The four samples
Each card links to the annotated page for that assignment, where two or three passages are quoted and the reasoning behind them written out.
Process Safety
Model answer
Overfill Protection in Bulk Hydrocarbon Storage: A Critical Evaluation of HAZOP and Layer of Protection Analysis
A 2,720-word MSc process safety essay that asks why structured hazard analysis failed at two sites that had it. Three annotated extracts show how to turn a method description into an argument about where the method breaks.
- Critique generated from the student's own worked HAZOP table
- Swiss-cheese model used to do work, not as decoration
- Quantitative ALARP anchors, correctly labelled as guidance
Process Control
Model answer
Temperature Control of an Exothermic Jacketed CSTR: PID Tuning, Cascade Structures and Instrumentation Selection
A 2,330-word MSc control report that compares three tuning methods by simulation and then argues that tuning constants matter less than structure and instrumentation. Three annotated extracts show how to make a comparison produce a recommendation.
- Comparison converted into a transferable decision rule
- Cascade introduced as the answer to a demonstrated weakness
- States when its own FOPDT model would not hold
Nanomaterials
Model answer
From Flask to Plant: Engineering Challenges in Scaling Up the Synthesis of Inorganic Nanoparticles
A 2,370-word MSc essay arguing that nanoparticle scale-up fails for engineering reasons, not chemical ones. Three annotated extracts show how to locate a single controlling variable and build an argument on it.
- One controlling variable identified, then everything judged against it
- Mechanism traced to the symptom the engineer actually sees
- Names a standard heuristic and scopes where it fails
Energy Engineering
Model answer
Low-Carbon Hydrogen Production Pathways: A Comparative Techno-Environmental Assessment
A 2,300-word MSc comparative assessment of blue, green and biomass-derived hydrogen that handles a live academic disagreement without ducking it. Three annotated extracts show how to weigh contested evidence.
- Handles a live academic disagreement by locating its assumptions
- Conclusions are conditional, and the conditions pick the technology
- Attacks its own comparative framing before a marker can
Industrial Automation
Model answer
Integrating Industrial Robots and IEC 61131-3 PLC Control in Industry 4.0 Manufacturing Cells
A 2,317-word MSc technical report on PLC–robot integration that refuses the easy conclusion about collaborative robotics. Three annotated extracts show how to qualify an industry claim with evidence rather than enthusiasm.
- Bounds the fashionable claim instead of repeating it
- Cites Stuxnet for the mechanism it proved, not its fame
- Ends in a committed design, not a summary
Applied AI
Model answer
Deep Reinforcement Learning for Robotic Manipulation: Algorithms, Sim-to-Real Transfer and Safe Deployment
A 2,181-word MSc critical literature review that organises the field around one unsolved problem rather than by chronology. Three annotated extracts show the difference between reviewing a literature and listing it.
- Organised by open problem rather than by paper
- Names the hidden engineering judgement in domain randomisation
- Treats safety as structural to RL, not an ethics footnote
Machine Learning
Model answer
Convolutional Neural Networks versus Vision Transformers for Object Detection in Mobile Robot Perception
A 2,097-word MSc essay that compares two architectures and then questions whether the benchmark everyone cites means anything for a robot. Three annotated extracts show how to interrogate your own evidence base.
- Questions whether its own benchmark measures the right thing
- Converts 100 ms of latency into 15 cm of travel
- Comparison table whose first row explains the others
Control Engineering
Model answer
State-Space Modelling and Linear Quadratic Regulator Design for an Inverted Pendulum on a Cart
A 2,255-word MSc design report on the classic inverted pendulum that explains why the textbook problem resists the textbook controller. Three annotated extracts show how to justify a design choice rather than assert it.
- Shows PID's limit here is architectural, not a tuning failure
- Spends an earlier result to license a later design choice
- Gives every mathematical result its physical meaning
Public Health
Model answer
Tackling Health Inequalities through the Social Determinants of Health: Upstream Public Health Action in England
A 2,521-word MPH essay that tests a theory against one real policy rather than describing both. Three annotated extracts show how to evaluate an intervention you broadly support.
- Criticises the evidence for its own position, on method
- Narrows to one policy and evaluates it properly
- Thesis derived from evidence and hedged to match it
Epidemiology
Model answer
From Association to Causation: A Critical Evaluation of Epidemiological Study Designs
A 2,314-word MPH critical essay on the question epidemiology exists to answer. Three annotated extracts show how to use a historical case to teach a method rather than to tell a story.
- Uses the most familiar case in the field to teach method
- Designs introduced as answers to specific objections
- Treats Bradford Hill as historical, not current practice
Health Economics
Model answer
Financing Universal Health Coverage: A Comparative Analysis of the Health Systems of the United Kingdom and India
A 2,333-word MPH comparative report on two health systems that could hardly be less alike. Three annotated extracts show how a shared framework makes an unequal comparison fair.
- Compares functions, not outcomes, so the comparison is fair
- Decomposing coverage produces a finding aggregates hide
- Defends its choice of comparators on structural grounds
Health Leadership
Model answer
Leading Complex Change in Public Health: Leadership Theory Applied to the UK COVID-19 Vaccination Programme
A 2,321-word MPH case study that applies seven leadership theories to one programme and concludes that none of them is sufficient. Three annotated extracts show how to make theory answer to evidence instead of the reverse.
- Reaches a negative finding and treats it as the result
- Tests theory against the case rather than illustrating it
- Analyses the failure that stresses its own conclusion
Database Systems
Model answer
Designing a Secure Relational Database for a Private Outpatient Clinic: Data Modelling, Threat Analysis and Defence in Depth
An MSc Data Science report that designs a clinical database and then argues against its own obvious optimisation. Three annotated extracts show how security reasoning changes a design decision.
- Rejects a performance optimisation on security grounds
- STRIDE threat model scored and specific to this system
- Separates technical controls from organisational ones
Machine Learning
Model answer
Interpretability versus Accuracy in Clinical Machine Learning: Comparing Logistic Regression, SVMs and Random Forests
An MSc Data Science report that recommends the model which did not win on accuracy, and justifies it properly. Three annotated extracts show how to let context override a leaderboard.
- Recommends against the best-scoring model, and proves nothing is lost
- Treats classification as a choice of operating point
- Closes the data-leakage route that would void the comparison
Data Visualisation
Model answer
Clarity and Integrity in Data Visualisation: Redesigning a Healthcare Performance Chart
An MSc essay that redesigns a real-world chart and then asks what the redesign still hides. Three annotated extracts show how to engage with a canonical theory instead of reciting it.
- Treats Tufte as contested, and explains why the dispute persists
- Ethics section turns on its own redesign, not a textbook example
- Declares data provenance before critiquing
Big Data and Cloud
Model answer
A Cloud Big Data Architecture for a UK Grocery Retailer: Evaluating Distributed Processing Frameworks and Service Models
An MSc report that challenges the assumption cloud is cheaper, and shows when it is not. Three annotated extracts show how to evaluate an architecture against a specific business, not in general.
- Makes 'cloud is cheaper' conditional, and names the condition
- Evaluates against three named workloads, so it can decide
- Cost critique feeds back into the proposed design