Temperature, vibration, pressure, speed, and alarm readings at uneven sampling intervals.
MISSION 02 · INDUSTRIAL ANALYTICS
IN PRODUCTION
Northstar
Manufacturing
Build an investigation system that unifies production evidence, accelerates defect analysis, and communicates causal uncertainty honestly.
01 / CORE QUESTION
What is driving the defect increase—and what should the plant do?
Northstar produces precision industrial components. Surface defects have increased across one production line, but quality engineers, maintenance supervisors, operators, and procurement teams disagree about the cause.
The mission does not ask the learner to identify one predetermined root cause. It asks them to create a reproducible investigation workflow, distinguish correlation from evidence of causation, and recommend the next operational action with explicit uncertainty.
DECISION TO IMPROVEWhich production factor should the plant investigate or change first, and what evidence justifies that priority?
02 / EVIDENCE PLAN
Production truth must be reconstructed.
The evidence room will force the learner to reconcile inconsistent identifiers, timestamps, units, and maintenance history before drawing conclusions.
Defect category, severity, inspection station, image reference, and disposition.
Work orders, component changes, downtime, calibration events, and technician notes.
Supplier, composition, received date, storage history, and production-run usage.
Crews, handoffs, training status, schedule changes, and manual interventions.
Part family, recipe, line, machine, timestamps, quantity, scrap, and rework.
Conflicting explanations from quality, maintenance, operations, and procurement.
Read-only production databases, restricted network access, and role-based visibility.
EXPECTED DATA DEFECTSClock driftUnit mismatchMissing calibrationDuplicate run IDsLate inspection resultsManual corrections
03 / REQUIRED BUILD
A reproducible defect investigation workflow.
The system should make evidence easier to inspect without pretending that a dashboard can prove causation by itself.
Canonical production-run model
Connect machine, material, shift, maintenance, and inspection evidence to one traceable unit of analysis.
Data-quality pipeline
Validate units, time alignment, identifier matching, missingness, and calibration boundaries.
Investigation dashboard
Compare runs, filter hypotheses, inspect source records, and preserve original measurements.
Correlation analysis
Rank associations with sample size, uncertainty, confounding warnings, and reproducible methods.
Traceability view
Show exactly which records and transformations support each investigation finding.
Investigation log
Record hypotheses, evidence reviewed, decisions, ownership, and follow-up tests.
04 / HIDDEN COMPLICATION
The most important sensor changed meaning.
The complication is released after the first analytical design checkpoint.
REVEAL PRODUCTION COMPLICATION+
A temperature sensor was recalibrated during an undocumented maintenance window. Values before and after the change are not directly comparable, and the suspected defect increase overlaps the calibration boundary.
Required response: choose and defend a treatment—separate periods, estimate a correction, exclude affected observations, or retain them with explicit uncertainty. Revise the investigation workflow, dashboard, and recommendation accordingly.
Separate periods
Preserves raw evidence but reduces sample size and trend continuity.
Estimate correction
May recover comparability but introduces model-dependent assumptions.
Exclude the sensor
Avoids false precision but may remove the strongest operational signal.
Retain with warning
Keeps context visible while requiring downstream users to handle uncertainty correctly.
05 / COMPLETION CRITERIA
The answer cannot come from one chart.
The completed mission must demonstrate analytical validity, operational usefulness, and restraint.
Multiple plausible hypotheses remain at the first checkpoint.
The learner documents data lineage and transformation logic.
Evidence quality changes the strength of the recommendation.
The dashboard exposes original records and uncertainty.
The analysis is reproducible from the submitted repository.
The recommendation distinguishes correlation from causation.
The architecture respects on-premises and access constraints.
The final plan includes a follow-up experiment or validation action.
06 / PRODUCTION STATUS
What is being built now.
This page is the public mission specification. Full learner and reviewer materials will be added in staged releases.
Operating challenge
Core decision, mission boundaries, evidence categories, and learning emphasis.
Evidence pack
Telemetry, defects, maintenance, materials, shifts, and stakeholder interviews.
Starter repository
Python data pipeline, DuckDB model, investigation notebook, and React interface shell.
Assessment package
Checkpoint prompts, pilot evidence, scoring anchors, and reviewer-only guide.
EARLY ACCESS
Pilot Northstar when the evidence room opens.
Early participants will receive the mission pack at pilot pricing and provide structured feedback on difficulty, clarity, and realism.