Rosetta makes an AI system answer five questions
A Rosetta Review is one visible moment. The stack keeps the underlying state governable across memory, change, trust, learning, and action.
“Why did it act?” refers to the governance-relevant state that shaped the action — evidence, goals, constraints, uncertainty, pressures, and arbitration — not private model internals or a transcript of hidden reasoning.
What did it know?
MemorySemantic state
The complete case: claims with sources, goals, open disagreements, and declared gaps — structured state that any conforming model can read, held in the institution's own storage.
What changed?
LedgerGoverned mutation
Every change to the case is a validated, append-only transaction. Any earlier state can be reconstructed; nothing is edited in place, and alterations at the storage level are detectable.
What deserved trust?
TrustEpistemic verification
Claims carry earned standing — from self-reported through independently verified. Unexamined evidence caps standing no matter how many reviewers agree; disputes stay attached until worked through.
What did it learn?
GrowthGoverned learning
Finishing the task is not enough. What was tried, what changed, and whether the improvement transfers are recorded and independently checked. Authority expands with verified capability and contracts when the evidence changes.
Why did it act?
JudgmentCognitive governance
Goals, constraints, uncertainty, and pressures are weighed explicitly before consequential action, producing a disposition and a record. Nothing consequential happens without a decision that can be inspected.
Technical detail — scope and boundaries
These five descriptions are behavior-level: what a conforming system does, never how. Formats, schemas, rules, and thresholds are specified in the protocol documents, which are patent filings on record, disclosed under NDA. The stack described here is the five-protocol family; runtime operating-state layers above it are not represented on this site.