Try it yourself, free
The fastest way to distrust a claim about a model is to check it yourself. So the point of entry is not a sales call — it is a reading you can look at before you have committed to anything. Two doors are open right now, one per house, and each answers a different question about the model you are weighing.
Ardora's atlas — and a free read
Start at the Ardora atlas. It is an open, family-navigable field guide to open-weight models: their dispositions, their lineage, and — where both a finetune and its base are charted — how the finetune moved off its base. You can browse it without asking anyone for anything, and see how the model you are considering sits among its kin.
If the model you care about isn't charted, Ardora will read one of yours free and you keep the field-note — a characterization with its replayable witness and its stated ceiling. It answers "what is this?", never "is it safe?" — that boundary is the whole design. Use it to decide whether a candidate is even the right animal before you spend a security budget on it.
Protora's audit — the open door to assurance
When the question turns from what is this to is there something in here I have to act on, that is Protora. The Audit is open now: bring a finetune and the house reads what it did — including what it was shaped to hide — proves it with a replayable witness you can re-run on your own hardware, and hands back a signed dossier. Where it can't prove something, it abstains and says so.
You do not have to take the dossier on faith, which is rather the point: every finding ships with a reproduction you can run with ordinary tooling, on your own hardware. See the receipts and coverage stamps in the record, and the head-to-head framing in the bench. If a receipt claims a safety guardrail was stripped, you can replay the witness and watch the effect fire for yourself — the claim never rests on our word, which is the only basis on which a security decision should be made.
What's staged — said plainly
Some surfaces are on the roadmap and not yet launched. Do not plan around them as if they were live:
- Self-serve Detect — running the read yourself, unattended — is staged. Today the read is delivered as the Audit engagement, not a button you press alone.
- Monitoring (a diff on every release) and Drift Guard (an alarm during training) are staged.
- Portfolio analytics is staged.
They are described on the product map so you can plan for them — but the open entry points today are the two above.
A concrete first step
Pick the door that matches your question.
- Weighing which model to build on, or want to know what a model is? Open the atlas, find your candidate, and if it isn't there, claim the free read.
- About to ship a finetune, or just downloaded one you can't fully account for? Bring it to Audit and get back what it did, with a witness on every finding.
Both start with a model and end with something you can read for yourself — which is the only kind of proof worth starting from. Neither door asks for a commitment before it shows you something real, and that is deliberate: the entry point is a reading, not a pitch. When you need the deeper work — excision with proof, which is trust-gated, or continuous coverage across releases once Monitoring lands — the ladder climbs from there. Browse models to see what has already been read, then bring the one you are weighing.
