This section is for the platform's own operators. Platform → Intelligence health is a read-only page, visible to the super-admin only, that shows whether the intelligence side of the platform is keeping up — using only what the system has actually recorded, with the time each fact was recorded.
- Warehouse. The current data generation and when it was published, with the recent generations behind it. The warehouse refreshes every 15 minutes; if the newest generation is well past that, the page says STALE and states the 15-minute expectation beside it. If warehouse v2 is switched off, the page says OFF — that is a setting, not a fault. Companies marked inactive are left out of the refresh on purpose and are never counted as stale.
- Semantic layer (Cube). Blossom does not keep a live health check for Cube, and the page says so plainly rather than pretending to monitor it. What it can show is the most recent recorded evidence that Cube answered: when the definitions corpus was last indexed, since indexing reads Cube's catalogue.
- Retrieval corpus. How many passages are indexed for tenant documents, this manual, and the definitions corpus, when each was last updated, and which embedding space they sit in. Passages left in an older embedding space are flagged ORPHANED — Ask AI cannot read them until the corpus is re-indexed.
- Companies. For every company: whether it is active, the AI Analyst switch, whether a live AI provider is configured (shown as configured or not — never the address, key or model), the Ask planning and Ask→Act switches, which AI managers are on, and the last manager report the system filed for it. A switch that is off simply reads off; off is a choice, not a failure.
The page has no buttons and fixes nothing — it is where an operator looks before deciding whether anything needs doing at all.