Four things nobody else has together
1. We run the nodes
Most tools resell a node-provider API, so their evidence inherits somebody else's terms of service and retention policy.
We operate twelve-plus blockchain full nodes ourselves. When someone asks how you know a transaction propagated when it did, the answer is we observed it at this height at this time — not a vendor told us.
A fair question for any vendor: is your chain evidence first-party?
2. We publish calibrated error rates
Any system that assigns risk will be wrong sometimes. The obligation is to know how often and say so.
Two instruments make it possible. A closed-loop chain where we author the transactions and implement the typologies ourselves, so we hold the complete answer key and can score our detectors honestly, including where they fail. And a honeytoken programme that mints bait assets never meant to move, so anything touching one is a confirmed fact rather than an inference.
No competitor we have surveyed has either one.
This is not academic. When attribution methodology met a serious defence in the Tornado Cash prosecution, the absence of published vendor error rates was exactly the attack.
You can watch us do this in public. One of our free datasets documents a defect in our own CoinJoin detector: roughly three quarters of its output sits at denominations where batching and dust are far more likely than mixing. We publish the distribution that exposes it and the filtered subset we stand behind, rather than the flattering headline.
3. Glass box, always
If a score cannot be broken into the features that produced it, an analyst cannot defend it to a prosecutor. If an alert cannot cite the rule that fired, a regulator cannot examine it.
AI here is analyst-time compression with an audit trail, not an oracle.
4. Civil liberties: an account, not a promise
Every vendor in this market has a page saying they respect civil liberties. Ours is an account of what our own audit found.
In August 2026 we found Nexus Prism inferring ethnicity from surnames and geography, per individual, across property records — and using it to rank people. Not a dormant column: there was an index built to make that ranking fast, and a scoring rule worth three points.
Migration 726 dropped the table — 4,059,250 rows of inferred ethnicity, gender and nationality about named property owners. Migrations 756 and 757 then went back for the 13,801 scores that had already been served to subscribers through a published feed. The two models that produced them now sit behind a gate that fails closed: if it cannot read its own authorisation record, it refuses rather than proceeds.
Aggregate, tract-level measurement is retained deliberately. Measuring disparate impact in your own output is the obligation, not the offence — and a test asserts that it stays permitted, so a later sweep cannot quietly remove the thing that lets us audit ourselves.
The least flattering part: our first fix reported success and was wrong. It searched the served feed for the wording one of our two pipelines uses; every affected record came from the other one. The pattern was correct, correctly spelled, and matched nothing — so it truthfully reported zero records naming the score while 12,673 of them said it in plain text. A count of zero from a pattern that never matched is indistinguishable from a job well done. The second migration exists because we went looking again.
A fair question for any vendor: not whether they have a civil-liberties policy, but what their last audit found.
And what we are not
We are not the deepest chain tracer, and we will not pretend to be. The incumbents have spent years building labelled address corpora we have not caught up to. We are not trying to replace them — we are the darknet context, the privacy-coin work, and the correlation they do not carry.
We have no reference customers yet. We built the whole platform to production before asking anyone to buy it. That is the trade.
Some of what we sell is engineered but not yet shipped. Our sovereignty practice is at specification stage with reference builds behind it, not a catalogue you can order from. We say which is which on the sovereignty page.
If you catch a vendor overselling once, you should stop believing everything else they told you. We would rather tell you the gaps ourselves.