Study architecture
Select experimental, observational, survey, simulation, qualitative, mixed-method or computational design according to the research question.
WORKFLOW STAGE 02
Method & evidence.
Translate each objective into a transparent study design with appropriate data, controls, baselines, evaluation metrics and reproducible implementation.
Study DesignData ProtocolBaselinesReproducibility
Design the test before running it.
The methodology must make the proposed claims testable and separate real improvements from noise, leakage or unfair comparison.
Sampling & provenance
Document source, population, eligibility, sample size, consent or licence, missingness, class distribution and representativeness.
Preprocessing
Specify cleaning, normalization, augmentation, feature engineering and train/validation/test separation. Fit transformations on training data only.
Baselines & controls
Include standard methods, recent strong baselines, ablations and appropriate control conditions under identical data and evaluation settings.
Metrics & statistics
Predefine primary and secondary outcomes, effect sizes, confidence intervals, statistical tests and multiple-comparison correction where required.
Execution record
Save code, seeds, versions, hardware, hyperparameters, stopping rules, checkpoints and configuration files.
Method details by research type.
Technical checks change with the evidence source.
| AI / machine learning | Leakage-free splits, tuning only on validation data, class imbalance, calibration, ablation, external testing and compute reporting. |
|---|---|
| Experiments | Controls, randomization, replication, instrument calibration, environmental conditions, uncertainty and measurement error. |
| Simulation / CFD | Governing equations, boundary conditions, solver settings, convergence, mesh independence and experimental or benchmark validation. |
| Surveys / SEM | Sampling frame, power, instrument validity, reliability, common-method bias, model fit and discriminant validity. |
| Time series | Chronological splits, rolling evaluation, seasonality, leakage prevention and comparison with naive/statistical baselines. |
| Qualitative studies | Sampling rationale, interview protocol, coding framework, reflexivity, saturation and trustworthiness procedures. |
Avoid: selecting metrics after seeing results, testing on training data, hiding failed runs, tuning baselines less carefully than the proposed method, or describing implementation without enough detail to reproduce it.
Need a technically defensible methodology?
Share your objectives, data and target contribution for a focused method review.