The confident wrong answer is the enemy.
A green checkmark with nothing behind it is worse than no answer at all — it moves people to act on false certainty. Every product here is designed against that one failure mode. This is the thinking underneath.
What each answer actually looks like.
Proven true
Not proven — held distinct
Proven risk
Four honest answers, not one hopeful one.
Across the products, a result is only ever one of four things. The distinction between them is the product. Most of the value is in refusing to collapse "unknown" into "fine".
Proven true
The claim is backed by reproducible evidence — a sandbox run, a signature that recovers, a rule that a machine can re-check. This is the only state that carries a green mark.
Not proven
Inputs are missing, providers disagree, or a scan couldn't complete. The answer is unknown — held distinct from a pass, never quietly rounded up to one.
Proven risk
Evidence shows a real problem — a matched CVE, a basket item added without consent, a stale verdict. The finding points back at exactly what triggered it.
Yours to act on
Even a verified result is advice, not an action. Where a tool touches live systems, authority to act stays with your team and your governance — by design.
The rules every product inherits.
These aren't slogans; they're constraints the code is written to satisfy. When a new product joins OneZero, it adopts them — that shared discipline is the point of the network.
Evidence over reputation
Stars, popularity and vendor claims are signals about perception. We measure behaviour instead, and show the measurement.
Reproducible or it doesn't count
A result another party can't independently re-derive isn't a result. Signatures, hashes and deterministic rules exist so verification doesn't require trusting us.
Fail closed, in the open
When something can't be proven, the safe answer is the honest one: unknown, surfaced clearly — not a silent success that looks identical to a real pass.
Seeded is not verified
Bootstrapping data is labelled and capped everywhere it appears. Example and cold-start content is never presented as production evidence.
Advise, don't seize
Our tools inform decisions; they don't take irreversible action on your behalf. The human, or your governance, stays in the loop.
The marketing obeys the same rule
What we publish about a product is bounded by what its code does today. If a capability is a goal, we call it a goal.
Check it, prove it, sign it, remember it.
The five products map to four moments in the life of a system. Each hands the next a signed, portable artifact — so evidence travels between stages instead of being re-established from scratch. You can adopt any single stage on its own.
Vet what you're about to run
Sandbox an open-source AI tool, measure its real footprint, scan it, and score it before it reaches production.
→ VerditNxtGen · Verdit CLI
Verify changes against reproducible rules
Deterministic checks and evidence graphs establish what's true — not what a model guessed — before anything ships.
→ Causa · verification engine
Emit a portable, verifiable verdict
Results are cryptographically signed so a contributor, a smart contract, or an auditor can check them independently.
→ Axiom · signed receipts across the stack
Turn every outcome into shared knowledge
Diagnoses, fixes and verified outcomes accumulate, so the same failure costs less to resolve each time it recurs.
→ Causa · knowledge layer · BizOpsTool
The principles are only worth the proof.
Read exactly what each product claims — and what it deliberately doesn't — or bring us a real system to run against.