As of this morning, July 22, 2026, DefiLlama's listing for Bedrock uniBTC, a wrapped and restaked Bitcoin product, shows a 24 hour TVL change of minus 37 percent: $307.6 million at the day's first datapoint, $193.2 million by the 16:26 UTC refresh. On a screen that reads as a nine-figure run on a Bitcoin reserve. Our incident monitor flagged it within the hour, the same way it would flag a real redemption run.
It is not a run. We recomputed the actual reserve token on chain before writing a word, which is the entire point of this piece. uniBTC's supply on Ethereum moved by 0.43 of a token over the past seven days, about $28,000 on a $195 million float, and the token trades within 0.6 percent of Bitcoin itself. Nothing was redeemed. The crash exists in the data pipeline, not in the protocol. It is also the third time in eight days that a nine-figure crash appeared on a dashboard and not on a chain, which is why this is worth a full walk rather than a shrug. Earlier this week we published the mirror-image case: USD.AI's very real 24 percent supply contraction, which the same kind of recompute confirmed as one large redemption honored at par. The method does not care which way the answer comes out. That is what makes it a method.
What the dashboard showed, three times
July 14: DefiLlama's stablecoin feed showed BlackRock's tokenized treasury fund BUIDL contracting 22 percent in 24 hours, from $3.69 billion to $2.88 billion, which would have been an $810 million single-day redemption run on the largest tokenized fund in existence. July 15: the Lombard LBTC listing dropped 57.5 percent in a day, from $591 million to $251 million. This morning, July 22: Bedrock uniBTC, minus 37 percent, $307.6 million to $193.2 million. Three different issuers, three different asset classes, eight days, roughly $1.3 billion of combined phantom outflow.
Each of those numbers was live on a public dashboard that risk desks, allocators, listing committees, and monitoring bots read programmatically. Ours included: our own incident monitor consumes the same DefiLlama feeds, and it fired a candidate alert on all three. The difference between a candidate and a headline is the recompute that follows.
What the chain shows for Bedrock today
The reserve token behind the listing is uniBTC, an 8 decimal ERC-20 at 0x004E9C3EF86bc1ca1f0bB5C7662861Ee93350568 on Ethereum. A raw totalSupply call at block 25,589,934 this morning returns 2,984.999 uniBTC. The same call against archive state seven days earlier returns 2,985.426. That is a net change of minus 0.43 uniBTC, minus 0.014 percent, over the week in which the dashboard says a third of the product left. We ran the read against three independent public RPC endpoints and got the identical value from each. Over the 24 hours in which the listing lost $114 million, the supply change was three ten-thousandths of one token, and it was a small mint, not a burn. On price, uniBTC trades at 0.994 of Bitcoin, a discount comparable to WBTC's ordinary spread. A real $114 million forced exit leaves a mark on at least one of those two surfaces. There is none on either.
So where did the dashboard number come from? The listing's own per-chain breakdown answers it. Of the $114.4 million single-refresh drop, 98.6 percent is one token family, the Locked FBTC positions the adapter tracks, disappearing from exactly three chains at once: $85.5 million of it on Ethereum, $26.6 million on BOB, $0.7 million on Mantle. The very same token on BNB Chain kept reporting an unchanged 246.99 tokens through both snapshots, and every other balance in the listing is frozen to the second decimal across days: 1,716.63 native BTC, 990.88 M-BTC. Reserves do not exit like that. Balances that vanish from three chains in a single tick while surviving on a fourth are a read that failed, not value that moved. The same listing's BOB component had already printed zero twice this week, on July 16 and July 21, and recovered both times.
Two honest possibilities remain for the missing component, and we flag both because as of publication the reading has not yet corrected. Either the adapter's balance calls failed on three chains, in which case the figure should snap back within a refresh or two the way Lombard's did, or DefiLlama deliberately removed the Locked FBTC positions from the listing, for example to avoid double counting collateral tracked elsewhere, in which case the listed figure stays lower permanently. The second would be a methodology change, not an outflow. In neither case did anyone redeem $114 million from Bedrock yesterday, and that is the claim a reader of the minus 37 percent headline would otherwise walk away with.
The other two, briefly
Lombard, July 15, was the clean snap-back case. The listing fell from $591 million to $251 million and was fully recovered to $587 million by July 17, while the underlying BTC token count ran continuous through the whole episode. Anyone who de-risked, delisted, or wrote a headline off the July 15 print was reacting to a component that had dropped out of an aggregation for two days. Lombard's listing is flat this week, down about 1 percent, which also tells you today's Bedrock reading is not some shared methodology change sweeping every Bitcoin product at once.
BUIDL, July 14, was the more interesting shape, because there the new, lower number was the correct one. The entire $810 million "contraction" was one chain, Aptos, going from a reported $821.9 million to $16.1 million, while the fund's seven other chains sat flat. The on-chain supply of BUIDL's Aptos deployment reads 16,121,612 units, matching the new figure exactly. There was no burn: the only Aptos activity for weeks had been a daily yield deposit of about $1,500, and the last large movement was a $162.7 million withdrawal eighteen days earlier. The feed had been carrying a stale $806 million phantom on that chain for weeks and finally corrected it. Which means the dashboard was materially wrong before the crash, in the quiet direction, the one nobody alerts on. A $1,500 daily accrual is a plausible treasury yield on $16 million and an absurd one on $822 million, so the overstatement was even visible from inside the feed itself, if you did the arithmetic.
TVL is an aggregation, not a proof
None of this is a knock on the protocols. Bedrock's supply is flat and its token is at par; Lombard's reserves never moved; BlackRock's fund did nothing at all. And it is not a dunk on DefiLlama either, which maintains adapters across hundreds of chains for free, publishes its methodology, and gets the aggregate overwhelmingly right. The point is structural. A TVL figure is the output of a pipeline: RPC calls, third-party adapters, price feeds, deduplication rules, all of it revised without an audit trail a holder can replay. It is a directionally useful estimate published as if it were a measurement. A proof of reserves is the opposite object: a specific claim, about specific addresses, that a third party can recompute and either reproduce or refute. The failure mode of confusing the two runs in both directions, and this week produced both: three phantom crashes that made healthy reserves look like runs, and in BUIDL's case a stale overstatement that made the reserve look $806 million larger than it was until the correction landed. If your risk process, your listing decision, or your redemption finger is wired to the first kind of number without a recompute step, weeks like this one are what it will trade on.
The five minute check
The recompute that settled all three cases is the same one every time, and it needs no infrastructure. Find the token's official contract address from the issuer's documentation, not from a search result. Call totalSupply and decimals on it through two or three independent public RPC endpoints, so no single provider can lie to you, and compare against the figure from a day or a week ago. Glance at the market price against the peg or the underlying. If supply is continuous and price is at par, the crash you are looking at lives in a pipeline, and if supply really fell or price really broke, you now know that from the chain rather than from a chart. We keep free hosted versions of exactly this at /verify-anything for arbitrary claims and /verify-vault for ERC-4626 vaults. The limit cuts the other way too, and it should be said plainly: a flat supply read proves the crash was phantom, but it does not prove the reserve behind the tokens is sound. That takes the reserve side, which is what a proof of reserves regime, with all its own limits, exists to cover.
Where Kerne stands
We publish the object we are arguing for, so the interest is declared. kUSD's on-chain collateral on Base is covered by an hourly proof of reserves signed with an EIP-191 key that any holder re-derives against raw chain state at /verify and /api/por, with no attestor in the path, which is precisely the property the dashboards above cannot offer: when our number is wrong, you can catch us, byte for byte. The boundaries are listed in the same breath. The hedge leg runs at pilot scale against a small disclosed founder-custodied float and is self-reported rather than independently re-derivable; our first external audit delivered its initial report on July 20 with remediation underway and the final report pending; and our own advertised-versus-realized row sits first, and currently worst, on the honesty index. We consume DefiLlama like everyone else, and this morning's alert is the standing reminder of why the recompute step is not optional for us either. Nothing here is an offer of any token.
If holders trade on your numbers
The check we ran on Bedrock this morning is the same one we sell as a fixed-scope service: a Disclosure Integrity Audit, $499 flat, 72 hours, a signed findings summary you can publish, covering whether your public claims match your chain before someone else checks for you. The worked example is free: our scrvUSD projected-versus-realized read.
Sources and reproduction, all as of July 22, 2026, 17:30 UTC: the Bedrock uniBTC series and per-chain, per-token breakdown from api.llama.fi/protocol/bedrock-unibtc (drop between the 00:00 and 16:26 UTC datapoints; change_1d minus 37.09 percent on api.llama.fi/protocols); Lombard LBTC series from api.llama.fi/protocol/lombard-lbtc (July 14 through 17); the BUIDL episode from the DefiLlama stablecoin feed of July 14 and an Aptos supply read of the fund's fungible asset, 16,121,612 units. uniBTC supply reads: eth_call totalSupply on 0x004E9C3EF86bc1ca1f0bB5C7662861Ee93350568 at block 25,589,934 and at the block seven days prior, via publicnode, dRPC, and 1RPC, all returning 2,984.999 and 2,985.426 respectively; decimals 8; price ratio from CoinGecko simple price, uniBTC at 0.994 of BTC against WBTC's 0.999. DefiLlama readings are point-in-time and routinely revised, which is part of the point; recompute before relying on any figure here, including ours.
Verify it yourself
Run the same check on any reserve, or have it run for you.
Paste any issuer's signed attestation into the free verify tool and recover the signer, rehash the figures, and check freshness in your own browser. For a machine-signed, point-in-time read of an address you name, delivered on the page in about two minutes, the instant self-serve read is $29; a human-reviewed read is $149. A teardown like this one, commissioned on any target you name, is $499. An independent read of a counterparty you hold or allocate to is $2,500. Attestation tooling, not an audit, and not a solvency opinion.