Module 1 capstone
A Documented, Reproducible Model
Choose a small system, design a model, build it transparently, estimate its parameters, verify and validate it, and archive it.
§ Project brief
The Module 1 capstone is a small end-to-end modelling project that exercises every skill introduced in the module. You will choose a system, classify the model, design a three-layer architecture, estimate at least one parameter from data, produce verification and validation evidence, and package the work so that another analyst could rerun it.
The project is deliberately small. The purpose is to demonstrate mastery of the process, not the sophistication of the model. A model with one parameter, carefully verified and validated, is worth more than a complex model with no verification. Choose something you understand thoroughly and can defend at every step.
Process over product
The capstone is assessed on the rigour of the modelling process and the clarity of the documentation. The model itself may be simple. The evidence that you built it well is the deliverable.
§ Learning outcomes addressed
This capstone assesses
- Distinguish observational and explanatory, deterministic and stochastic, static and dynamic models.
- State the purpose and system boundary of a model.
- Design a three-layer spreadsheet architecture.
- Estimate a model parameter from data with reported uncertainty.
- Produce verification and validation evidence.
- Package a model for reproducibility by another analyst.
§ Suggested scenarios
Choose one of the following, or propose your own. Each scenario has been selected because it is small enough to complete in the time available and rich enough to exercise the full lifecycle.
Cooling of a beverage
Model the cooling of a cup of coffee under Newton's law of cooling. Estimate the cooling rate constant from temperature data. Verify against the analytical solution. Validate by prediction on held-out observations.
Compound interest projection
Model the future value of a savings account with periodic contributions. Estimate the effective annual return from historical data. Verify against a closed-form expression. Validate against actual account balances.
Population growth of a small colony
Model exponential or logistic growth in a small population. Estimate the growth rate from observations. Verify against theory. Validate by projecting one generation forward.
Spread of a rumour in a closed group
Model the diffusion of information in a small closed group. Estimate the transmission rate from observations or assumed behaviour. Verify conservation of group size. Validate qualitatively against the expected sigmoid curve.
§ Deliverables
| Deliverable | Weight | What it should contain |
|---|---|---|
| Model specification | 15% | Purpose, boundary, classification, variables, parameters, units, assumptions, claim-support scope. |
| Spreadsheet model | 20% | Three-layer architecture with separated inputs, calculations, and outputs. Documentation sheet. |
| Parameter estimation report | 15% | Estimation method, data source, point estimate with uncertainty, residual check. |
| Verification and validation report | 25% | At least three verification tests and one validation exercise. Disconfirming evidence stated in advance. |
| Reproducibility package | 15% | Model file, data, documentation, readme, and a tested independent rerun. |
| Technical summary | 10% | One-page summary for a technical audience: what was built, what it says, what it does not say. |
§ Milestones
- Choose the system and state the question Write a one-paragraph purpose statement. One hour.
- Write the model specification Classification, boundary, variables, units, assumptions, claim-support scope. Two hours.
- Build the three-layer spreadsheet Separated inputs, calculations, and outputs, with a documentation sheet. Three hours.
- Estimate parameters Fit the parameter, compute uncertainty, check residuals. Two hours.
- Produce verification and validation evidence Three verification tests and one validation exercise with pre-stated criteria. Two hours.
- Assemble the reproducibility package Model file, data, readme, and an independent rerun test. One hour.
- Write the technical summary One page. One hour.
§ Rubric
| Criterion | Excellent | Satisfactory | Needs work |
|---|---|---|---|
| Model specification | Every element stated explicitly with justification. | Every element stated, but justification missing for some. | One or more elements missing or vague. |
| Architecture | Clean three-layer separation; no hard-coded constants; documentation complete. | Three layers present but one rule violated. | Layers mixed or documentation missing. |
| Parameter estimation | Estimate with uncertainty; residual structure examined. | Estimate with uncertainty but no residual check. | Point estimate only; no uncertainty reported. |
| Verification & validation | Multiple diverse tests; disconfirming evidence stated in advance; all criteria met. | Some tests; criteria stated but not in advance. | One test; criteria set after the result. |
| Reproducibility | Independent rerun tested and passed; documentation complete. | Documentation complete but independent rerun not tested. | Documentation missing a critical element. |
| Technical summary | Self-contained; claims supported by evidence; limitations stated. | Self-contained but some claims unsupported. | Not self-contained or claims overstated. |
§ Common pitfalls
Watch for these
- Choosing a system that is too complex. The goal is to demonstrate the process. A simple model done well is worth more than an ambitious one left incomplete.
- Skipping the specification. The specification is the foundation. Without it, nothing downstream can be justified.
- Reporting a parameter estimate with no uncertainty. A number without a spread is not a modelling result.
- Verification tests that pass trivially. A test that would pass for any input is not a test.
- Validation on calibration data. Validation requires held-out evidence.
- Claiming more than the evidence supports. The claim-support scope should follow from the verification and validation evidence, not precede it.
§ Submission
What to submit
- A single ZIP archive named
M1-capstone-yourid.zip. - The archive contains: the spreadsheet model, the model specification document, the parameter estimation report, the verification and validation report, the reproducibility package (including the readme), and the technical summary.
- Submit through the course LMS before the deadline listed on the course schedule.