Coin Metrics has rebuilt its historical Ethereum Standard Flow Metrics from the network’s first block, a change that could alter how analysts interpret past exchange outflows and test trading strategies. The update does not establish that any strategy’s returns changed, but it highlights the importance of using the correct historical data version and publication timing.
According to CryptoSlate, Coin Metrics announced on Oct. 1 that it had recomputed Ethereum’s Standard Flow Metrics using its latest information as part of an Ethereum Point-in-Time, or PIT, release. The affected data covers ETH flow metrics at both daily and hourly frequencies, and the revised history is available for backfilling.
Why historical exchange flows can change
The issue concerns how blockchain analytics providers identify wallets belonging to exchanges and other entities. Standard metrics use addresses currently known to belong to an entity and apply each address’s history from the first time it held a nonzero balance. If a provider later identifies another wallet as belonging to an exchange, past transactions involving that address can be incorporated into the historical series.
That means a chart downloaded today may describe an earlier date using information that was not available at the time. The historical date itself remains unchanged, but the information used to calculate the value may have been updated.
Coin Metrics’ PIT methodology applies a different rule. It uses the addresses known to belong to an entity during each historical period. An address begins contributing from the date it was discovered, while later wallet discoveries do not rewrite earlier PIT intervals.
This distinction matters for backtesting. A backtest replays a trading rule over historical data, so the relevant question is not only what the provider now believes happened, but what information would have been available when the hypothetical decision was made.
Coin Metrics first outlined the recomputation on Sept. 28 and expected the Ethereum work to be completed by Sept. 30. Its completion notice was posted on Oct. 1 at 17:04 UTC. The notice date alone, however, does not establish when every individual historical value became available.
Outflows are not automatically trading signals
An exchange outflow records movement relative to wallets attributed to an exchange. It does not, by itself, demonstrate that investors bought an asset, that prices will rise, or that a trading strategy would have been profitable. Those conclusions require additional evidence and a defined set of trading rules.
CryptoQuant’s documentation for ETH Exchange Flows explicitly warns that its endpoint does not provide PIT accuracy. Historical values can change as exchange wallets are discovered, added and validated through periodic clustering updates. CryptoQuant says its automatic updates are scheduled for Tuesdays at 00:00 UTC and that values can change slightly, particularly for recent observations.
For researchers, saving only the original query dates is therefore not enough if the same mutable endpoint is queried again later. The dates of the observations may be identical even though the underlying wallet attribution and calculated values have changed.
A previous example from Glassnode illustrates how data revisions can affect a backtest, although it involved Bitcoin rather than Ethereum. In a hypothetical test covering Jan. 1, 2024, to March 9, 2026, Glassnode used Binance’s BTC exchange balance to buy when a five-day moving average fell below a 14-day average and sell when the shorter average moved back above the longer one. The test began with $1,000 and included a 0.1% fee per trade.
Glassnode then repeated the same test with PIT balances, keeping the rules, dates, parameters and fees unchanged. The provider reported weaker performance using PIT data than using revised balances. The result has not been independently replicated in the material available here, and it does not prove that Ethereum will show the same outcome. Its relevance is methodological: holding the trading rule constant allows researchers to isolate the effect of the data version.
What an Ethereum test would need
A credible assessment of Coin Metrics’ Ethereum rebuild would require paired datasets from the same provider and metric. Analysts would need retained pre-rebuild Standard values alongside the post-rebuild Standard series, with matching exchange coverage, time intervals and observation dates.
To assess trading implications, the study would also need to account for when the information could actually be retrieved. A historical value may be unchanged after publication but still be unusable in a test if the strategy is assumed to act before that value was available. Publication delays and relevant availability timestamps therefore form part of the information set.
The trading rule, parameters, evaluation period and transaction costs should remain fixed between the two data versions. The analysis should then separate four possible outcomes: changed input values, changed signals, changed trades and changed returns. A revised dataset may contain different historical numbers without altering the decisions of a particular strategy.
PIT coverage also has its own limits. Glassnode’s documentation says PIT history begins only when tracking for a metric began. It notes that before July 2025 coverage was limited to Bitcoin, Ethereum and selected tokens and metrics, while tracking expanded across all platform metrics from July 2025. Its documentation also says that computed_at timestamps have been recorded since September 2024 when available, and that API publication follows computation with a delay. These details apply to Glassnode’s products, not automatically to Coin Metrics’ ETH series.
Current market picture
Moneyiar market data at 09:49 on Oct. 5, 2026, showed Ethereum at $2,716.6635, up 0.44% over 24 hours and 2.87% over seven days. Shiba Inu was $0.0000059486, up 4.17% in 24 hours and 6.23% over seven days. Uniswap stood at $9.0195, down 0.20% over 24 hours but up 1.73% over seven days. Pepe was $0.0000044984, up 4.81% over 24 hours and 8.32% over seven days.
The central question raised by the rebuild is therefore one of measurement rather than a confirmed market effect. Revised Standard history may offer a fuller picture of exchange-related supply movements using today’s wallet knowledge, while a trading backtest must preserve the information available at each decision point. Any claim that Ethereum outflows produced a repeatable trading advantage still requires a fixed-rule comparison using documented data vintages and publication times.
