Groups with enough follow-up
36
Weekly groups old enough to reach this week
ON-CHAIN USER RETENTION STUDY
Does NEAR keep its users?
When an account signs its first observed NEAR transaction, does it come back in the weeks that follow? We grouped accounts by the week they first appeared and tracked their activity over time.
THE DATA
Explore the numbers
Change the account filter or choose how many weeks later to check. Every number, chart and row updates together.
Accounts in those groups
4,627,825
Across groups with complete follow-up
Accounts active in selected week
18,989
Signed a transaction in the selected week
Return rate
0.41%
About 4 in every 1,000 accounts
Highest-returning group
—
Among groups with enough follow-up
Early vs recent change
+1.29 pp
95% range includes no change
How many came back from each week?
Each point is one group. Select a point to see its exact count and return rate.
Select a point to inspect a group.
How many accounts were in each group?
Log scale: each step is 10×. This keeps very large and typical weeks visible together.
Each bar represents one week.
How does activity fade over time?
Share of accounts still signing transactions at each follow-up week.
Only groups old enough for each week are included.
All 40 weekly groups
A dash means the group has not yet had enough time to reach the selected follow-up week.
| First seen in week | Accounts in group | Returned | Return rate | Follow-up |
|---|
Retention at key milestones
Each row uses only groups old enough to reach that exact week. Change the account filter above to recalculate.
| Follow-up | Eligible groups | Accounts | Active | Rate |
|---|
These are exact-week snapshots, not continuous activity across the entire period.
Five largest weekly groups
Compare the same five weeks under no filter, the >50 filter and the >20 filter. Week-four returns use the >50 definition.
| First seen | All accounts | After >50 filter | After >20 filter | Week-four active (>50) | Week-four rate |
|---|
ⓘ
A few very large weeks dominate the pooled result.
For the headline filter at week 4, the pooled return rate is 0.41%, the average weekly-group rate is 2.41%, and the median weekly-group rate is 2.31%. The five largest groups contain 88.4% of accounts but only 24.0% of week-4 returns. Their pooled return rate is 0.11%, compared with 2.68% across the other 31 groups. The first three February groups alone contain 75.0% of mature accounts. Across the five groups, the >20 filter removes only 3,355 of 4.091 million accounts (0.082%) compared with no filter. High-frequency filtering does not explain the spikes. One-time automation or campaign activity remains possible, but the aggregate CSV cannot identify the cause.
At week 4, about 4 in every 1,000 accounts in the selected groups signed another transaction.
ASK THE DATA
Find a quick answer
Choose a question. The answer uses the filters and follow-up week selected above.
Choose a question to see an answer based on the current data.
HOW TO READ IT
Three things to know
No research background needed. These are the rules behind every number.
Accounts whose first observed signed transaction fell in the same week. There are 40 groups, each containing many accounts.
The account signed at least one transaction in that exact follow-up week, not merely at some point during the weeks before it. We count accounts, not people. One account is not necessarily one person.
No signed transaction was seen for that account in the 12 weeks before the study window. An older inactive account can therefore look new.
What does “percentage points” mean?
It is the difference between two percentages. If a rate moves from 2% to 3%, that is an increase of 1 percentage point. It is not the same as a 1% increase.
Why compare the first eight and latest eight groups?
We compare average return rates in the earliest eight and latest eight groups that have a complete follow-up. This stops one unusually good or bad week from deciding the trend. The 95% range shows uncertainty around the difference.
What does the 95% range tell us?
If the range crosses zero, the data do not clearly distinguish a rise from no change.
Why remove accounts with many first-week transactions?
Some accounts may be automated or created for short-term activity. We repeat the calculation with different transaction limits to see whether the result changes. These limits are checks, not proof that an account is a bot.
What are the study's limits?
This counts accounts that sign transactions. It does not count people, accounts that only receive tokens, or accounts that act only through relayers. A 12-week look-back cannot distinguish a genuinely new account from an older account dormant for longer. Retention alone does not establish revenue, fees, token value or product-market fit.
INVESTMENT DILIGENCE
Usage is not value capture
Retention is only one part of an investment case. The token must also capture economic value, and the price must be reasonable relative to that value.
CoinGecko showed NEAR around $3.36, a $4.40B market cap and $107.1M TVL. DeFiLlama's chain-only NEAR TVL was $125.08M, while its $276.2M "NEAR Protocol Combined" figure includes assets across multiple chains, including $110.31M on Ethereum. Compare chain-only figures, not the combined total. NEAR Intents separately reported $176.7M TVL across 26 chains. These scopes can overlap and must not be added. NEAR Intents processed $3.63B DEX volume in 30 days, generated $5.83M fees and captured $1.81M revenue. Across the broader NEAR Protocol view, 30-day fees were $5.95M and revenue $1.93M. Captured revenue was about 32.4% of gross fees. DeFiLlama labels $1.81M as holders revenue used for NEAR buybacks since 23 Feb 2026. That is buyback-based value capture, not a burn. A simple 30-day annualization is $21.7M, around 0.50% of market cap, not a guaranteed yield. Trailing-year fees of $54.14M and revenue of $9.64M imply about 81× market-cap/fees and 454× market-cap/revenue. At the same market cap, 100× revenue requires $43.8M a year (4.5× current), 50× requires $87.5M (9.1×), and 25× requires $175M (18.2×). These are arithmetic hurdles, not price targets. The $3.63B volume is not revenue. A recent chain-data analysis estimated about 88,000 NEAR issued per day versus 890 NEAR burned per day after protocol version 87 took effect on 4 Nov. That is roughly 1% of issuance offset by gas burn in the first eight days. The sample is short, but it shows that 100% of eligible gas being burned does not mean NEAR is deflationary. Net issuance remains strongly positive at that observed rate.
The active maximum annual issuance remains 2.5%. HSP-029 passed a vote to ramp it to 1.6% over 24 months, but its veto window runs to 16 November, followed by a 90-day grace period and a required protocol upgrade and validator adoption. The developer gas rebate ended on 4 November under HSP-027, so eligible execution fees are now fully burned rather than split 70/30.
Retention driven by incentives, concentration in a few contracts, product revenue that does not reach NEAR holders, issuance dilution, bridge/security incidents and stronger competing networks. The 16 Oct postmortem says the $3.8M incident came from a deposit-notification bug in the NEAR Intents contract, not an Omni bridge bug. Twenty-three failed notifications created unbacked balances. The team says all funds were returned on 2 Oct and affected users were made whole; the notification feature was removed on 7 Oct after another critical report. The recovery is reported by the team, but the incident exposed contract-accounting and monitoring weaknesses. It is not evidence of a NEAR base-layer consensus failure.
Current verdict: valuation is not yet underwritten. One day of fees or revenue is not enough to estimate sustainable annual cash flow.
CoinGecko market data ↗DeFiLlama combined protocol data ↗NEAR Intents economics ↗NEAR chain TVL ↗NEAR holders revenue ↗Issuance vs burn analysis ↗NEAR Intents postmortem ↗Removed deposit notification feature ↗HSP-029 issuance proposal ↗HSP-027 gas rebate removal ↗nearcore releases ↗NEAR Revenue dashboard ↗Full valuation framework ↗
SOURCES
Where the numbers come from
The dashboard reads the supplied aggregate results. The query and analysis code are public.
Data snapshot and download
Source table: bigquery-public-data.crypto_near_mainnet_us.transactions. Study window: 2 February to 2 November 2026. Burn-in: 12 weeks. Activity means at least one signed transaction in a Monday-start UTC week.
The supplied file contains aggregated cohort counts, not account addresses. This page reproduces calculations from the supplied CSV. The dashboard is a published snapshot, not a live chain-data feed.