Verify — the fifth stage of the convergence loop
Verify is the safety stage: nothing is accepted without evidence. After Unisona acts, it checks the result against external reality before treating it as true. This is the step most AI systems skip — and the one that keeps Unisona honest.
What it means in Unisona
The rule is simple and strict: every important claim needs a claim, its evidence, a confidence level, and a source. An answer that can't be backed by something real — a tool result, a cited page, a file that was actually read — is flagged, not asserted. Unisona would rather say "I don't know" plainly than improvise, and it never invents sources, URLs, or facts about you.
In practice this shows up as grounding (tying answers to real evidence), fact-checking, and a drift canary that watches for the model drifting into confident-but-unanchored territory. Verification is deliberately placed before a result becomes an input to the next step, so a mistake gets caught instead of compounding.
Where it sits in the loop
Verify scrutinizes the output of Act. What survives — with its evidence and confidence attached — is passed to Converge to be recorded.
The convergence loop
Observe → Remember → Reason → Act → Verify → Converge — then back to Observe.
- Observe — take in the world
- Remember — recall what matters
- Reason — decide what to do
- Act — do it with real tools
- Verify — check it against reality (you are here)
- Converge — record the outcome and improve
See the North Star briefing for the whole picture.