NOTQIN Industrial Intelligence Operating Thesis 2026

What this is. The canonical thesis behind everything NOTQIN builds, sells, and learns. Frames NOTQIN not as “AI dashboards for factories” but as a data-driven decision layer that connects machine behavior → production flow → quality → logistics → maintenance → energy → business impact for Moroccan automotive suppliers.

What this binds. Existing vault assets had the pieces (UNS standard, sector packs, customer anatomy, Tier-2 sprint, PFE topics, founder curriculum) but no single doc said here is the mental model and the staged build path that ties them together. This file is that doc.

How to use it. Read once end-to-end. Then keep §4 (5-layer model) and §5 (9-stage roadmap) open as the working scaffolding. Every NOTQIN deliverable should answer: which layer does this serve, which stage does this advance, and which target question (§0) does it help a customer answer?

What it supersedes. Nothing is deleted. This doc unifies what was scattered across NOTQIN Project Management OS §1, NOTQIN GTM — Aeronautics & Automotive §2, the Customer Anatomy Pack, and Reading List — NOTQIN Founder Curriculum 2026. Treat earlier surface-level framings (“NOTQIN = MQTT broker + dashboards”) as deprecated in favor of this thesis.

0. The target questions — what Moroccan suppliers must be able to answer

The whole system is justified by whether a supplier on the AMICA Tier-2/3 scorecard can answer these six questions in under a minute, with evidence:

Which machine or process is creating the most contract risk?
Which defects are causing OEM rejection or rework?
Which line stoppages threaten OTIF delivery?
Which maintenance actions actually reduce downtime?
Which energy or process changes improve margin?
Which data proves compliance during customer audits (IATF 16949, AS9100D, CBAM, CNDP)?

If a NOTQIN deliverable does not shorten the path from question → answer → action for at least one of these, it should be challenged in review.

1. Foundation — manufacturing is a system, not a dashboard

Before any AI, mental model is ISA-95 / IEC 62264. The same five-level stack that the OT-UNS reference card and the Morocco UNS Standard are rooted in:

Level	Domain	Examples
4	ERP / business planning	orders, finance, scheduling commitments
3	MES / MOM	production execution, quality, maintenance
2	SCADA / HMI	supervisory control, alarms, recipes
1	PLCs / sensors / actuators	discrete control, IO scans
0	physical process	machines, motors, conveyors, operators, material

NOTQIN job description, in one line: connect Level 0–2 reality with Level 3–4 decisions, with the Unified Namespace as the connective tissue and the MES / business layer as the audience.

This is why every customer pitch (auto, aero, cement, agrifood) is built from the same architecture — only the Sparkplug B topic overlays, KPIs, and compliance envelope change. See packages/sector-packs/<sector>/ in the NOTQIN repo.

2. Literature stack — the only reading list that matters

Do not start with random YouTube. Build from standards, credible industry reports, and serious textbooks. The Reading List — NOTQIN Founder Curriculum 2026 is the curated long form; this section is the why-each-one-matters tagging of that list.

A. Standards and frameworks

Area	Source	Why it matters for NOTQIN
IT/OT integration	ISA-95 / IEC 62264	Defines how ERP, MES, SCADA, and shop-floor systems should communicate. Root of Morocco UNS Standard v0.1.
Manufacturing KPIs	ISO 22400	Defines KPIs for Manufacturing Operations Management — formulas, units, time behavior, user groups. Should be cited verbatim in every NOTQIN KPI definition.
Automotive quality	IATF 16949	Global automotive QMS standard, required across the supply chain. Drives the PPAP / APQP / FMEA / Control Plan / SPC data we must capture. See Appx C — Quality.
Industrial cybersecurity	NIST Smart Manufacturing Cybersecurity	Factories need security without destroying safety, reliability, latency. Anchor for our CNDP + DGSSI + IEC 62443 posture.
Smart manufacturing architecture	NIST Smart Manufacturing Systems Design and Analysis	Real-time control + data analytics across the extended enterprise. Underwrites our Stage 4 data engineering choices.

B. Industry transformation references

WEF Global Lighthouse Network — real Industry 4.0 cases at scale. Lighthouse case data is the gold standard for ROI claims in investor decks.
NIST data analytics for smart manufacturing — identifies two barriers: (i) selecting the right analytics tools and (ii) integrating them with data acquisition and decision-support systems. The second barrier is exactly NOTQIN’s wedge. In factories, the hard part is not the model — it is the integration into real decisions.

C. Books to study, in order

#	Book	Lens it gives
1	Factory Physics — Hopp & Spearman	flow, variability, bottlenecks, cycle time, WIP, throughput
2	The Goal — Goldratt	bottleneck thinking, Theory of Constraints
3	Toyota Production System — Taiichi Ohno	lean, waste, flow, jidoka, JIT
4	Statistical Quality Control — Montgomery	SPC, control charts, process capability
5	Introduction to Statistical Learning — JWHT	practical ML foundations
6	Designing Data-Intensive Applications — Kleppmann	distributed systems, pipelines, event streams, storage, consistency
7	Practical MLOps	deployment, monitoring, retraining, model governance
8	Industrial Network Security (Byres/Green)	OT security without breaking safety

Cross-reference: many of these already appear in Reading List — NOTQIN Founder Curriculum 2026; ones that don’t should be added in the next quarterly review.

3. Moroccan automotive context — why this market, why now

Morocco’s automotive sector is the strongest first market because it is export-oriented, quality-sensitive, logistics-sensitive, and increasingly EV-coupled.

Tanger Med describes its automotive platform as an ecosystem with nearly 120 automotive operators — wiring, metal, stamping, seats, interiors, plastic injection — serving Moroccan manufacturers and European assembly plants.
AMICA 2025 figures: >260 automotive suppliers · 3 OEMs · >280,000 sector jobs · 69% integration rate · 1,000,000 installed production capacity · >€15 B automotive exports.
Reuters (2024 data): Morocco’s automotive exports ≈ 157 B MAD in 2024; national target raises local sourcing from 69% → 75% by 2030.

What this implies for NOTQIN:

More local integration → more local suppliers → more quality pressure → more traceability pressure → more productivity pressure → more need for the data layer we are building.

This is the macro tailwind under the Auto Tier-2 narrow thesis. The numbers above belong on slide 2 of every investor and AMICA conversation.

4. Core mental model — five layers from physical to decision

Industrial intelligence is one chain, five layers. Every NOTQIN feature is some unit of work on this chain.

Physical process → Signal → Event → KPI → Decision → Economic outcome

Layer 1 — Physical process

Machines, lines, tools, operators, materials. Examples: injection molding, stamping press, CNC, wiring harness board, conveyor, compressor, paint booth, test bench. Question: What physical transformation is happening?

Layer 2 — Signals

Raw data from reality: cycle time, machine status, alarms, vibration, temperature, pressure, current, energy consumption, part count, reject count, operator ID, batch ID, tool ID, material lot ID. Question: Which signals represent the health, flow, quality, and cost of the process? NOTQIN layer: MQTT / Sparkplug B tags on the Unified Namespace — see OT-UNS Reference Card.

Layer 3 — Events

Signals become events: machine stopped, changeover started, defect detected, maintenance opened, batch consumed, part scrapped, shipment delayed, customer complaint received. Question: What happened, when, where, why? NOTQIN layer: the canonical event model (see §5 Stage 4 below) stored in TimescaleDB / Postgres event store.

Layer 4 — KPIs

Events become business metrics. Use ISO 22400 formulas verbatim to avoid the usual “everyone’s OEE is different” trap.

Examples: OEE (availability × performance × quality), MTBF, MTTR, scrap rate, rework rate, first-pass yield, OTIF, energy per unit, cost of poor quality, schedule adherence. Question: What does this mean economically?

Layer 5 — Decisions

KPIs and predictions drive actions: stop the line, inspect a batch, reschedule production, dispatch maintenance, reorder spares, escalate to quality manager, adjust process parameter, notify logistics, create audit evidence. Question: What should the factory do now? NOTQIN layer: NOTQIN Agent recommendations + factory-app workflow + CBAM / IATF 16949 audit packs.

The product test. Every NOTQIN screen should be tagged with the layer it serves. A screen that only renders Layer 2 signals without a path up to Layer 5 is decoration, not industrial intelligence.

5. Roadmap — 9 stages from foundation to optimization

This is the staged build path. It is both the founder curriculum (what to learn) and the product roadmap (what to ship). Maps onto the wave system in NOTQIN Project Management OS §8.

Stage 0 — Choose one narrow use case

Do not start with “build an industrial AI platform.” For Moroccan automotive suppliers, the best initial use cases are:

Downtime intelligence — which stoppages hurt production and delivery most?
Quality defect intelligence — which machine/material/shift/tool creates defects?
OEE and bottleneck intelligence — which line is constraining throughput?
Traceability and audit evidence — can the factory prove what happened for each batch/order?
Energy-per-unit intelligence — which lines or processes consume abnormal energy?

NOTQIN MVP choice: downtime + OEE + root-cause intelligence for one production line. Connects machine data, operator events, business losses, and ROI — and is what AMICA Tier-2 suppliers are scored on by Renault / Stellantis (see Appx C — Quality PPM/OTD).

Stage 1 — Learn manufacturing operations

Build an Obsidian note per concept under 02_Knowledge Base/Industrial Intelligence/Concepts/ with: definition, formula, factory example, required data, possible data sources, economic impact, common mistakes.

Concepts: OEE · takt time · cycle time · throughput · bottleneck · WIP · changeover · scrap · rework · MTBF · MTTR · preventive / predictive / corrective maintenance · line balancing · first-pass yield · production order · work order · BOM · routing · batch / lot / serial traceability.

Reference template:

OEE = Availability × Performance × Quality
Required data:    planned production time, downtime,
                  ideal cycle time, total count, good count
Economic impact:  lost output, overtime, delivery delay, scorecard risk

Stage 2 — Learn automotive quality reality

For IATF 16949 suppliers, quality is not optional. Study: PPAP · APQP · FMEA · Control Plan · SPC · MSA · 8D · customer-specific requirements · nonconformity · containment · corrective action · supplier scorecards · PPM defects · OTID/OTIF · line-stop escalation.

Deliverable — quality data map (the spine of NOTQIN’s quality module):

Defect detected → defect type → part number → machine → tool
                → operator → shift → batch/material lot
                → containment action → root cause
                → corrective action → customer impact

Cross-links: Appx C — Quality for the Renault/Stellantis-specific score formulas; Aero MRO — Wrong-Part Detection on Test Cell 2026 for the aero-side variant (AS9100D / NADCAP / AOG).

Stage 3 — Learn industrial data acquisition

Topics: PLC basics · sensors / actuators · HMI · SCADA · OPC UA · Modbus TCP · MQTT · industrial gateways · barcode scanners · edge computers · machine tags · sampling rates · event-based vs time-based · signal quality · timestamp sync · network segmentation.

Goal is not to become a full automation engineer — it is to capture reliable data without disturbing production.

Minimal architecture (canonical NOTQIN stack):

PLC / machine / sensor
  → OPC UA or Modbus gateway
  → Edge collector
  → MQTT broker (HiveMQ / EMQX) — Sparkplug B encoded
  → Stream processor
  → Time-series DB (TimescaleDB / InfluxDB)
  → Event store (Postgres)
  → Analytics layer
  → Dashboard / alert / workflow (factory-app + NOTQIN Agent)

Deliverable — a simulated line on a laptop: Python simulator → MQTT → TimescaleDB → FastAPI → dashboard. Simulating: machine status, part count, downtime reason, reject count, temperature, vibration, energy. This is already shipped in the Wave A-G simulator (cement, textile, aero, agrifood, auto). Re-use, do not rebuild.

Stage 4 — Learn data engineering for industrial systems

Topics: event-driven architecture · Kafka / Redpanda · MQTT · schema design · time-series modeling · dimensional modeling · data lake vs warehouse · stream processing · data quality rules · late-arriving data · idempotency · exactly-once vs at-least-once · feature stores · lineage · observability.

Industrial-data peculiarities to memorize: sensors lie · operators forget · machines change state fast · PLC tags are badly named · timestamps drift · data is missing · ERP data arrives late · downtime reasons are political.

Canonical event model (every NOTQIN sector pack must conform):

{
  "event_id": "uuid",
  "site_id": "factory-01",
  "line_id": "line-01",
  "machine_id": "press-03",
  "event_type": "machine_stopped",
  "timestamp_start": "...",
  "timestamp_end":   "...",
  "duration_seconds": 420,
  "production_order_id": "PO-123",
  "part_number": "PN-456",
  "shift_id": "night",
  "operator_id": "op-07",
  "reason_code": "tool_jam",
  "source": "plc",
  "confidence": 0.92
}

Star schema (fact / dim split):

facts: fact_machine_state, fact_downtime_event, fact_quality_event, fact_production_count, fact_energy_consumption
dims: dim_machine, dim_line, dim_part, dim_operator, dim_shift, dim_tool, dim_material_lot, dim_supplier

Stage 5 — Analytics before AI

Most teams fail here by jumping to ML too early. NOTQIN’s discipline: descriptive → diagnostic → predictive → prescriptive, in that order.

Descriptive (“what happened?“): downtime by line, scrap by part, OEE by shift, defects by supplier batch, MTTR by maintenance team.
Diagnostic (“why?“): defect spike after tool change, downtime concentrated on one machine, energy abnormal during idle periods, night-shift restart drag.

Tools: SQL · Pandas · Grafana / Superset / Power BI · SPC · Pareto · correlation · cohort · control charts.

Baseline dashboards NOTQIN must ship before claiming AI:

Production flow
Downtime Pareto
OEE
Quality defects
Maintenance reliability
Energy per unit
Customer scorecard risk

Stage 6 — Statistical process control

SPC is more credible in factories than random ML. Topics: mean / variance / SD · control limits · X-bar · R · p · c · Cp/Cpk · out-of-control rules · MSA.

Use SPC for: process drift detection, quality stability, abnormal cycle times, abnormal energy consumption, defect-rate monitoring.

Deliverable — notebook detecting: cycle-time drift · defect-rate shift · energy-per-part anomaly · machine-temperature instability. Run control charts before reaching for deep learning.

Stage 7 — Predictive maintenance, honestly

Predictive maintenance is attractive and easy to oversell. Deloitte frames it as connecting machines to reliability professionals and using advance insights to avoid failures.

Honest language NOTQIN uses:

“We identify failure patterns earlier, prioritize high-risk assets, and measure impact through controlled baselines.”

NOT “we reduce downtime by 50%” unless we have evidence.

Topics: failure modes · condition monitoring · vibration analysis basics · thermography basics · current signature analysis · remaining useful life · anomaly detection · supervised classification · survival analysis · maintenance work orders · CMMS integration · false positives / negatives · alert fatigue.

Public datasets for learning only: NASA turbofan, bearing vibration, hydraulic system, SECOM, UCI predictive maintenance. Real value requires local machine context, real maintenance history, and labeled failures. Sales pitch must reflect this.

Stage 8 — Optimization

Industrial intelligence becomes powerful when it recommends actions. Topics: linear / integer / constraint programming · scheduling · inventory optimization · reorder points · safety stock · queueing theory · simulation · digital twins.

Use cases: production scheduling, spare-parts planning, maintenance scheduling, workforce assignment, energy peak shaving, delivery-risk planning.

Deliverable — NOTQIN Optimizer module:

Input:  production orders, machine capacity, changeover times,
        due dates, maintenance windows
Output: recommended schedule, predicted late orders,
        bottleneck line, overtime requirement

Stage 9 — Causal thinking

Factories are full of misleading correlations. “Night shift has more defects” may really be: harder orders at night · older machine at night · maintenance team absent · humidity · different material lot.

Topics: causal graphs · confounding · interventions · A/B where possible · diff-in-diff · root-cause analysis · 5 Whys · Ishikawa · DoWhy / EconML basics.

NOTQIN rule — for every detected problem, surface a causal hypothesis card in the UI, not a verdict:

Observed:       Defects increased after 22:00.
Possible causes: operator skill, machine temp, material lot,
                 humidity, maintenance condition, product mix.
Data needed:    shift roster, machine state, environmental
                data, material batch, part number, tool condition.

This is what separates NOTQIN from black-box “AI alerts” that lose customer trust on the third false positive.

6. The MVP — downtime intelligence for one line

The first serious project. Mapped to the Auto Tier-2 sprint.

Problem

Moroccan automotive suppliers know downtime is expensive but lack reliable, granular, structured evidence:

Which stoppages happen most?
Which stoppages cost most?
Which machine is the bottleneck?
Which downtime reasons are real vs guessed?
Which interventions reduce recurrence?
Which events threaten customer delivery?

Minimum data required

Machine status (running / stopped / idle / fault) · timestamp · line ID · machine ID · production order · part number · good count · reject count · downtime reason code · operator / shift · maintenance action.

Optional but high-leverage: energy · vibration · temperature · tool ID · material lot · customer order · delivery due date.

This minimum is deliberately small so it can be captured on a single line in a MOWAKABA-subsidized pilot in 6–8 weeks, billed in MAD, hosted in-country per CNDP / DGSSI.

7. 90-day success metrics — operational, not vanity

What we measure in the first 90 days of a pilot. No “monthly active users” — operational evidence only.

Baseline (Day 0–30):

% machine states automatically captured
% downtime events with reason codes
Average time to classify downtime
Top recurring downtime causes
Baseline OEE, MTBF, MTTR
Lost production minutes / units / estimated MAD value

Improvement (Day 30–90):

Reduction in repeat downtime
Reduction in unclassified stoppages
Reduction in mean repair time
Improvement in schedule adherence
Reduction in scrap / rework

These become the case-study numbers for customer #2. Cross-ref NOTQIN Pilot Playbook for the contractual side.

8. The Moroccan supplier problem map — five wedges

Strong thesis areas NOTQIN can lead with, ranked by Phase-1 priority:

A. Traceability (audit-driven, immediately fundable)

Problem: customer audits, batch genealogy, defect containment, PPAP / IATF 16949 evidence.
NOTQIN solution: part- or batch-level genealogy · machine / operator / material / tool traceability · audit-ready evidence package.
Targets: AMICA Tier-2/3, aero MRO (Aero MRO — Wrong-Part Detection on Test Cell 2026).

B. Supplier scorecard protection (loss-aversion, fast yes)

Problem: OEMs evaluate quality, delivery, responsiveness; a bad month costs the next program.
NOTQIN solution: early OTIF warning · PPM tracking · corrective-action tracking · delivery-risk forecasting.
Targets: Renault / Stellantis Tier-1 and Tier-2 in Tanger Med AFZ Kenitra.

C. Downtime and bottlenecks (highest ROI, our MVP)

Problem: hidden lost capacity · manual downtime logs · unclear root causes.
NOTQIN solution: automatic event capture · Pareto · repeated-failure detection · maintenance prioritization.
Targets: every Phase-1 customer.

D. Quality defects (scoped under §B/C)

Problem: defects discovered late · root cause unclear · weak SPC.
NOTQIN solution: defect correlation with machine / tool / material / shift · control charts · abnormal process detection.

E. Energy and cost (CBAM-aligned, Phase 2)

Problem: rising energy cost · idle waste · compressed-air leaks.
NOTQIN solution: energy per unit · idle consumption · abnormal energy signatures · peak-load alerts.
Targets: CBAM-exposed verticals — Sonasid, Managem Tizert, LafargeHolcim, OCP Safi via OCP Group. NOT textile.

9. How this thesis integrates with the rest of the vault

Existing artefact	Role in this thesis
Morocco UNS Standard v0.1 + v0.2 + 6 sector overlays	The Stage-3/4 data backbone — this thesis says why it exists in commercial language.
OT-UNS Reference Card	One-page glossary for Layers 1–3 of the 5-layer model. Hand to any new hire / customer engineer on day 1.
ISA-95 — Distilled · HiveMQ MQTT Essentials — Distilled	The Stage-1/3 founder-curriculum digests already exist — re-tag as required reading for this thesis.
Reading List — NOTQIN Founder Curriculum 2026	Long-form Stage 1–9 curriculum. §2 of this doc is the why-each-book-matters summary.
PFE Research Dossier — Management de Projet et SI	16 PFE topics map onto Stages 4–9. Use this thesis as the framing of the PFE viva.
Master — Customer Anatomy + Appx A-E	Stage-2 (quality), Stage-5 (analytics targets), Stage-7 (decision audience) all source from here.
Feasibility Study — NOTQIN Auto Tier-2 Scorecard 2026 + AMICA Tier-2 Target List 2026	The Stage-0 use-case decision and the Stage-6/7 90-day target customers.
Aero MRO — Wrong-Part Detection on Test Cell 2026	The Phase-3 sector variant of the §8.A traceability wedge.
NOTQIN Project Management OS §8 (Wave system)	Each wave should be tagged with the stage(s) it advances.
NOTQIN Go-to-Market OS + NOTQIN GTM — Aeronautics & Automotive	The §0 target questions become the cold-email opening lines and the demo script.
Sector packs (`packages/sector-packs/*` in repo)	Each pack = one Stage-3/4 instantiation of this thesis for a specific vertical.

Editing rule. When a new vault note touches industrial intelligence, link back here and tag which layer (§4) and stage (§5) it advances. When this thesis is edited, scan the table above and reconcile.

Action items

Open items

Which Stage-0 use case wins commercially? Thesis says downtime+OEE, but the Tier-2 sprint may surface traceability (§8.A) as the faster yes. Decide after H3–H7 interviews.
Do we cite ISO 22400 explicitly in customer SOWs? Pro: credibility, audit-friendly. Con: locks formula choices early.
Public datasets for Stage 7 demos — do we use them in sales demos at all? Risk: prospects see through “NASA turbofan” and lose trust. Need a Moroccan-context synthetic dataset (sector simulator may suffice).
Where does the NOTQIN Agent (MCP-backed) sit in this thesis? Currently implicit in Stage 5+9. Should it get its own §10 once the Agent’s role solidifies past Wave G?
Energy / CBAM wedge timing — §8.E is parked as Phase 2, but YNNA Steel CBAM deadline (Jan 2026) may force pulling it forward.

NOTQIN — OT Track

Explorer

NOTQIN Industrial Intelligence Operating Thesis 2026

NOTQIN Industrial Intelligence Operating Thesis 2026

0. The target questions — what Moroccan suppliers must be able to answer

1. Foundation — manufacturing is a system, not a dashboard

2. Literature stack — the only reading list that matters

A. Standards and frameworks

B. Industry transformation references

C. Books to study, in order

3. Moroccan automotive context — why this market, why now

4. Core mental model — five layers from physical to decision

Layer 1 — Physical process

Layer 2 — Signals

Layer 3 — Events

Layer 4 — KPIs

Layer 5 — Decisions

5. Roadmap — 9 stages from foundation to optimization

Stage 0 — Choose one narrow use case

Stage 1 — Learn manufacturing operations

Stage 2 — Learn automotive quality reality

Stage 3 — Learn industrial data acquisition

Stage 4 — Learn data engineering for industrial systems

Stage 5 — Analytics before AI

Stage 6 — Statistical process control

Stage 7 — Predictive maintenance, honestly

Stage 8 — Optimization

Stage 9 — Causal thinking

6. The MVP — downtime intelligence for one line

Problem

Minimum data required

7. 90-day success metrics — operational, not vanity

8. The Moroccan supplier problem map — five wedges

A. Traceability (audit-driven, immediately fundable)

B. Supplier scorecard protection (loss-aversion, fast yes)

C. Downtime and bottlenecks (highest ROI, our MVP)

D. Quality defects (scoped under §B/C)

E. Energy and cost (CBAM-aligned, Phase 2)

9. How this thesis integrates with the rest of the vault

Action items

Open items

See also

Graph View

Table of Contents

Backlinks