NEAR, Cross-Examined

The headline compares 30-day averages with a year ago. Other windows, same data:

• Daily active accounts, 30-day mean against 30, 90, 180, 365 and 730 days earlier: -22.9%, -32.2%, -53.7%, -77.9%, -94.0%.
• App fees: last 30 days against the 30 before -28.3%; last 90 against the 90 before +111.5%; last 365 against the 365 before +392.3%. The most recent month fell.
• The step in transactions on 2026-04-02: its cause is unknown, and no source we found names it. Transactions -74.9%, new accounts -84.1%, active contracts -92.0%, active accounts -3.9% (14 days after against 14 days before). That fits one automated actor stopping, but it is a guess.
• Definitions: app fees are the fees DefiLlama tracks for apps on NEAR, which may not cover every app. The app-fees-to-gas-fees multiple compares two different flows of money, so it shows scale only.

About 87% of the dollar growth is the price rise. In 2026-04 transactions fell 73% in one month while active accounts fell 10%; that month is 67% of the year's fall.

What we can and cannot say: this data cannot name the cause. The numbers are consistent with fewer small, everyday transactions and a larger average transfer: NEAR moved per transaction rose 7.3x (`paradox.near_per_txn_x`). No transaction-size distribution is measured here, so we do not know which kinds of transactions stopped, or whether larger transfers took their place.

So the paradox is not a contradiction. The data shows fewer transactions and more NEAR moved per transaction. That is consistent with a chain that lost many small transactions and kept larger ones, but it does not prove it.

On the due date, the thesis (NEAR's price is ahead of what its one paying product returns to holders) is WRONG if at least two of T1, T2, T3 are HIT. It is STRONGER if T4 or T5 is HIT. Anything else leaves it standing. T6 is a data check: if it is a MISS, nothing is scored until the two providers are reconciled.

gpt-oss-120b (OpenAI, open-weight): The data show a stark collapse in active accounts and transaction volume on NEAR, while the modest supply growth and high market‑cap multiples indicate that valuation is far outpacing genuine on‑chain usage; the ecosystem is heavily reliant on a single product (Intents) and app fees dwarf gas fees, suggesting limited sustainable demand.
Would change its mind: If 30‑day average daily active accounts and transaction counts both rose by more than 50% year‑over‑year while gas‑fee revenue and app‑fee revenue remained low, indicating a genuine resurgence of core usage.

DeepSeek-V4.1-Flash (DeepSeek): The evidence supports a bearish long-term case for NEAR as a general-purpose chain: core active accounts, transactions, new accounts, and active contracts have collapsed, gas fees are tiny versus issuance, and market cap is extremely high relative to gas fees. The main offset is NEAR Intents, where app fees and retained revenue are growing and buybacks have started, but that activity is concentrated, stablecoin liquidity is shrinking, and buybacks cover only a minority of issuance.
Would change its mind: A sustained multi-month increase in active accounts and transactions alongside broadening app-fee growth beyond NEAR Intents and stablecoin liquidity growth.

GLM-5.2 (Zhipu): NEAR's core on-chain usage has collapsed—active accounts and transactions are down 78–85% year-over-year and 97% from peak—yet the surviving activity is far more economically valuable per account, driven almost entirely by NEAR Intents, which generates growing fees (up 51% YoY), revenue (up 334% YoY), and token buybacks. The bull case rests on Intents becoming a durable cross-chain revenue engine, but the chain is a one-product economy (87% of app fees from Intents, only four protocols over $100/day), buybacks cover only a fifth of new issuance, and the market cap is hundreds of times annualized Intents revenue, making the setup fragile and richly valued.
Would change its mind: A sustained six-month recovery in monthly active accounts back above 200,000 (monthly.2025-11.active level) combined with apps.intents_share_30d falling below 0.50, indicating diversified organic adoption rather than single-product dependence.

Qwen3-30B (Alibaba): NEAR's ecosystem is undergoing a structural shift: user activity has declined sharply, but economic value per user has increased, driven by NEAR Intents and app fees. The chain is becoming more efficient and valuable per transaction, with growing app-level revenue and buybacks, though it remains highly dependent on a single product and faces challenges in sustaining growth amid low on-chain usage and modest supply inflation.
Would change its mind: If daily active accounts and transaction volume show sustained growth over the next 6 months, while app fees and Intents revenue continue to rise, it would indicate a recovery in organic user adoption.

• Nansen's Q3 2025 NEAR report found one app, HOT Wallet, had more than 100 times the active addresses of the next entity and 56.1% of transactions. That concentration is why account counts swing so hard: Nansen, Q3 2025.
• NEAR governance passed HSP-027 in 2026 to stop rebating part of gas fees to contract owners and burn them instead, so fees now offset issuance a little more than before: The Defiant. Over 365 days gas fees equal 1.0% of new issuance; the NEAR Intents buybacks are the larger offset, 20%.
• Price and USD volume move together, so transfer value is also shown in NEAR tokens (claim C9).

Every known weak point, with its source, is in WEAKNESSES.md. The short version: account and transaction counts come from one provider; NEAR Intents swap counts are sampled one Sunday a week; staking is not measured.

git clone the repository, then make thesis. It downloads the data (NEAR Intents needs a free partner key in INTENTS_JWT), recomputes every metric (out/metrics.json), re-runs the panel (out/panel.json, needs an OpenAI-compatible API key) and rebuilds this page. make verify needs no key and no network: from a fresh clone it checks the derived numbers against their recorded hashes, rebuilds the BigQuery numbers, checks each judge's citations and the documents, and scores the tripwires; with the source files downloaded it rebuilds every number. make verify-online downloads every source again and rebuilds all the numbers. The source files are not shipped because of their providers' terms; cd data && sha256sum -c --ignore-missing SHA256SUMS checks the ones you have.

Data: NearBlocks, DefiLlama, Intents Explorer, Google BigQuery public dataset (NEAR). Prices: Powered by CoinGecko. The raw files belong to their providers and are not shipped; their hashes are.

Built at NEAR HACKATHON by 3Dogs Nexus (3dogs.ai). Judges: gpt-oss-120b (OpenAI, open-weight), DeepSeek-V4.1-Flash (DeepSeek), GLM-5.2 (Zhipu), Qwen3-30B (Alibaba). Built on 3Dogs wisdom of crowds, three reviewers for speed, five when one stands alone. This entry uses a 4-judge panel.