Decision and terminology ledger
Offer a short choice and an editable free-text alternative for a consequential question. State a recommendation and its reason without pretending it is an owner's decision. The fields below are a summary pattern; a conversational host need not render a form, and an absent dashboard form must not be claimed as implemented.
| Question | Decision/status | Evidence or owner answer | Consequence | Revisit trigger |
|---|---|---|---|---|
| Who owns writes? | Settled: companion | Owner answer and writer trace | Offline browser changes wait | Adding a second writer |
| What counts as applied? | Settled: local acknowledgement | Persistence/replay evidence | Queue receipt is insufficient | Transport contract changes |
| Which preview feels right? | Open: owner | Three runnable variants | Specification waits on selection | Owner selects variant |
For a consequential accepted choice create DOC with data.kind=adr and body: context; options considered; chosen behavior; rationale; consequences; evidence; owner/date; supersedes; revisit condition.
For terminology create/update a glossary DOC: term; precise meaning; excluded meanings; implementation/record links. Preserve unrelated terms on update.
Example: define “Review” as completed agent work awaiting inspection, “Done” as owner-confirmed completion, and “deployed” as an independently evidenced delivery state. These are different concepts even when one task has all three events.