Thesis alignment

The final-year report supplies the research motivation: small-data supply-chain forecasting, external weather variables, interval type-2 fuzzy uncertainty, evolutionary optimization, and human-readable rules. The rebuilt system retains those testable ideas while correcting the prototype's methodological gaps.

Report or prototype statement	Repository treatment
2016 training and 2017 validation	Implemented as a strict year holdout
Real weather as external information	Canonical target-date weather interface implemented
Prototype-generated random weather	Removed
Full-data normalization	Replaced by training-only preprocessing
GA plus Pattern Search	Implemented in MATLAB and independently in the labeled Python IT2 backend
ARIMA/LSTM comparisons	Implemented as shared-split baselines
Interpretable fuzzy rules and heat map	Exported as rule and activation artifacts
Online model evolution	Not implemented; future work
Typhoon robustness and 3–5 day warning	Not established by available experiments
Six planning granularities	Not implemented
Causal weather effects	Not claimed; associations only

The original PDF is intentionally not included in the public repository. This page is the public traceability record between its concepts and current executable evidence.

The legacy prototype files are also not treated as reproduced evidence: they used combined-year normalization, synthetic weather, a non-independent split, and did not retain a completed fuzzy model artifact. They remain private historical material rather than inputs to published metrics.