Every result is an audit trail, not a claim.
This is the single line that separates Softinel from every scanner, chatbot, and generator. No output leaves the platform without its reasoning attached.
The anatomy of a Softinel finding.
The exact artifacts Softinel used to reach the conclusion: source lines, graph relationships, traces, manifests, contracts.
A measurable score, not a marketing adjective. Low-confidence findings are surfaced as questions, not answers.
Which analysis produced the finding, at which verification level, and against which snapshot of the model.
A short, plain-English explanation grounded in the evidence — never paraphrased around it.
What breaks, degrades, or leaks if this remains unaddressed — expressed in terms of the actual system.
A concrete next action, not "consider reviewing". Where safe, an auto-fixable patch is attached.
Toggle the pieces. Watch the badge change.
The interactive below shows what happens as each field of an evidence bundle gets added or removed. Notice how the badge changes state — high confidence with full evidence becomes a signed verdict; partial evidence stays a question.
The token refresh loop can complete before the session context is reattached.
One click. One URL. Anyone verifies.
An evidence pack that lives only inside the console is a receipt you cannot hand to anyone. So the console exposes a Publish share URL button on every verdict. Click it and the reporting service:
- Freezes the current bundle — Ed25519-signs the manifest, verdict, findings, and per-file Merkle root at that instant.
- Stores the whole signed pack verbatim under an opaque
pev_…token. - Returns a URL you can hand to anyone:
https://softinel.com/e/pev_…
The link renders exactly what you saw at publish time. If the underlying project changes later, the URL does not.
The receiver hits the URL, sees the signed bundle plus an openssl recipe. No Softinel login, no trust required — the JWS is valid or it isn't.
The console lists every live share with its view count. Revoke on demand — the public URL returns 404 within one HTTP request.
What we refuse to do.
Softinel will not surface a conclusion the platform cannot back up. When the evidence is thin, the system says so. When two analyses disagree, the verification kernel surfaces the conflict — it does not paper over it.
Language models help phrase findings in plain English. They never decide what is safe. The decision path goes evidence → verification kernel → verdict, and nothing in between can be talked into a different answer.
The word "AI" appears exactly once in our product description, and it appears in the negative sense: no AI in the decision path.
Connect your software. Softinel builds the model.
One canonical, connected model of your software. Every question about security, correctness, architecture, and change becomes a query against it — with evidence attached.