Evidence
Conversations, pasted transcripts, shared links, repository docs, notes, and mail enter the record with source metadata.
Control lifecycle
Evidence becomes trusted state through explicit boundaries.
AI remains probabilistic. ReasonTrail governs the context it receives, the changes it may propose, the actions that execute, and what is ultimately remembered.
Controlled stage
Bounded context
Intent is discovered, a project is selected, evidence is ranked, and the context package is constrained to a token budget.
Proposal
The configured AI provider or agent receives selected context and returns a proposal rather than an automatic change to trusted state.
Authorization
People and policy determine what may proceed. Sensitive Operator writes require scoped approval for the run step.
Verification
Execution outcomes, failures, retries, and evidence are persisted. Only verified results can update trusted memory.
The principle
Verified before remembered.
ReasonTrail owns the project record and keeps important writes reviewable. A model assertion is not treated as organisational truth merely because it was generated.
01Evidence↓
02Context↓
03AI / agent↓
04Proposal↓
05Authorization↓
06Execution↓
07Verification↓
08Memory