The refusal log.

The gate was written down before the data was pulled, which is the only way a gate means anything: pull 90 days of liquidation events, attribute each to its caller, count distinct liquidator addresses per venue, and build nothing until a venue shows fewer than five active liquidators and a modelled net margin above gas.

We then pulled every Morpho Blue liquidation on every chain Morpho Blue runs on, for the trailing 90 days. 23,188 events against an API total of 23,188, so the census is complete rather than sampled, keyset paginated and deduplicated on transaction hash and log index.

It fails in both directions at once. Where there is volume there is a crowd: Base has 341 distinct liquidators across 8,719 liquidations, Ethereum 171 across 1,205, HyperEVM 69 across 1,060. Where the field is thin there is no volume: Monad has 8 actors, Unichain 9 and OP Mainnet 5, and between them they produced 51 liquidations in 90 days at average gross values of $151, $138 and $43. Winning all three outright for an entire quarter is roughly $7,000 gross, before gas, slippage, infrastructure and tax.

The detail worth keeping is that Base's low top-three concentration, about 38%, is the discouraging reading rather than the encouraging one. 341 addresses sharing flow evenly is a commodity already competed down to its margin, not three incumbents to out-engineer.

All 7,410 Morpho markets were scanned. 270 of them, about 3.6%, have a deployed pre-liquidation contract, and those markets carry $4.68B of borrow between them.

That $4.68B is the number a pitch would lead with, so here is why it is not used that way. It measures deployment, not activity. A guard pass over the four highest-borrow pre-liquidation contracts on Base returned zero events in 90 days. On Ethereum only twelve contracts fired at all. The flow is in small markets and the headline is in large ones, and they are not the same markets.

Where flow does exist, the field is genuinely thinner than on standard liquidation. Base ran 962 events across 17 distinct actors, against 341 actors on the standard surface. That is the best ratio in either census. It is also not uncontested, because the top two addresses take 76% of it, and the Ethereum list is visibly professional.

Seventeen is not fewer than five. The gate failed. It would be easy to argue that the gate was calibrated for different economics, and that 17 actors across 962 events is a different proposition, and that argument may even be right. The honest thing is still to record the gate as failed rather than to quietly move it after seeing the data. Separately, realised margin after slippage and gas is unmodelled, and that is what actually decides this, not more counting.

471 markets saw at least one liquidation in the 90-day window. 251 of them, 53%, had exactly one distinct liquidator, and 185 of those still carry open borrow. The largest is a $3.93B borrow market served by a single address. It is a striking set of numbers and it points at the wrong conclusion.

Measured incidence of bad debt by liquidator count: one liquidator 27%, two 24%, three 23%, four 27%, five 29%, six 38%. Incidence is flat at roughly a quarter across the thin markets, and it rises as coverage increases, which is the exact opposite of the hypothesis.

The reason is that coverage is endogenous. Liquidators show up where there has been stress, and stress is what produces bad debt in the first place. Liquidator count is a proxy for activity, not for safety. The high-count cells are also samples of one to three markets and must not be over-read in either direction.

So coverage survives in the framework only as a continuity and dependency input. The question it is allowed to answer is what happens to a market if that one address stops showing up. It is not allowed to answer whether the market is likely to take a loss. A framework built on the first reading would have looked rigorous and been backwards.