Analytics industry profile (Autoconnecto)

Status: normative for Analytics product decisions.
Goal: industrial condition-monitoring and fleet analytics expectations — not invent metrics in ad-hoc chat.

This profile separates confirmed Autoconnecto behavior from target industry alignment.

---

Standards we align to

Layer	Industry reference	Product meaning
Metric exploration	BI / IoT metric explorers	Type + key + range → fleet KPIs, series, rankings
Exceedance / condition monitoring	PdM practice: exceedance frequency, duration, max excursion	“Which assets are unhealthy?”
Statistical process control (SPC)	Western Electric / Nelson rules (target)	Baseline ± limits; rule violations (later)
Alarm performance	ISA-18.2 / EEMUA 191 concepts	MTTA/MTTC, floods, chattering, noisy sources
Prediction	Classical forecasting	Secondary to exceedance; never the only risk signal
Asset health	ISO 55000-style likelihood view	Ranked health / risk for a device type or asset tree
Calculated metrics	Ingest-time calculated fields	Profile expressions on ingest

We do not claim formal certification against ISA/ISO. We do use their vocabulary and default UX so buyers recognize the product.

---

Layer model (mandatory)

1. Exceedance analytics     ← primary “devices with issues”
2. Alarm ops KPIs           ← operational noise / response
3. Metric Explorer          ← investigate any key/range
4. Prediction / forecast    ← “will it exceed again?” (drill-down + fleet assist)
5. SPC / learned anomaly    ← next (z-score today; rules/models later)
6. Health score             ← roll-up of 1–5 for ranking

Rule: Fleet “top risky” lists are driven by Layer 1 (+ 6), not by chart device caps and not by forecast-only scores.

---

Layer 1 — Exceedance analytics (normative)

Given a telemetry key, time range, and operator limits (above / below):

Metric	Definition	Why it matters
Breach episodes (crossings)	Count of transitions from OK → over limit	“How many times did it cross?” — primary industry signal
Breach sample count	Samples with value over limit	Intensity of exposure
Time over limit	Estimated duration in breach	Sustained problems vs spikes
Currently breaching	Latest sample over limit	Standing problem now
Max excursion (target)	Worst distance beyond limit	Severity (implement next)

Ranking order (fleet risk table)

1. Breach episodes (desc)
2. Breach sample count (desc)
3. Time over limit (desc)
4. Risk / health score
5. Forecast assist (will breach in horizon) — tie-break only

Without above/below, exceedance ranking is undefined. UI must warn and must not claim “failure risk.”

---

Layer 2 — Alarm ops (confirmed direction)

KPI	Status
MTTA / MTTC	Confirmed in Alarm ops
By severity	Confirmed
Noisy rules / devices	Confirmed
Chattering / flood indices	Target

---

Layer 3 — Metric Explorer (confirmed)

• Select device type + key + range (+ aggregation)
• Fleet KPI, Top-N, series, threshold anomaly count
• Saved views (private / tenant / shared)
• Asset rollup (hierarchy)

---

Layer 4 — Prediction (confirmed, secondary)

Surface	Cap	Role
Forecast chart	Strict (e.g. 5–12 devices)	Drill-down after exceedance ranking
Fleet risk assist	Full type (batched)	Optional “will breach again” flag
Engines	Linear · seasonal-naive (StatsForecast-ready)	Open-source, $0 API

Prediction must not replace exceedance as the main “risky devices” answer.

---

Layer 5 — SPC / anomaly (partial)

Capability	Status
Threshold + stale anomaly summary	Confirmed
Window z-score	Confirmed (one-shot on the selected window)
Learned watch	Confirmed. Opt-in per device and key. Mean and spread, then a warning alarm when the latest reading drifts. Clears when it returns.
Control chart / Nelson rules	Target
Unsupervised trained models	Target

---

Layer 6 — Health score (normative formula v1)

For each device in a typed fleet scan with limits set:

• Start from exceedance severity (episodes, samples, duration, standing breach)
• Add small assist from forecast-toward-limit
• Expose:
  • riskScore 0–100 (higher = worse)
  • healthScore 0–100 (higher = healthier) = 100 - riskScore

UI default sort: worst health / most crossings first. Pagination first page ≈ top 10 risky.

---

UX contract (non-negotiable)

1. Fleet risk ranking = full device type scan (batched), exceedance-first.
2. Forecast chart = capped drill-down; selection = Auto Top-N or manual multi-select.
3. Always show limits for both chart engines; enforce caps in UI + API.
4. Require / strongly prompt Above/Below for risk claims.
5. One-click Chart from a risk row to inspect.

---

API map (confirmed)

Concern	API
Chart forecast	POST /api/analytics/forecast
Forecast sentence	POST /api/analytics/chat (time-to-limit, next-period total)
Forecast limits	GET /api/analytics/forecast-limits
Fleet exceedance + risk	POST /api/analytics/fleet-forecast-risk
Threshold / anomaly summaries	POST /api/analytics/anomaly-summary, threshold-breach
Alarm ops	POST /api/analytics/alarm-ops-summary
Asset rollup	POST /api/analytics/asset-rollup

---

Capability gaps (honest)

Area	Autoconnecto now	Next
Rich explorer + viz	Series + hour×weekday heatmap	Calendar heatmaps, more view types
Calculated fields product	Ingest profile CF	Reprocess / richer funcs
Trained anomaly / prediction	Z-score + StatsForecast worker (optional) + seasonal/linear fallback	SPC Nelson rules; anomaly score telemetry
AI assistant	Ask: a sentence matched to this tenant's types and keys	Open language model
Exceedance-first fleet ranking	Yes (this profile)	Max excursion, chattering index

---

Change control

Any new Analytics feature must state which layer it belongs to and whether it is:

• backward-compatible
• dual-compat
• breaking

Update this doc + Analytics hub in the same PR. Rollback = feature flag / env kill switch where applicable.

---

Related

• Analytics hub
• Field validation
• Validation status (hardware bench — separate from this software profile)