Product Guide
Decisions
Review architecture findings, risks, debt, suggestions, and what-if ideas
Decisions is the review queue for AI-assisted findings. Each item includes its evidence, affected architecture elements, proposed action, and lifecycle state.
Decision Types
| Type | Purpose |
|---|---|
| Card assignment | Suggest an Architecture Card for an element, relationship, or workflow |
| Card removal | Remove a card that no longer fits the evidence |
| Documentation gap | Identify missing or thin documentation |
| Architecture concern | Surface a structural or operational concern |
| Pattern identified | Record a recognizable architectural pattern |
| NFR risk | Flag a quality-attribute risk |
| What-if idea | Propose a forward-looking improvement, optionally grounded in Data Model evidence |
| Risk finding | Promote a reviewed risk into the Architecture Handbook |
| Technical-debt finding | Promote reviewed debt into the Handbook |
| Decision suggestion | Create a proposed ADR for formal team review |
Lifecycle
| State | Meaning |
|---|---|
| Needs review | Waiting for a person to evaluate it |
| Applying | The approved side effects are in progress |
| Needs reconciliation | Application was interrupted and must be resumed or reconciled |
| Acknowledged | An advisory finding was accepted without a model mutation |
| Applied | The approved change completed |
| Dismissed | The finding was rejected with optional feedback |
| Expired | The review window ended before action |
Review Workflow
- Confirm the selected project version.
- Filter by state, type, severity, affected elements, or confidence.
- Read the reasoning and cited project evidence.
- Discuss ambiguous tradeoffs with Archie.
- Apply an actionable item, acknowledge an advisory item, or dismiss weak evidence.
- Resume safely when an item needs reconciliation rather than creating a duplicate change.
- Promote durable conclusions into the Handbook or Project Knowledge.
Approving a risk, technical-debt item, or decision suggestion does not make the suggestion unquestionable. It creates structured Handbook material that the team should continue to govern.
Generation And Cost Control
Analysis is debounced and deduplicated so rapid model edits do not create a storm of equivalent findings. Project-level reviews and scheduled analysis operate on bounded project evidence. What-if ideation runs less frequently than routine per-entity analysis.