NEAR ends a day more than 10% below its peak since 1 November. **Buy** needs a sealed lock, a quiet treasury, and buybacks **×9.2** today's (counted up to DefiLlama's revenue and fees) or growing +75% a year in NEAR and dollars (+56% with HSP-029; [yearly checkpoints](#what-would-change-our-mind) to 2030). Next checks: 16 November, 9 and 11 December. | **The whole token.** At Hyperliquid's, Jupiter's and Aave's 24×–39× revenue, NEAR Intents would be worth 10%–16% of NEAR's $6.39B if the policy holds (16%–39% at their fully diluted multiples; October's prices). The rest, mostly a layer-one premium, is not rated, and the chain's own use under it is falling ([claim E](#e-the-chain-is-used-less-and-its-use-pays-holders-almost-nothing)). **What this repository adds:** a rating that is code and re-reads mainnet every day (Avoid can fire any day; Buy needs a lock that does not exist) · predictions hashed on chain before they are judged, one of them judged automatically by a daily GitHub Action · the same rules run on four other protocols · NEAR's own use counted from blocks through BigQuery, as the track asks ([claim E](#e-the-chain-is-used-less-and-its-use-pays-holders-almost-nothing)). ![Who holds each lever, from the fee a user pays to the NEAR a holder keeps: the revenue wallets move with a few signatures and nothing locks or burns; only issuance is a protocol rule](docs/levers.svg) ## Why Hold | Question | Answer | The number | |---|---|---| | How much of the new NEAR do the buybacks offset? | A small part | 10.8% of it, counted in NEAR on the two buybacks wallets | | Does a rule guarantee it? | No: it is a choice | nothing locks or burns it, and 2 of 4 signers can move it | | Is the demand real? | Yes, but most fees stay with the apps that route swaps | users paid $206,049 a day in fees (DefiLlama) | | Are holders paid more elsewhere? | Yes | on DefiLlama's holders' revenue (NEAR: all three revenue wallets): NEAR 0.40% of its value a year (almost all NEAR circulates), Hyperliquid 4.09% of circulating value, 1.03% fully diluted | | Did the buybacks grow, or the price? | Both | NEAR averaged $4.88 in the window, ×3.6 April's; per day, October's buybacks kept ×2.1 April's NEAR, ×7.5 its dollars | | Is the chain's own use growing? | No, except NEAR Intents | daily transactions: -45% in a year without each day's 10 busiest signers, -83% with them (-45%: BigQuery; -83%: NearBlocks, matched by BigQuery) | **What staking does to your share:** stakers receive the new NEAR minus the treasury's tenth (a protocol rule; 10.1% of net new NEAR, measured), so your share of all NEAR grows +2.44% a year staked (before the validator's commission) and falls -2.40% unstaked. ![For the NEAR issued each day, the buybacks keep about a tenth; all fees would cover under half](docs/coverage.svg) **[Live dashboard](https://nikolaskahris-png.github.io/near-thesis-data/dashboard/)** · [Try it in two minutes](#try-it-in-two-minutes) · [One-page thesis](THESIS.md) · [Every detail](docs/DETAILS.md) · [Terms](#terms) · Window: 2 October → 1 November 2026; any other date is named where used
The dashboard's first screen: the rating, Hold, and the drain test read from NEAR mainnet
The rating, and the drain test reading NEAR mainnet live every day.		 The levers picture on the dashboard
Every lever from fees to holders, and the signatures that move it.
The rating panel: this window's tests and the Buy and Avoid rules
Hold today; code states what would make it Buy or Avoid.		 The same rules applied to four other protocols
The same rules run on Hyperliquid, Pump.fun, Aave and Jupiter.
## Try it in two minutes 1. **Open the [dashboard](https://nikolaskahris-png.github.io/near-thesis-data/dashboard/).** You see the rating, Hold, and the drain test read live from NEAR mainnet: the NEAR the buybacks hold, against the line 10% below its peak. 2. **On the "Buybacks wallets" card, press Check live.** Your browser reads mainnet and compares with the saved answer. 3. **In "Re-run it", move the sliders.** They start at today's 10.8%. Press "Hypothetical: buybacks locked, fees ×9.2 → Buy" to see what Buy takes. 4. **Run `python scripts/rebuild.py`.** It ends with `REBUILD: PASS`: every number on this page recomputed offline, no network, no keys (responses that list people's accounts are read from redacted copies; the originals stay out of git, their SHA-256 recorded, and `--archival` re-fetches them), and PREREG.md checked against the hash anchored on mainnet. ## The thesis in five claims Sources for A–D: [long version](docs/DETAILS.md#v5-the-thesis-in-four-claims). ### A. The demand is real NEAR Intents moved $4.02B in the 30 days to 7 November (after the window; DefiLlama), 15th among protocols by DEX volume. **Against.** Volume and fees are DefiLlama's; NearBlocks, indexing the chain, sees 66% of that volume before the private path. Evidence, tables and checks: [EVIDENCE.md](EVIDENCE.md#a-the-demand-is-real) ### B. It is small Each day the buybacks wallets keep 9,570 NEAR, against 88,392 new NEAR issued: coverage 10.8%. **Against.** Coverage is in NEAR and revenue in dollars, so coverage falls when the price rises; in dollars the buybacks kept ×7.5 as much a day in October as in April. Evidence, tables and checks: [EVIDENCE.md](EVIDENCE.md#b-it-is-small) ### C. It is a policy, not a rule Nothing in code burns, locks or requires the buybacks. Each buybacks wallet moves with 2 of 4 signers. What flows in depends on off-chain revenue-share deals, and it has moved both ways: the share of users' fees DefiLlama counts as revenue for NEAR Intents was 15.4% in March, 8.1% in April and 34.3% in October, while the contract's fee stayed at 1 pip (last approved change proposed 2025-10-02). What has been proposed for the revenue points away from a lock: a co-founder's [sovereign-fund post](https://gov.near.org/t/near-governance-discussion-sovereign-fund/42419) (3 August 2026) and a House of Stake draft (8 November 2026) that would put part of the fees into US Treasury bills ([sources](docs/DETAILS.md#how-likely-a-rule-is)). **Against.** The policy has been kept: April to October the three revenue wallets kept 96.9% of DefiLlama's revenue, so the missing rule blocks Buy, not Hold. Evidence, tables and checks: [EVIDENCE.md](EVIDENCE.md#c-it-is-a-policy-not-a-rule) ### D. It did not stop when the flow went private From 27 October swaps began to settle privately on FAR. To 8 November, the NEAR kept a day by the three revenue wallets changed by -3% (-19% in dollars, mostly the price). The NEAR figure is inside the noise (-28% to +28%); a stop would be near -100%. **Against.** Thirteen days show capture did not stop, not that it held level; of five start dates tried, the worst gives -13%. Evidence, tables and checks: [EVIDENCE.md](EVIDENCE.md#d-it-did-not-stop-when-the-flow-went-private) ### E. The chain is used less, and its use pays holders almost nothing Leaving out each day's 10 busiest signers, daily transactions changed by -45% in a year (-83% with them; NearBlocks and BigQuery agree on every day), and all gas paid in that year was 0.8% of a year of new NEAR. Successful relay calls into NEAR Intents rose +125% a day from October to 1–11 November, but these are calls, not dollars, and the rating does not use them. **Against.** The track's suggested test, whether Intents' volume or the price leads the chain's use, finds nothing over 52 weeks, but that many weeks give it little power: "not detected" is not "no effect". Evidence, tables and checks: [EVIDENCE.md](EVIDENCE.md#e-the-chain-is-used-less-and-its-use-pays-holders-almost-nothing) ## What a holder earns to 2031 Only staking moves your share of all NEAR. To 1 Jan 2031, with HSP-029 and a 10% commission: **+6.3%** staked, -7.4% not staked. No price target ([paths to 2031](docs/DETAILS.md#v4-what-a-holder-earns-to-2031)). NEAR is valued at 248× the revenue DefiLlama counts for NEAR Intents (the three revenue wallets); Hyperliquid, Jupiter and Aave at 24×–39× theirs: at those multiples NEAR Intents would be worth $0.61B–$1.02B, if the policy holds. Layer-one tokens are priced on more than fees; the other 84%–90% of NEAR's value, mostly that premium, is not rated here (Intents' share is 16%–39% at the peers' fully diluted multiples; [more](EVIDENCE.md#what-a-holder-earns-to-2031)). ## The rating as rules `scripts/rating.py` applies them daily to each 30-day window ([`data/rating.csv`](data/rating.csv)). One question: does a rule, rather than a choice, pay holders for the NEAR issued? | Rating | When | Read from | |---|---|---| | **Avoid** (overrides Buy) | **Either** the bought NEAR is more than 10% below its highest since 1 November (at once, until back within 10%).
**Or** this window and the one 30 days earlier each fail a test: (1) the three wallets keep under 75% of DefiLlama's revenue (a window missing DefiLlama days is judged on test 2 alone, if every day has a price); (2) the buybacks keep April's $6,270 a day or less, unless fees fell (or lack data) while the private path's stablecoins grew. | the chain; DefiLlama for test 1 | | **Buy** | **All three:**
1. A published rule sends the buybacks to a lock, which we then declare in `daily.LOCKS` (none yet), and the chain shows it cannot send NEAR out: no contract and no keys, or a reviewed code hash (`AUDITED_CODE`, none yet) with no full-access key and every function-call key limited to the lock with a finite allowance. The code checks the declared lock, not how much NEAR is in it. Read on a real chain: `near-thesis-lock-1311.testnet`, a testnet account whose only key we deleted by hand, has 0 keys and no contract and reads as sealed at block 277,435,278 (`scripts/testnet_lock.py`).
2. `treasury.near` opened no new multisig request in the window (its request nonce is the same at both ends and it has no full-access key; a request opened earlier and confirmed inside the window would not be seen).
3. The buybacks cover issuance, or at the yearly check on 1 November have grown +75% a year (+56% once HSP-029 is in force: `HSP029_IN_FORCE` in `scripts/rating.py`, set by hand after the release), in dollars and in NEAR. A yearly pass then holds until the next check, without item 2 being checked again. | the chain; DefiLlama caps what counts | | **Hold** | Everything else, unjudgeable windows included. | | - **Is Hold built in?** Upward, yes, until a lock exists: 54 threshold settings × 7 months (April to October) = 378 ratings, all Hold (378). That is the finding: what holders lack is a rule. Downward, no: Avoid can fire any day (the drain test, added on 12 November; run on April to October at month ends it would not have fired, as the buybacks wallets' NEAR grew every month, `data/coverage_monthly.csv`), and the same rules give Hyperliquid, whose bought tokens sit at a keyless address, Buy on one of the four ways we count its new tokens ([peers](EVIDENCE.md#the-same-rules-on-four-other-protocols)). - **Price.** Until a lock exists, no price gives Buy; a lower price can only push toward Avoid (test 2 fails at $0.655 or less if the buybacks keep this window's NEAR). With a lock, the yearly check counts dollars too. Backtest: [the rating in full](docs/DETAILS.md#the-rating-in-full). Four other protocols: [EVIDENCE.md](EVIDENCE.md#the-same-rules-on-four-other-protocols). ## What would change our mind - **16 November:** PREREG predictions 1–3 are judged; HSP-029, the issuance cut, leaves its timelock at 2026-11-16 16:48 UTC at the latest: until then any one member of House of Stake's Security Council can veto it or end the timelock early, and the timelock's end executes nothing on chain. It cuts issuance only once validators run a release. - **9 December 2026:** if from 9 November the buybacks keep less than half of $46,692 a day, PREREG-3's P6 (0.90) is scored wrong, and claim D with it. - **From 11 December at the earliest:** the two window tests can give Avoid (the windows ending 1–10 November passed and are final). - **Each 1 November, 2027 to 2030:** Buy's growth test, against the [yearly checkpoints](#what-would-change-our-mind) below. - **Any day:** a drain gives Avoid. - **Any day:** a sealed lock, no new treasury request and 100% counted coverage (best month so far 15.3%) gives Buy. - **Any three months running:** coverage of 25% or more ends "small" ([the full list](THESIS.md#what-would-change-our-mind)). - **Predictions** are hashed on mainnet before they are judged ([all, with results](EVIDENCE.md#predictions-anchored-before-they-are-judged)). **Buy's yearly checkpoints.** On each 1 November `scripts/rating.py` compares the 30-day window ending that day with the pace that makes the buybacks cover issuance on 1 January 2031: +75% a year (+56% once HSP-029 is in force) from the window ending 1 November 2026, 9,570 NEAR and $46,692 a day. At or above both floors, with `treasury.near` quiet in that window, the growth test passes until the next check: Buy on every day the lock is sealed, unless Avoid fires. Below either floor it fails for that year, and Buy then needs full coverage. If DefiLlama lacks a day in that window, no check is made and the last result stands. | 1 November | Floor, NEAR a day | Floor, dollars a day | With HSP-029 in force: NEAR · dollars | |---|---|---|---| | 2027 | 16,716 | $81,560 | 14,928 · $72,835 | | 2028 | 29,244 | $142,685 | 23,315 · $113,754 | | 2029 | 51,083 | $249,239 | 36,368 · $177,445 | | 2030 | 89,231 | $435,365 | 56,731 · $276,798 | **0.25** that the buybacks go to such a lock by 2031: our judgment, from three peers and NEAR's proposals so far ([sources](docs/DETAILS.md#how-likely-a-rule-is)). ## Check it with one command Python 3.10+, standard library only, no network, no keys: ``` python scripts/rebuild.py ``` It runs every test, including an independent second implementation, and recomputes every number from the saved raw responses ([each step](AGENTS.md); [against the chain](docs/DETAILS.md#verify-live)). ## How NEAR is used - **Read.** From mainnet by archival RPC: the buybacks wallets, new NEAR issued, signers and keys, `treasury.near`, protocol upgrades. - **Written.** Each prediction file's SHA-256 is anchored in SocialDB (`social.near`) before it is judged. - **Every day.** A GitHub Action re-reads the chain and re-runs the rating; the dashboard reads the drain test live. ## What we can't measure The fees (NEAR's datasets on Dune; only revenue is checked on chain), whether DefiLlama's volume counts one or both legs of a swap (not checked), volume and fees on FAR, why NEAR moved, the NEAR AI credits' dollar value, and peers (from their docs, DefiLlama and CoinGecko). ## Terms - **Buybacks wallets:** `buybacks.multisignature.near` and `1csfundsadmin.sputnik-dao.near`, where NEAR bought with NEAR Intents' fees is kept. - **Front-end fund:** `fefundsadmin.sputnik-dao.near`, holding the fees of NEAR Intents' own front end; with the two above, the **three revenue wallets**. - **Coverage:** NEAR the buybacks wallets keep (inflows less outflows) ÷ new NEAR issued (minted less gas burnt), over the same 30 days. - **FAR:** a private instance of the NEAR protocol, run by a few validators ([docs](https://docs.near-intents.org/integration/market-makers/confidential-intents.md)); NEAR Intents' contract there is `intents.far`. - More, including HSP-027 and HSP-029: [Terms](docs/DETAILS.md#terms). ## Timeline and disclosure - Research: 22–31 October 2026, in private notes outside this repository. Code and data: from the first commit, 2026-10-31 13:51:52 UTC. The [rules](https://nearhackathon.com/HACKATHON.md) (FAQ Q-001) count work from the hackathon's announcement on 19 October: "Earlier commits are baseline". - Submitted state: the tag `hackathon-submission` (FAQ Q-003). After the deadline, 14 November 23:59 UTC, the only commits are the daily workflow's (`.github/workflows/daily.yml`, author `github-actions[bot]`): new data rows in `data/` and the rating they give, as this README promises; no code or text changes. - Sources: NEAR archival RPC, NearBlocks, NEAR's public BigQuery dataset, DefiLlama, SEC filings and NEAR's docs, linked where used. Raw responses are saved byte for byte; those listing people's accounts are kept private, with their SHA-256. - Rules changed after anchoring: the drain test (Avoid on a fall of the bought NEAR) was added on 12 November, after the predictions in `PREREG*.md` were anchored, so it is not pre-registered. Every rule change is in the git history of `scripts/rating.py`. - AI: code and text were written with Claude Code (Anthropic), directed by the author; those commits are marked `Co-Authored-By: Claude`. The demo video's voice is synthetic (ElevenLabs text to speech), read from the script, with the same words as subtitles. - Libraries: none beyond the Python standard library. `docs/sign*.html` load NEAR Wallet Selector 10.1.4 from esm.sh. - This is the author's only submission to NEAR HACKATHON. License: [MIT](LICENSE).