Coin Metrics has rebuilt Ethereum’s historical Standard Flow Metrics, creating timing complications for backtests that utilize exchange outflows as a trading signal.
As outlined in the crypto data provider’s Oct. 1 notice, Ethereum Standard Flow Metrics were recomputed from the network’s genesis block using its most current information as part of the Ethereum Point-in-Time release. This update impacts all ETH Flow Metrics across daily and hourly frequencies, with corrected history available for backfilling.
This introduces a practical distinction for investment research. A chart retrieved today illustrates past flows using insights obtained at a later date. Conversely, a backtest simulating a trading strategy across history requires the exact data accessible at the moment each decision would have taken place. These information sets differ, even when their observations share identical dates.
The notice omits specific revision quantities and any comparison of ETH strategies. The immediate takeaway is the necessity of tracking data vintage—the specific version of the data utilized in a test—while any subsequent impact on returns must still be measured.
Why the same historical date can tell a different story
Coin Metrics’ flow methodology clarifies this distinction. Standard metrics aggregate all addresses currently attributed to an exchange or tracked entity, with each address’s history calculated from its first nonzero balance. Consequently, past values can be updated whenever additional entity addresses are discovered.
In contrast, its Point-in-Time (PIT) series relies exclusively on addresses recognized as belonging to the entity during that specific historical timeframe. An address contributes data starting only from its discovery date, meaning subsequent discoveries do not alter earlier PIT intervals. The provider offers daily and hourly PIT alternatives to standard exchange-flow metrics.
Attribution is the core driver of this issue. Transfers can be retrospectively linked to an exchange once the provider identifies the relevant wallet. While this comprehensive reconstruction helps analyze past supply movements using current address coverage, determining what a trader could have realistically observed requires the address data and values accessible at that prior point in time.
Coin Metrics initially announced the planned recomputation on Sept. 28 to preserve this distinction, anticipating ETH completion by Sept. 30. The final completion notice was published on Oct. 1 at 17:04 UTC, though notice timing alone does not establish the availability of every impacted value.
Two separate comparisons must also be maintained. Standard versus PIT evaluates different address-knowledge rules, whereas comparing retained Standard history from before and after the rebuild measures this particular revision. Because PIT represents a distinct attribution method, preserving a copy of pre-rebuild Standard values is necessary to retain that specific version of the product.
CryptoQuant explicitly notes in its ETH Exchange Flows documentation that its endpoint lacks PIT accuracy. It cautions that historical metrics can shift as exchange wallets are uncovered, incorporated, and validated via routine clustering updates.
CryptoQuant schedules automated updates for Tuesdays at 00:00 UTC weekly, noting that figures can experience minor shifts, particularly for recent observations. Each provider’s revisions necessitate individual measurement and update tracking.
For analysts, simply retaining an old query date proves insufficient if historical metrics are pulled again from a mutable endpoint. Observation dates may remain static while the underlying information used to generate them changes.
Furthermore, interpreting outflows requires caution. Withdrawals simply track movement relative to attributed exchange wallets, meaning claims regarding buying behavior or profitable trading demand additional evidence.
Glassnode’s BTC illustration isolates data-vintage risk
Glassnode highlighted this challenge using a March 13, 2026, hypothetical backtest. The test utilized Binance’s BTC exchange balance to trigger market entry when a five-day moving average fell beneath a 14-day average, and exit when the shorter average moved above the longer one.
Covering the period from Jan. 1, 2024, through March 9, 2026, the backtest initiated with $1,000 and incorporated a 0.1% fee per trade. Glassnode reran the test utilizing PIT balances while keeping signal logic, parameters, dates, and fees constant, ultimately reporting weaker performance with PIT data than with revised balances.
This comparison demonstrates that keeping rules constant while altering data variants changes outcomes. Historical balance patterns reconstructed using future knowledge can trigger entirely different decisions than patterns built strictly from contemporaneous data.
While Glassnode provided this BTC balance outcome—and the test itself remains unreplicated in this analysis—its relevance to ETH lies in the measurement approach: keeping rules fixed while comparing data vintages. Evaluating ETH signals and return impacts requires a dedicated experiment.
The element of publication timing introduces an additional constraint. Glassnode’s PIT documentation outlines two specific limitations regarding the concept of replaying the past.
First, PIT history is only available starting from the date tracking commenced for a given metric. Prior to July 2025, coverage was restricted to BTC, ETH, and specific tokens and metrics, before expanding across all platform metrics in July 2025. Metrics added at that time do not automatically gain earlier PIT observations simply because standard historical data exists.
Second, the timestamp assigned to an observation does not necessarily reflect when a trader could have accessed it. Glassnode states it has logged relevant computed_at timestamps since September 2024, omitting the field when unavailable, and notes that API publication follows computation with a delay.
While an unchanged historical value accounts for subsequent revisions, replaying a trading decision also requires positioning the input after its actual publication time. A test that acts before inputs were accessible relies on future information.
For Coin Metrics’ ETH series, this necessitates tracking each metric’s initial monitoring date and historical customer availability. Similarly, Glassnode’s coverage timelines and publication disclosures apply exclusively to its own offerings.
The evidence needed to measure an ETH trading effect
Assessing this rebuild demands paired observations from the identical provider and metric, matching exchange coverage, intervals, and dates. For revision analysis, this involves comparing retained pre-rebuild Standard values alongside post-rebuild Standard history. For trading evaluations, it also requires an information set verifiably accessible at each decision point.
Rules must remain static across comparisons, maintaining identical entry and exit criteria, parameters, and evaluation windows. Availability cutoffs, execution timing, and trading costs must also be factored into tests. Altering strategies simultaneously with data changes leaves the source of performance variances unclear.
Comparisons should clearly differentiate between modified input values, altered signals, changed trades, and resulting returns. Revisions can impact a dataset without necessarily altering the decisions dictated by a specific rule.
Ultimately, a definitive follow-up requires a paired ETH dataset and a fixed-rule replay that separates data modifications from trading adjustments. While revised history can characterize supply using modern address intelligence, claiming that outflows provided a usable trading edge demands reproducible inputs, publication timelines, and documented trading decisions.
Frequently Asked Questions
Why do Ethereum exchange outflow charts change over time?
Charts can change when data providers like Coin Metrics identify additional wallets belonging to an exchange and update their historical records retroactively from the network’s first block.
What is the difference between Standard Flow Metrics and Point-in-Time (PIT) metrics?
Standard metrics use all currently known addresses for an entity and can be rewritten when new addresses are found. PIT metrics use only the addresses known to the entity during that specific historical interval and do not rewrite past data when new discoveries are made.
How do historical data revisions affect trading backtests?
Backtests rely on the information available at the exact time a decision would have been made. If a backtest uses revised data that includes future knowledge or undiscovered wallets, the test results may not accurately reflect what a trader could have actually achieved.
Are all crypto data providers consistent with Point-in-Time accuracy?
No. For example, CryptoQuant explicitly notes that its endpoint does not support PIT accuracy and that historical values can change as exchange wallets are periodically discovered and updated.
