What qualifies (methodology 2026-09-19)
Open vaults where we measured automated execution from their own timing, or which are declared or sit on a venue built to run software. A name alone does not qualify. A vault held wholly by another tracked vault is excluded: it cannot be bought and its capital is inside its parent. Automated execution is a machine at work, not evidence that a model decides.
Changed 2026-09-19
- Renamed from "machine-managed capital" to "capital under automated execution"; the membership rule is unchanged in kind and now stated.
- Morpho Vault V2 entered the universe: 200 vaults, of which 84 qualified on allocator timing, counting only allocations the vault’s own allocators called.
- Vaults held wholly by another tracked vault (HLP’s sub-vaults) left the universe as not investable.
- A vault classified on its name alone no longer qualifies.
The failures are in it
Constituents are chosen on their size at each rebalance and on a return observed by then, and a vault that dies keeps its loss in the index permanently. That is the one thing every index in this asset class gets wrong: pick today’s survivors and the history looks wonderful, because the ones that blew up were quietly left out of it. 2 constituents died inside the series and their loss is in the line above.
Return means profit, not deposits
A vault’s account value rises when somebody deposits, so a series built on it measures fundraising. Returns here are trading profit against the capital that was there to earn it, compounded through the venue’s own observations. A vault whose venue does not report profit separately is excluded rather than guessed at.
One vault cannot be the index
Weights are capital-proportional but capped at 20%, because one vault can otherwise dominate the capital under automated execution that we track. The series also does not begin until 5 constituents can be measured; 6 earlier periods were too thin and are not published.
What this index has not been tested against
Point-in-time membership and dead constituents are handled and tested. Three things are not, and a reader quoting the level should know which: it charges no fees, so it is a gross series where a holder would have paid a management and performance cut; it assumes a rebalance can be executed at the observation, which a vault with a deposit cap or a withdrawal queue would not have allowed; and the constituent set is drawn from the vaults we read, so a venue we do not cover is absent rather than weighed and rejected. Until those are tested it is a research index, and that is how it should be cited.
What we could not see
Coverage is the share of index weight actually observable in a period. The worst here was 100%. A gap in our data is not a flat month, so it is printed rather than smoothed over.