Industrial Data Product · Synthetic French Network

CosmaOps Data Trust & Risk Lab

Manufacturing, Quality, EHS and environmental performance become decision-ready only when the underlying information systems are decision-ready too.

A reproducible synthetic-data case study connecting MES, QMS, EHS and EMS source contracts to formula-driven KPIs, decision confidence, cross-domain intervention priorities and scenario economics across a French cosmetics manufacturing network.

8 synthetic sites16 production lines18 months3,456 data-quality controlsExcel + Python + SQLISA-95-aligned architecture
MES / QMS / EHS / EMS
Data contracts
KPI mart
Decision confidence
Risk priority
Intervention simulator
Executive KPI Control Tower

Twelve indicators, one decision surface

Latest month: June 2026. Numbers use tabular numerals and consistent formatting throughout the site.

Network OEE
79.6%
Availability × Performance × Quality
Right First Time
96.1%
First-pass batches / total
Median Cpk
1.36
Process capability, network
TF 12m
6.10
Lost-time accidents × 1e6 / hours
TG 12m
0.084
Days lost × 1,000 / hours
Decision Confidence
96.8%
Weighted MES/QMS/EHS/EMS
Decision-Grade Lines
50.0%
≥95% weighted, all sources ≥95%
P1 / P2 Lines
2
Requiring immediate intervention
Failed Data Contracts
4
Site × domain, June 2026
Annualised Value Loss
€41.84m
Synthetic economic sensitivity
Monthly Good Units
40.8m
40,762,544 units, June 2026
Top Priority Score
52.4
FR03-L2, dominant capability risk

The network average is stable; the decision risk is local. Two lines require P2 intervention, while half of the network still falls short of the stricter decision-grade evidence threshold.

Analytical Chain

From source-system evidence to a defensible intervention decision

Six deterministic stages. Each stage exposes typed data and reproducible outputs. Validation labels below refer to internal project QA, not external certification.

P0
Validated

Source governance

MES, QMS, EHS and EMS owners, dependencies and SLAs mapped.

P1
Validated

Data contracts

Completeness, validity, consistency, timeliness, uniqueness and lineage.

P2
Validated

KPI mart

OEE, RFT, Cpk, TF/TG, CAPA, energy, water and waste intensity.

P3
Validated

Decision confidence

Fitness-for-use per decision with minimum-source floor.

P4
Validated

Risk radar

Weighted multi-domain risk, dominant risk, concurrence multiplier.

P5
Validated

Intervention simulator

Benefit, payback, residual risk, confidence uplift.

Performance & Quality

Trends, value loss and site-level accountability

Monthly network trends and the site scorecard together separate systemic movement from localised issues.

Monthly network trends

OEE · RFT · Confidence

Value loss & P1/P2 lines

Synthetic sensitivity
Formula
OEE
Availability × Performance × Quality.
Formula
Right First Time
first-pass batches / total batches.
Formula
Cpk
minimum distance to specification / 3σ.

FR03-L2 is not an average-network problem. It is a local capability case: Cpk 0.54, RFT 86.8%, monthly value loss €195k and intervention score 52.4.

Site scorecard — click any column header to sort8 sites
SiteOEERFTMedian CpkDecision conf.Priority scoreAnnualised lossDominant action
FR0378.1%95.3%1.0195.6%52.43€4.65mRestore process capability and CAPA flow
FR0675.5%99.0%1.3396.5%35.98€6.93mStabilise equipment and changeovers
FR0282.6%96.5%1.3696.6%27.80€4.01mClose high-potential EHS actions
FR0881.1%96.6%1.2297.8%24.80€5.35mClose high-potential EHS actions
FR0579.0%93.3%1.3596.9%18.89€5.55mClose high-potential EHS actions
FR0779.2%97.4%1.2895.8%16.83€5.96mValidate data before operational action
FR0180.0%94.4%1.4497.9%5.23€4.77mClose high-potential EHS actions
FR0481.1%95.7%1.4697.5%3.71€4.61mClose high-potential EHS actions
Data Trust

A KPI is decision-grade only when its evidence is decision-grade

Six data-quality dimensions, transparent weights and a source-floor gate. Without decision-grade evidence, an operational escalation is a validation task first.

Completeness
20%
Validity
20%
Consistency
20%
Timeliness
15%
Uniqueness
10%
Lineage
15%

Data-contract heatmap · June 2026

Fail Warn Pass Strong
MES
QMS
EHS
EMS
FR01
97.7%PASS
97.9%PASS
97.9%PASS
97.8%PASS
FR02
97.3%PASS
97.5%PASS
94.0%FAIL
98.0%PASS
FR03
97.6%PASS
92.0%FAIL
97.1%PASS
97.2%PASS
FR04
97.1%PASS
98.0%PASS
97.2%PASS
98.0%PASS
FR05
96.7%PASS
95.7%PASS
97.7%PASS
98.4%PASS
FR06
92.8%FAIL
97.7%PASS
97.8%PASS
98.9%PASS
FR07
98.5%PASS
97.1%PASS
97.4%PASS
85.8%FAIL
FR08
97.9%PASS
98.3%PASS
97.3%PASS
97.6%PASS
Confidence logic
  • Manufacturing = 75% MES + 25% QMS
  • Quality = 20% MES + 80% QMS
  • Network = 30% Manuf. + 30% Quality + 25% EHS + 15% EMS
  • Decision-grade ⇔ weighted ≥95% AND every required source ≥95%
  • If min-source < 90% → "Validate data before operational action"
Case · FR07-L1

FR07-L1 shows why sequence matters. At 95.8% network confidence, a weak EMS contract makes environmental variance a validation case before it becomes an operational escalation.

Cross-domain Risk Radar

Line-level prioritisation, not a generic network programme

Filter, search and toggle between table and chart. Ranking is driven by a weighted multi-domain score with a concurrence multiplier.

Priority
Confidence
Search
#LineSiteFamilyOEERFTCpkTF12TG12Dec. conf.ClassValue lossScorePriorityDominant riskDecision gate
1FR03-L2FR03Lip & Eye76.7%86.8%0.540.000.00095.6%Guarded€195k52.43P2Restore process capability and CAPA flowRestore process capability and CAPA flow
2FR06-L1FR06Shampoo & Care67.3%100.0%1.220.000.00096.5%Guarded€374k35.98P2Stabilise equipment and changeoversStabilise equipment and changeovers
3FR02-L2FR02Cleansers82.0%95.9%1.4438.840.27296.6%Guarded€184k27.80P3Close high-potential EHS actionsClose high-potential EHS actions
4FR08-L2FR08Lip & Eye79.9%93.6%1.0819.650.23697.8%Decision-grade€257k24.80P3Close high-potential EHS actionsClose high-potential EHS actions
5FR05-L1FR05Serums80.2%95.5%1.2239.300.84596.9%Decision-grade€170k18.89MonitorClose high-potential EHS actionsClose high-potential EHS actions
6FR07-L1FR07Face Care83.0%98.4%1.350.000.00095.8%Guarded€213k16.83MonitorInvestigate resource-intensity varianceValidate data before operational action
7FR07-L2FR07Cleansers75.4%96.2%1.210.000.00095.8%Guarded€284k14.67MonitorStabilise equipment and changeoversValidate data before operational action
8FR05-L2FR05Dermo Care77.8%91.3%1.480.000.00096.9%Decision-grade€293k10.71MonitorRestore process capability and CAPA flowRestore process capability and CAPA flow
9FR08-L1FR08Foundation82.3%98.6%1.360.000.00097.8%Decision-grade€189k6.20MonitorInvestigate resource-intensity varianceInvestigate resource-intensity variance
10FR01-L1FR01Shampoo & Care78.9%94.2%1.510.000.00097.9%Decision-grade€224k5.23MonitorClose high-potential EHS actionsClose high-potential EHS actions
11FR06-L2FR06Hair Color83.4%98.2%1.440.000.00096.5%Guarded€203k4.66MonitorInvestigate resource-intensity varianceInvestigate resource-intensity variance
12FR03-L1FR03Foundation79.5%100.0%1.480.000.00095.6%Guarded€192k4.20MonitorInvestigate resource-intensity varianceInvestigate resource-intensity variance
13FR02-L1FR02Face Care83.2%97.3%1.270.000.00096.6%Guarded€151k3.87MonitorRestore process capability and CAPA flowRestore process capability and CAPA flow
14FR04-L1FR04Fine Fragrance82.5%94.4%1.560.000.00097.5%Decision-grade€175k3.71MonitorClose high-potential EHS actionsClose high-potential EHS actions
15FR04-L2FR04Gift Sets79.7%96.8%1.360.000.00097.5%Decision-grade€208k3.26MonitorInvestigate resource-intensity varianceInvestigate resource-intensity variance
16FR01-L2FR01Hair Color81.1%94.7%1.380.000.00097.9%Decision-grade€174k3.09MonitorClose high-potential EHS actionsClose high-potential EHS actions

The five highest-loss lines represent 33.8% of current monthly value loss. Prioritisation should be line-specific, not a generic network programme.

Intervention Simulator

Levers, benefits, payback and residual risk

Adjust the five intervention levers around the validated scenario. Outputs are deterministic sensitivities, not forecasts or approved business cases.

Illustrative decision-support logic. Scenario outputs are deterministic sensitivities, not forecasts, approved business cases or operational commitments.

Scenario
Intervention levers
Downtime reduction2.0%
Scrap reduction35.0%
EHS overdue reduction0.0%
Energy reduction0.0%
Data-quality uplift2.0pp
Annual benefit
€139k
Interactive sensitivity
Payback
13.8 mo
Investment €160k
Recovered units
51,146
Annual
Scrap avoided
187,103
Annual units
Priority score
52.4 → 38.2
Decision confidence
95.6% → 97.6%
Interpretation

Value case driven by recoverable capacity and quality loss.

Portfolio comparison — validated scenarios

Benefit (k€) · Residual score
Executive Insights

Six findings, six decisions

Findings are stated in engineering terms. Decisions translate them into what a leadership team should sequence next.

01
I01

Operational loss is concentrated

Finding

The five highest-loss lines represent 33.8% of current monthly value loss.

Decision

Prioritise line-level interventions instead of a network-wide generic programme.

02
I02

Quality drift is economically material

Finding

FR03-L2 combines Cpk 0.54, RFT 86.8% and an intervention score of 52.4.

Decision

Stabilise the process window and accelerate repeat-deviation CAPA closure.

03
I03

EHS leading indicators precede lagging outcomes

Finding

FR02-L2 shows a high action backlog and TF12 of 38.84.

Decision

Use overdue high-potential actions as the intervention trigger, not near-miss volume alone.

04
I04

Data confidence changes the decision sequence

Finding

FR07-L1 has 95.8% decision confidence; environmental variance should be validated before operational escalation.

Decision

Restore EMS lineage and completeness, then reassess resource intensity.

05
I05

The simulator separates value and risk cases

Finding

Predictive maintenance sprint has the shortest modelled payback at 2.9 months, while EHS/data scenarios primarily reduce decision risk.

Decision

Govern the portfolio with two lenses: financial return and control-risk reduction.

06
I06

Decision-grade coverage remains incomplete

Finding

Only 50.0% of lines meet the 95% decision-confidence threshold in the latest month.

Decision

Treat data contracts as a product backlog with named owners, SLAs and measurable remediation value.

Engineering Architecture

Excel · Python · SQL — one deliverable, three lenses

A formula-first Excel control model, a reproducible Python pipeline and a PostgreSQL-compatible SQL product, all aligned to ISA-95 source, calculation, decision and control layers.

Excel control model

  • 14 sheets across scope, calculation and QA.
  • 38,412 formula cells across the workbook.
  • Assumptions, scenarios, conditional formats, charts.
  • blue hardcodes · green cross-sheet · black same-sheet · amber editable

Python pipeline

  • Fixed seed 20260716.
  • Generates masters, 288 line-month records, 3,456 DQ observations.
  • Publishes marts, JSON, charts, scenarios and automated QA.

SQL data product

  • PostgreSQL-compatible schema.
  • Dimensions, line-month fact, DQ fact.
  • KPI view, data-contract view, decision-confidence view.
  • ISA-95 source · calculation · decision · control layers.
Repository structure
config/
data/raw/
data/processed/
docs/
model/
notebooks/
qa/
sql/
src/
website_assets/
Methodology & Claim Boundaries

Inspired by public frameworks — not a certification or compliance tool

Every claim below sits inside a clear boundary. Thresholds and economic assumptions are transparent project choices.

OEE and its components follow the ISO 22400-2 conventions: availability, performance, quality. The implementation is derived from the standard's definitions but is not an assessment.

Formulas
availability
(planned time - downtime) / planned time
performance
ideal cycle time × total units / runtime
quality rate
good units / total units
oee
availability × performance × quality rate
right first time
first-pass batches / total batches
cpk
MIN((USL - mean)/(3σ), (mean - LSL)/(3σ))
tf 12m
lost-time accidents × 1,000,000 / hours worked
tg 12m
days lost × 1,000 / hours worked
data quality
weighted completeness, validity, consistency, timeliness, uniqueness and lineage
decision confidence
30% Manufacturing + 30% Quality + 25% EHS + 15% Environment dependency confidence
priority score
weighted domain risks with a cross-domain concurrence multiplier; capped at 100

The implementation is inspired by public frameworks. It is not a certification assessment, a compliance tool, an incident prediction model or an external benchmark. Thresholds and economic assumptions are transparent project choices.

QA & Reproducibility

Fifteen deterministic checks, all green

Reproducibility is a first-class output. The pipeline is seeded and every check reruns identically.

Python controls
15 / 15
Excel formula errors
0
Unique line-month keys
288
Scenarios complete
5 / 5
Synthetic claim boundary
PASS

QA checks

IDCategoryDescriptionObservedExpectedStatus
QA01ScopeSynthetic site count88PASS
QA02ScopeProduction line count1616PASS
QA03ScopeLine-month record count28816 × 18 = 288PASS
QA04KeysUnique line-month keys288288PASS
QA05ReconciliationGood + scrap equals total units288288PASS
QA06BoundsOEE within 0–100%0.639–0.878[0,1]PASS
QA07BoundsData-quality score within 0–100%0.738–1.000[0,1]PASS
QA08FormulaOEE identity reconciles0< 1e-10PASS
QA09France EHSTF calculation uses 1,000,000 hours6.1non-negativePASS
QA10France EHSTG calculation uses 1,000 hours0.084non-negativePASS
QA11ScenariosScenario count55PASS
QA12ScenariosAll scenario priority scores improve55PASS
QA13ClaimsNo real factory names usedAube, Camélia, Dune, Éclat, Iris, Nacre, Orme, Quartzsynthetic names onlyPASS
QA14ClaimsIndependent project positioning is explicitindependent synthetic-data projectrequiredPASS
QA15CompletenessNo nulls in core KPI mart00PASS
Formula reconciliation
OEE identity

Availability × Performance × Quality vs. reported OEE.

max |Δ| = 0.00e+00   (< 1e-10)
Reproducibility
  1. Install requirements.
  2. Run python src/pipeline.py.
  3. Inspect qa/qa_checks.csv.
  4. Open the advanced Excel model.
  5. Publish local JSON assets to the website.

From source-system evidence to an auditable intervention decision

CosmaOps demonstrates how industrial performance, quality, EHS, environmental intensity and data governance can be engineered as one decision product instead of five disconnected reporting streams.

Back to control tower
Independent synthetic-data analytical project · French cosmetics manufacturing context · All entities and values are invented · Methodology and limitations documented.
© 2026 CosmaOps Lab