Intent
Support interactive week selection without precomputing full intelligence for every historical period.
Context
The UI can ask for a specific week, and the orchestrator returns a cached intelligence snapshot if complete or builds it on demand from persisted data. The merge introduces a cache-aside path for week intelligence so dormant weeks are materialized only when the UI requests them.
Agentic profile
- System shape: hybrid
- Orchestration mode: event-driven
Agent-to-agent interaction
- Present: true
- Mechanism: shared-state
- Evidence: The orchestrator rebuilds missing week intelligence into one shared snapshot state only when the cache contract requires it.
Tool protocols
- MCP: absent
- Tool calling: present
- Evidence: Intelligence steps are invoked lazily on cache miss rather than eagerly for every week.
Optimisation target
- Primary: latency
- Secondary: cost, quality
- Notes: Defers expensive intelligence work until a week is actually requested.
Simplicity vs autonomy
- Position: balanced
- Rationale: The build is autonomous once triggered, but only triggered when the cache contract demands it.
Forces
- Historical intelligence can be expensive to precompute.
- Week-level UI navigation must remain responsive.
Solution
Implement get_or_build_intelligence_snapshot(...) as a cache-aside function that checks completeness first, reconstructs week context from repositories on miss, runs only intelligence-related steps, stores the result, and returns it.
Implementation signals
- New
get_or_build_intelligence_snapshot(...)path in orchestrator. - Tests verify the same week is not rebuilt twice.
Evidence
- The orchestrator rebuilds intelligence lazily from persisted weekly context and stores the result for reuse.
- Tests verify on-demand build on the first request and a cache hit on the second request for the same week.
Consequences
Benefits
- Avoids eager computation for rarely viewed weeks.
- Reuses one orchestrator pathway for both precomputed and on-demand intelligence.
Costs
- Adds another execution path that must stay consistent with the main run path.
- Requires a completeness check to avoid returning partial snapshots.
Failure modes
- Incomplete snapshot detection can be wrong.
- Repository reads can fail after the original upload session is gone.
Reuse notes
- Apply when an agentic system supports interactive drill-down into periods or entities that should not all be materialized eagerly.
- Use it when latency matters more than maximizing autonomous generation everywhere.
Confidence
High. Confidence is high because the pattern is evidenced directly in the PR132 code paths and tests, with breadth consistent with a stable pattern rather than speculation.