Language-Keyed Intelligence Cache

Intent

Avoid stale cross-language reuse when intelligence is rebuilt lazily or week selection changes.

Context

Cached intelligence snapshots are stored and retrieved by week plus language, and cache misses trigger rebuild instead of language fallback reuse. The merge uses (week_start, language) as the intelligence cache key so multilingual outputs are reused safely.

Agentic profile

• System shape: hybrid
• Orchestration mode: event-driven

Agent-to-agent interaction

• Present: true
• Mechanism: shared-state
• Evidence: Cached week intelligence is shared by the orchestrator and snapshot service using the same language-aware key.

Tool protocols

• MCP: absent
• Tool calling: absent
• Evidence: Cache lookup and invalidation are controlled inside the orchestrator and snapshot service rather than through external tools.

Optimisation target

• Primary: latency
• Secondary: quality, reliability
• Notes: The cache prevents unnecessary rebuilds while avoiding stale language reuse.

Simplicity vs autonomy

• Position: balanced
• Rationale: Caching is simple operationally, but keyed by language to respect multilingual agent outputs.

Forces

• Intelligence and UI text are language-specific.
• Week selection should be fast, but cached content must remain semantically valid.

Solution

Key intelligence snapshots by both week and language, clear matching keys during invalidation, and return None when a week exists only in another language so the orchestrator rebuilds it.

Implementation signals

• SnapshotService cache uses Dict[Tuple[date, str], Any].
• Orchestrator returns None if only another-language snapshot exists for the week.

Evidence

• The snapshot service manages cache entries keyed by both week and language.
• The orchestrator refuses to reuse a cached week snapshot when it belongs to the wrong language.

Consequences

Benefits

• Prevents stale localized UI content.
• Supports cheap repeated week views in the same language.

Costs

• Increases cache footprint.
• Makes cache invalidation slightly more complex.

Failure modes

• Language-specific cache misses can increase rebuild frequency.
• Incomplete invalidation can leave stale entries for some languages.

Reuse notes

• Use when agent outputs are localized and cache reuse must remain language-safe.
• 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.