Analytics hub

Normative product contract: Analytics industry profile (exceedance-first fleet risk, explorer, alarm ops, prediction as secondary).

The Analytics sidebar item (/analytics) covers fleet analysis and industrial condition-monitoring expectations.

What you can do today

Capability	Where
Ask (a sentence, or a suggestion from this tenant's fleets)	Analytics → Ask
Metric Explorer (type + key + range + aggregation)	Analytics → Metric Explorer
Time-bucket series charts	Explorer (auto / 5m / 1h / 1d)
Top/Bottom rankings + fleet KPIs	Explorer
Threshold anomaly count	Explorer (Above/Below)
Linear forecast	Explorer + POST /api/analytics/forecast (engine=linear) — capped
Seasonal / StatsForecast predictive	Explorer Predictive + optional forecast-worker sidecar
Pattern heatmap (hour × weekday)	Explorer
Z-score statistical anomalies	Explorer + POST /api/analytics/statistical-anomaly
Asset hierarchy rollups	Explorer + POST /api/analytics/asset-rollup
Alarm ops (MTTA/MTTC, noisy rules)	Analytics → Alarm ops
Shared saved views	Analytics → Saved views (private / tenant / shared emails)
Learned watch (opt-in alarm when a reading leaves its normal band)	Ask, or POST /api/analytics/watches
Time-to-limit and next-period total, in one sentence	Ask
Vibration level, peak, crest, and dominant frequency	Ask, on a vibration key or a named reading
Profile calculated fields on ingest	Device Profiles → Calculated fields
Analytics Solutions	Phases A–C packs

How to ask

The device type and the reading are dropdowns on the Analytics page. Run uses those dropdowns.

Ask matches the sentence to this tenant’s device types and keys. The number is calculated from stored telemetry. Ranking, a linear forecast, a vibration check, and a learned watch are what that calculation does. An open language model is not part of the answer.

Ask uses the same names inside a sentence:

1. Fleet: say the device type with spaces. energy_meter is “energy meters”. ht_sensor is “ht sensors”. machine is “machines”.
2. Reading: say the telemetry key with spaces. voltage_vln_avg is “voltage vln avg”. temperature is “temperature”.
3. One device: use the name shown on the device, such as avh1. The long id is not the name.
4. Or press a suggestion. Those lines are filled from the fleets and readings this tenant already has.

Example: “Which machines have the highest temperature in the last 24 hours?”

A watch stays off until someone asks. Example: “Watch temperature anomalies on Machine 3.” It learns the normal from about 30 readings, then opens a warning alarm when the reading drifts, and clears the alarm when the reading returns. Scoring runs every 10 minutes. Nothing is watched for a tenant until a watch exists.

If the sentence names neither a fleet nor a reading, the answer lists what this tenant has.

Forecast mode UX

Fleet risk ranking answers “which of my 500 devices look worst?” using the industry profile: crossings → samples over limit → time over limit, then score/forecast assist. Set Above/Below or ranking is not exceedance-based.

Forecast chart is drill-down only (Linear/Predictive, Top-N auto or manual multi-select, capped device count). Use Chart on a risk row to inspect one device.

Limits: GET /api/analytics/forecast-limits · APIs: POST /api/analytics/forecast, POST /api/analytics/fleet-forecast-risk.

Surface	Default max devices	Purpose
Linear chart	12	Detail line chart
Predictive chart	5	Detail seasonal chart
Fleet risk scan	1000	Rank whole type

Capability map

Capability	Now	Later
Metric Explorer	Yes	Richer calendars / heatmaps
Calculated fields	Profile expressions on ingest	More expression functions
Asset rollups	Hierarchy rollup API + Explorer	Multi-metric trees
Anomaly	Threshold + stale + one-shot z-score, plus an opt-in learned watch	Trained models
Prediction	Linear sentence (time-to-limit, next-period total) + StatsForecast worker (AutoETS) with fallback	Batch cache / accuracy UI
Saved views	Server private/tenant/shared + local fallback	Folders / permissions UI
Visualizations	Bar, line, spectrum, hour×weekday heatmap	Calendar / more view types
Ask	Sentence uses this tenant's device types and keys	Open language model

APIs (confirmed)

• POST /api/analytics/forecast — history + linear forecast points per device
• POST /api/analytics/statistical-anomaly — mean/stdDev/z-score vs window baseline
• POST /api/analytics/asset-rollup — aggregate a key over asset root + descendants
• GET/POST/DELETE /api/analytics/saved-views — visibility: private | tenant | shared
• GET /api/analytics/prompts — suggestion sentences from this tenant's fleets and numeric keys
• POST /api/analytics/chat — one sentence to a ranking, forecast sentence, vibration report, or watch
• GET/POST /api/analytics/watches and DELETE /api/analytics/watches/:id — learned watches. Off until created.

Migrations / env

• 0039_analytics_saved_views.sql
• 0040_analytics_saved_views_sharing.sql
• 0044_analytics_watches.sql
• CALCULATED_FIELDS_ENABLED (default true) — see backend/ENVIRONMENT.md

Field validation

API + UI smoke checklist: Field validation.

Related

• Phase 4 · Phase 5 · Phase 6
• Validation status (hardware bench tracker)