Coin Metrics has rebuilt Ethereum’s historic Commonplace Circulation Metrics, elevating a timing drawback for assessments that deal with change outflows as a buying and selling sign.
The crypto knowledge supplier’s Oct. 1 discover says it recomputed Ethereum Commonplace Circulation Metrics from the community’s first block utilizing its latest data as a part of its Ethereum Level-in-Time launch. The affected scope is all ETH Circulation Metrics at every day and hourly frequencies, with corrected historical past accessible for backfilling.
That creates a sensible distinction for funding analysis. A chart downloaded in the present day can describe previous flows utilizing data acquired later. A backtest, which replays a buying and selling rule by way of historic knowledge, wants the data accessible when every choice would have occurred. These are completely different data units, even when their observations carry the identical dates.
The discover provides no revision quantities or ETH technique comparability. The quick consequence is a must determine knowledge classic, that means the model of the info used within the check; any impact on returns nonetheless requires measurement.
Why the identical historic date can inform a distinct story
Coin Metrics’ stream methodology makes the excellence concrete. Commonplace metrics use all addresses at present recognized to belong to an change or different tracked entity, with every deal with’s historical past beginning at its first nonzero stability. Previous values will be restated when further entity addresses are recognized.
Its Level-in-Time, or PIT, collection as an alternative makes use of addresses recognized to belong to the entity throughout the historic interval. An deal with contributes from its discovery date, and later discoveries don’t rewrite earlier PIT intervals. The supplier paperwork every day and hourly PIT counterparts to Commonplace exchange-flow metrics.
The underlying challenge is attribution. A switch will be assigned to an change retrospectively as soon as the supplier identifies the pockets. That fuller reconstruction could also be helpful for analyzing previous provide actions with in the present day’s deal with protection. Establishing what a dealer might have acknowledged requires the deal with data and values accessible at that earlier second.
Coin Metrics had outlined the recomputation on Sept. 28 to take care of that distinction, anticipating ETH completion on Sept. 30. Its completion discover was posted Oct. 1 at 17:04 UTC; discover timing alone doesn’t date each affected worth’s availability.
Two comparisons should additionally keep separate. Commonplace versus PIT assessments completely different address-knowledge guidelines. Retained Commonplace historical past from earlier than and after the rebuild is the comparability wanted to measure this specific revision. PIT is a definite attribution methodology. A duplicate of pre-rebuild Commonplace values preserves a selected model of the Commonplace product.
CryptoQuant’s ETH Alternate Flows documentation explicitly warns that the endpoint doesn’t assist PIT accuracy. It says historic values could change as change wallets are found, added and validated by way of periodic clustering updates.
CryptoQuant schedules computerized updates for Tuesday at 00:00 UTC every week and says values can change barely, particularly latest observations. Every supplier’s revisions require their very own measurements and replace information.
For an analyst, retaining an outdated question date is subsequently inadequate if the historic values are fetched once more from a mutable endpoint. The dates of the observations could stay the identical whereas the data used to assemble them adjustments.
The interpretation of an outflow additionally wants restraint. A withdrawal measures motion relative to attributed change wallets. A declare about shopping for or worthwhile buying and selling requires further proof.
Glassnode’s BTC illustration isolates data-vintage danger
Glassnode provided an illustration of the issue in a March 13, 2026, hypothetical backtest. It used Binance’s BTC change stability to enter the market when a five-day transferring common fell under a 14-day common and exit when the shorter common rose above the longer one.
The check coated Jan. 1, 2024, by way of March 9, 2026, beginning with $1,000 and charging 0.1% per commerce. Glassnode stated it repeated the check utilizing PIT balances whereas holding the sign logic, parameters, dates and costs unchanged. The supplier reported worse efficiency with PIT knowledge than with revised balances.
The helpful comparability is that the rule stayed mounted whereas the info variant modified. A historic stability sample reconstructed with later data can set off completely different choices from a sample constructed from contemporaneous data.
Glassnode provided this BTC stability outcome, and the check stays unreplicated on this evaluation. Its relevance to ETH is the measurement strategy: maintain the rule mounted and examine the info vintages. ETH sign and return results require their very own experiment.
The supply clock provides an extra constraint. Glassnode’s PIT documentation provides two limits to the shorthand promise of replaying the previous.
First, PIT historical past exists solely from the date monitoring started for every metric. Earlier than July 2025, protection was restricted to BTC, ETH and chosen tokens and metrics; monitoring expanded throughout all platform metrics from July 2025. A metric added then doesn’t purchase earlier PIT observations merely as a result of common historic knowledge exists.
Second, the timestamp connected to an commentary shouldn’t be essentially when a dealer might retrieve it. Glassnode says it has recorded related computed_at timestamps since September 2024, omitting the sector when unavailable, and that API publication follows computation with a delay.
An unchanged historic worth addresses later revision. Replaying a buying and selling choice additionally requires inserting the enter after its precise publication. A check that acts earlier than the enter might be accessed nonetheless makes use of data from the long run.
For Coin Metrics’ ETH collection, which means documenting every metric’s first monitoring date and historic buyer availability. Glassnode’s protection dates and publication disclosures apply to its personal merchandise.
The proof wanted to measure an ETH buying and selling impact
Measuring this rebuild requires paired observations from the identical supplier and metric, with matching change protection, intervals and dates. For the revision query, which means retained pre-rebuild Commonplace values alongside the post-rebuild Commonplace historical past. For the buying and selling query, it additionally means an data set demonstrably accessible at every choice time.
The rule should stay mounted throughout the comparability: the identical entry and exit circumstances, parameters and analysis window. Availability cutoffs and execution timing belong within the check, alongside buying and selling prices. In any other case, altering the technique whereas altering the info would depart the supply of any efficiency distinction unclear.
The comparability ought to then distinguish modified enter values from modified alerts, modified trades and adjusted returns. A revision can matter to the dataset with out altering a selected rule’s choices.
The decisive follow-up is a paired ETH dataset and a fixed-rule replay that separates knowledge adjustments from buying and selling adjustments. Revised historical past can describe provide with in the present day’s deal with data. A declare that outflows provided a usable buying and selling edge requires reproducible inputs, publication timing and buying and selling choices.
