Straight answers, including the unflattering ones

Treasuries, funds and protocols that have to choose where to put stablecoins and then defend the choice. The person we are most useful to is the one who writes the forum post arguing for an allocation — or who answers for it a quarter later.

It is not for anyone looking for yield recommendations, and not for anyone wanting us to hold their money. We do neither.

Venue risk gets assessed three ways today, and each fails differently.

• TVL and APY reward the loudest venue, not the soundest. Both look backwards, and neither tells you who holds the admin keys.
• A one-off report is right the week it is written. A governance vote or an upgrade the month after makes it quietly wrong, and nothing tells you it has.
• A black-box score cannot be interrogated, so it cannot be defended. “Their model said so” is not a rationale a delegate accepts.

The expensive failure is not the bad allocation. It is that after a loss nobody can reconstruct what was known at the time, so every decision gets re-judged in hindsight and the honest answer to “did you know?” becomes “we are not sure what we knew.”

An assessment of one venue: a 0–100 score where higher is riskier, the band it falls in, the evidence behind each of the six dimensions, the date and block height it was computed at, the methodology version, and the observation set it ran on — hashed, so you can recompute the number yourself.

The score is not the product. Being able to recompute it is. A number you cannot check is an opinion with a decimal point.

• The assessment — score, band, per-dimension evidence, date, block height, methodology version. The document you forward to a delegate.
• The observation set — every input the score was computed from, hashed. This is what makes it checkable rather than assertable.
• What would move it — which dimension sits nearest a band edge, and what change would push the venue across it. A verdict becomes something you can monitor.
• An implied cap — the position size the score supports against your own risk budget. The real question is never how risky a venue is; it is how much.
• Re-assessment — the same venue scored again when something material changes. Scores go stale on governance changes, upgrades and oracle swaps.

Some should. If you have analysts and few enough venues to watch, this is work you can own, and owning it beats buying it.

It stops being true at the point where it becomes a standing cost. Doing it properly means reading audits, governance threads and admin configurations for every venue, then doing it again whenever any of them change — roughly an analyst-week per venue, repeated. That work is nearly identical across everyone who needs it, which is the definition of something worth doing once.

It is a real market with capable incumbents. The difference is which job each one takes.

• Risk-parameter firms run deep simulation work for large protocols, optimising a protocol’s own parameters. We assess venues on behalf of the allocator, and we publish the weights rather than only the conclusions.
• Vault curators take the decision off your hands entirely and are paid on performance. That is a discretionary manager. We sell the method so you keep the decision.
• Dashboards and data vendors supply inputs, and the good ones supply them well. Nobody is answerable for what the inputs add up to, which is the part a treasurer actually needs.

You should not have to. That is why the weights are published, the thresholds are published, and the observation set ships with every score. Hold the same inputs and you get our number, or you do not — and if you do not, we want to hear about it.

We publish our mistakes too. The first calibration error is still on the method page with the wrong number left in: exit liquidity was scored on withdrawable share alone, which punished large lending markets for lending and rated an $840m market 69/100. Eighteen passing tests did not catch it. A method nobody has found a fault in is usually a method nobody has checked.

We do not have a price yet, and we would rather say so than publish a number we invented. It will be set against what the first customers actually use — per venue, per watchlist, or a flat subscription.

If you have this problem, talk to us early. Early conversations shape the product and get priced accordingly.

No — and you never will be able to. Diligentia deploys its own capital only. We hold no outside money, we take no custody, and there is no vault to deposit into.

That is a structural choice rather than a stage we grow out of. The regulatory weight in this business comes from handling other people’s money, and we have chosen to handle none.

No. It means the things we measure looked better than at a venue scoring higher, on the date we measured them. Venues with low scores have failed before and will again.

A score is a statement about what was observable on a date. It is not a prediction and not a guarantee, and anyone selling you the second thing is selling something they do not have.

Pre-launch. Specifically, so nobody has to guess:

• No capital is deployed. The book is not running yet.
• No external audit has been commissioned.
• No track record, no returns, no clients. There is nothing to report because nothing has happened yet.

What does exist, and can be checked:

• The rubric is built and tested, with its six weights published and calibrated against four reference profiles.
• It has already caught one error in itself, and that correction is published.
• The execution contracts are written and tested against live venues on Base, including fork tests that measure real positions rather than assert that nothing reverted.

We would rather be read as early than as further along than we are. The second one only works until somebody checks.