Week-Scoped Repository Reconstruction

Intent

Make historical weeks behave the same as current weeks when intelligence or charts are requested.

Context

The orchestrator now fetches all runs for each affected week from the repository and rebuilds summary/trend context for that week before invoking intelligence steps. The merge reconstructs intelligence and visualization contexts from repository-backed week state instead of relying only on newly uploaded runs.

Agentic profile

• System shape: hybrid
• Orchestration mode: sequential

Agent-to-agent interaction

• Present: true
• Mechanism: shared-state
• Evidence: Week-specific context is rebuilt once and then shared across the downstream intelligence steps.

Tool protocols

• MCP: absent
• Tool calling: present
• Evidence: The orchestrator reconstructs durable week state before invoking intelligence generation.

Optimisation target

• Primary: quality
• Secondary: reliability
• Notes: Repository-backed reconstruction ensures intelligence uses the full week state, not just the latest upload slice.

Simplicity vs autonomy

• Position: balanced
• Rationale: Agents still decide over the context, but the system now reconstructs that context deterministically from persisted history.

Forces

• Upload batches may contain only a subset of runs relevant to a week.
• Historical week navigation must not depend on the original upload session state.

Solution

Use run repository lookups and week-specific summary/trend reconstruction helpers to prepare each week context before intelligence and visualization execution.

Implementation signals

• Orchestrator fetches all_week_runs from run_repo.get_runs_for_week(...) before per-week intelligence.
• SnapshotService adds build_agent_context_for_week(target_date).

Evidence

• Both upload-time and on-demand intelligence paths rebuild week-specific state from repositories.
• A dedicated helper constructs summary and trend context centered on one selected week.

Consequences

Benefits

• Improves consistency between current and historical week behavior.
• Makes chart and insight generation less sensitive to upload path shape.

Costs

• Increases repository reads during orchestration.
• Adds more context-building logic to the service layer.

Failure modes

• Repository gaps can still force fallback to session runs.
• If persisted runner identity is wrong, the wrong week data may be reconstructed.

Reuse notes

• Use when agent execution should depend on durable period state rather than only the latest request payload.
• Use it when quality 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.