The dollar figure is the least interesting number in last week's ETF tape.
US spot Ethereum ETFs printed a net outflow of $319.97 million across the five trading sessions ending October 16. US spot Bitcoin ETFs printed $228.58 million. Ethereum "bled more," by roughly 40 percent, and that sentence traveled everywhere within an hour of the SoSoValue print.
Normalize both by assets under management and the gap stops being 40 percent. It becomes something closer to 900 percent.
That inversion — dollar ranking versus AUM ranking — is the only piece of information in the dataset a reader cannot reconstruct from the headline. Everything else is arithmetic that closes to the cent, which is precisely why the framing deserves an audit rather than a summary.
The arithmetic, before the interpretation
Let me reproduce the tape. BTC: −$65.40M, +$175.62M, −$261.46M, +$5.06M, −$82.39M. Sum: −$228.57M against a reported −$228.58M. ETH: −$85.40M, +$24.84M, −$129.51M, −$60.22M, −$69.68M. Sum: −$319.97M against a reported figure of roughly −$320M.
Both close. The weekly totals are internally consistent, and I will treat them as directional evidence.
Now the cumulative check, which almost nobody ran. Aggregate US spot BTC ETF net inflow stood near $57.8 billion on October 2 and $56.88 billion on October 16 — a drawdown of $920 million. The prior week accounted for $680 million of it. Last week's print accounts for $228.6 million. Combined: $908.6 million, against the $920 million the cumulative series implies. The $11.4 million residual sits inside rounding tolerance on a figure quoted to three significant digits.
So the tape is clean. Which means the problem is not the data. The problem is what the article does with it.
Where the signal actually lives
Here is my working AUM frame, and I want to flag it explicitly as an estimate rather than a source figure — the article supplies no AUM at all, which is its single largest structural omission.
US spot BTC ETFs hold roughly 1.3 to 1.4 million BTC. At the $80,000 anchor the article itself implies, that is $104 to $112 billion. Last week's $228.58 million outflow is therefore 0.20 to 0.22 percent of AUM.
US spot ETH ETFs hold roughly 6 to 8 million ETH. At $2,300, that is $13.8 to $18.4 billion. Last week's $319.97 million is 1.74 to 2.32 percent of AUM, call it 2.0 percent.
The ratio lands between 8x and 11x. Ethereum's relative bleed is an order of magnitude worse than Bitcoin's, and the headline number — 40 percent — understates the spread by a factor of twenty.
Extend to the three-week window. ETH ETFs have now printed three consecutive negative weeks totaling roughly $1 billion. Against a $16 billion AUM midpoint, that is 6.3 percent of the wrapper's asset base gone in fifteen sessions. BTC's two-week drawdown of $920 million against a $108 billion base is 0.85 percent.
That is the number the piece should have led with. It did not, because it ranked in dollars instead of in percentages. When the flow tape speaks, we listen for the discrepancies — and this one is a unit error masquerading as a headline.
The churn ratio, and why it matters more than the total
There is a second derived metric the source never computes and that I now consider mandatory in any flow post-mortem: gross flow divided by net flow.
BTC last week: gross absolute flow across the five sessions was $589.93 million. Net was $228.58 million. Net flow was 39 percent of gross, meaning 61 percent of the week's movement was self-cancelling churn.
ETH last week: gross was $369.65 million. Net was $319.97 million. Net was 87 percent of gross. Only 13 percent cancelled.
Read that pair again. Bitcoin's tape shows a $241 million reversal inside 24 hours (−$65.40M to +$175.62M) and then a $437 million reversal the next day (+$175.62M to −$261.46M). No allocation desk operates that way. Configured capital does not flip $437 million in a single session. That is the signature of basis-trade inventory and authorized-participant arbitrage, not conviction.
Ethereum's tape shows almost no churn. One green day, four red days, and the red days are distributed evenly across the week rather than concentrated. That is what a persistent seller looks like, and it is a materially different signal from Bitcoin's noise.
When code speaks, we listen for the discrepancies. Here the discrepancy is not in the totals. It is in the texture.
The basis trade contaminates the directional read
A large share of spot ETF flow is not directional at all. The structure is: buy the spot ETF, short the CME futures contract, harvest the annualized basis. When the basis compresses — and it compresses every time futures open interest gets crowded — the position unwinds on both legs. The ETF leg redeems. The tape prints an outflow. Nothing about the holder's view of Bitcoin changed.
I built a version of this model in 2024, after the spot approvals, trying to isolate whether institutional accumulation was actually driving price. It was not, at least not through the channel everyone assumed. What the custody data showed was a reduction in exchange-resident supply, not a price impulse. That distinction is the same one being lost here.
The article treats outflow as "bleeding," which is a directional claim smuggled in as a description. It is not supported. A $228 million outflow on a $108 billion base could equally be authorized participants destocking inventory ahead of a creation cycle.
Ethereum's real problem is the container, not the mood
Here is the part I would have led with, and it has nothing to do with last week.
A spot ETH ETF cannot stake. That is not a market condition. It is a product specification. The holder carries the full price exposure of ETH and forfeits the staking yield — call it 3 percent annualized in the current regime — while also paying the management fee. For the low-fee vehicles at 0.15 to 0.25 percent, the annual drag versus self-custodied, staked ETH is roughly 3.2 to 3.3 percent. For ETHE at approximately 2.5 percent, the drag exceeds 5.5 percent per year.
Bitcoin has no native yield. There is nothing to forfeit. The BTC ETF's 0.25 percent fee buys custody, a tax wrapper, and brokerage access, and costs the holder no opportunity yield whatsoever. On pure product economics, the BTC ETF is strictly dominant over spot BTC for any account that cannot self-custody efficiently. The ETH ETF is strictly inferior to spot ETH for any account that can.
That asymmetry is structural and permanent. It explains the persistence of ETH outflows better than sentiment does, and it will keep explaining them until staking is approved inside the wrapper. I flagged this in my 2024 work as the most under-priced variable in the ETH ETF complex. It remains under-priced, because it is a specification defect and not a narrative one.
The data hygiene problem nobody flagged
One more thing, and it is small in size and large in implication.
The article states that October 13 was the week's only green session, at +$24.84 million. It separately reports a Monday net outflow of $85.40 million. Those two claims are incompatible if October 13 is the Monday of that week.
I ran the calendar. October 13 falls on a Monday in 2025. It falls on a Tuesday in any year where October 12 is a Monday — and among years in which both spot BTC and spot ETH ETFs existed, that leaves exactly one candidate: 2026.
So the contradiction resolves, but only by assuming a trading week of Monday October 12 through Friday October 16, and only by assuming a year the article never states. Under that reading the daily sequence is internally coherent: Monday −$85.40M, Tuesday +$24.84M, Wednesday −$129.51M, Thursday −$60.22M, Friday −$69.68M.
I will take the resolution. I will also take the lesson: the article's own date label is unreliable enough that I had to reconstruct the year from a day-of-week mapping. That downgrades daily attribution to medium confidence even though the weekly totals survive intact.
And the sourcing
Seven of eighteen discrete data points carry a SoSoValue attribution. Eleven carry nothing. SoSoValue is a second-order aggregator — generally accurate on US spot ETF daily flows, but subject to T+0 and T+1 revision and not a primary source. Primary would be each issuer's daily NAV and creation-redemption disclosure, or Bloomberg, or Farside. CryptoPotato is a crypto-native mid-tier outlet with no proprietary data pipeline. Composite reliability: medium. Usable for direction. Not usable as the evidentiary base for a $56 billion claim.
I learned in 2017, reverse-engineering a testnet whose original audit had missed three integer overflows, that the question is never whether data exists. It is who produced it and what they were incentivized to leave out. Two risks the piece never mentions: custody concentration, since the underlying sits predominantly with one custodian, and authorized-participant concentration, since a handful of desks clear the creation and redemption flow. Both are the actual single points of failure in the wrapper. Neither appears.
What the outflow does not tell you
Flows follow price far more reliably than price follows flows. Weekly ETF data is a momentum-confirming series, and treating it as a leading indicator is a category error made every Monday.
The drawdown is 1.6 percent of the cumulative stock over two weeks. That is not bleeding. That is a rounding event amplified by a reporting cadence that manufactures narrative out of noise. And a redemption is not necessarily supply hitting the market — coins leaving an ETF wrapper can land in cold storage just as easily as on an exchange order book.
The honest read: Bitcoin's ETF complex is churning and slightly shrinking. Ethereum's is bleeding persistently, for a reason that lives in the product specification rather than the price chart. Those are different diagnoses, and only one of them is fixable by a regulatory decision.
Next week's signal
Watch the normalized ratio, not the dollar print. If ETH's weekly outflow stays above 1 percent of AUM for a fourth consecutive week, the persistence is structural and the cumulative drawdown crosses 8 percent of the wrapper's base. If it mean-reverts below 0.5 percent, last week was basis unwind plus a crowded long getting flushed.
Watch the churn ratio on BTC. If net flow climbs back above 60 percent of gross, real configuration is returning. If it stays near 40 percent, you are reading a market maker's inventory report, not a positioning statement.
And watch the staking docket. Everything else is a weekly print that expires in five sessions.
When the ledger speaks, we listen for the discrepancies. Last week the discrepancies were not in the numbers. They were in which numbers got quoted.