Everything here is computed from DefiLlama's own public API on 2026-10-26. The scripts are in the public repo. This page exists mostly to stop you from over-reading the headline.
| Term | Exact meaning |
|---|---|
| fee listing | A protocol that appears in api.llama.fi/overview/fees at all. 3,069 of them. |
| BROKEN | Non-zero fees on at least one day in the last 12 months, and exactly 0 on each of the last 7 days. 411 protocols. |
| MISSING | At least $300,000 of Base TVL, and absent from both the fee overview and the DEX-volume overview. 80 protocols out of 908 Base listings. Bridges and CEXs excluded — their "fees" are not a product surface. |
| lost / day | Mean daily fees over the 30 days before the feed stopped. Not extrapolated, not modelled. |
| lost to date | That mean × days dark. An upper bound on "what the leaderboard never counted", not money anyone owed. |
Ten of the ten largest and most recent breaks were not adapter bugs.
We ran each adapter through its own harness and then hit its data source directly. In every case the source genuinely reported zero:
FeePaid event in two days at the contract, $266M TVL, zero fees collected. Nothing to fix.pricePerShare unchanged for 96 hours. Nothing to fix.Mass-patching that list would have produced false claims against $266M protocols. So the audit ships two numbers: $3,333,964/day is the aggregate measured hole across all 411 rows, and $23,735/day across 18 protocols is the subset that broke recently on protocols that still hold TVL. Only the second one is a work queue. The first one is a headline, and we label it as one.
The biggest dollar figures in the list belong to protocols that are probably dead. That is not a flaw in the measurement, it is what the data says — and it is why the free number and the work queue are different numbers.
| Protocol | Dark | Lost / day | TVL at last read | What it probably is |
|---|---|---|---|---|
| Avalon USDa | 352d | $559,748 | $3 | wound down — not a repair |
| Lava | 303d | $459,262 | $0 | wound down — not a repair |
| Pyth Pro | 47d | $291,779 | $0 | unknown — check the source |
| USD AI | 7d | $258,434 | $266.2M | verified: source genuinely zero, do not patch |
| Kaito Capital Launchpad | 272d | $154,803 | $0 | wound down — not a repair |
The ranking that matters for work is the second one. The 18 recent breaks on protocols that still hold TVL account for $23,735/day between them, and every one of those rows ships with a reproduction command so you can check us before you spend a minute on it.
The pack sorts three ways — overall priority (money × live TVL × freshness), BROKEN by measured $/day, MISSING by TVL. A $1.6B protocol with no adapter and a $154k/day stale feed are not the same job and should not share one rank.
Two commands per protocol, both in the pack. For example, DefiLlama's own summary endpoint:
curl -s 'https://api.llama.fi/summary/fees/wbtc?dataType=dailyFees' | jq '.total7d'
And the adapter harness, from a clone of DefiLlama/dimension-adapters:
npx ts-node --transpile-only cli/testAdapter.ts fees/wbtc
If the harness prints real numbers while the API shows zero, you have a publish-side or chain-mapping problem, not an adapter problem. If both print zero, check the source before you write anything. That single distinction is the difference between an audit and a slander.
null rather than guessing.confidence: verified. Every one carries the command to overturn it.