A Safety-Aware Retrospective Audit of a Certainty-Factor Expert System for Formalin-Risk Screening

Abstract

Observable characteristics can support early food-safety triage, but they cannot chemically confirm formaldehyde. This study presents a safety-aware retrospective technical audit and algorithmic reconstruction of a legacy web-based certainty-factor (CF) expert system previously used as a prototype for food screening. The available evidence comprised interface artifacts, a food-category summary, legacy rule values, and one manually reconstructable fish scenario; source code, laboratory measurements, and a detailed fish calculation trace were unavailable. Expert and user CF values were multiplied for each positive evidence item and aggregated iteratively. The fish contributions of 0.42, 0.40, 0.12, and 0.28 produced an exact reconstructed score of 0.7795072, consistent only with the prototype summary value of 78%; a detailed interface-level trace for fish could not be verified. A deterministic verification suite confirmed expected behavior for empty, zero, unit, single-evidence, order-invariance, and invalid-input cases in the reconstructed reference algorithm. The audit also identified unsafe legacy remediation text and insufficient rule provenance. The revised architecture therefore separates screening from laboratory confirmation and introduces versioned rules, input snapshots, calculation lineage, and referral records. The contribution is a transparent audit framework for legacy rule-based screening systems. The CF score remains a rule-consistency measure and must not be interpreted as formalin concentration, contamination probability, predictive accuracy, or a substitute for validated chemical testing.

ABSTRAK

Karakteristik yang dapat diamati dapat mendukung triase awal keamanan pangan, tetapi tidak dapat mengonfirmasi formaldehida secara kimia. Penelitian ini menyajikan audit teknis retrospektif yang berorientasi keselamatan serta rekonstruksi algoritmik terhadap prototipe lama sistem pakar berbasis web dengan metode Certainty Factor (CF). Bukti yang tersedia terdiri atas artefak antarmuka, ringkasan kategori pangan, nilai aturan lama, dan satu skenario ikan yang dapat direkonstruksi secara manual; kode sumber, pengukuran laboratorium, dan jejak perhitungan terperinci untuk ikan tidak tersedia. Nilai CF pakar dan pengguna dikalikan untuk setiap bukti positif, kemudian diagregasikan secara iteratif. Kontribusi skenario ikan sebesar 0,42; 0,40; 0,12; dan 0,28 menghasilkan skor rekonstruksi tepat 0,7795072, yang hanya konsisten dengan nilai ringkasan prototipe sebesar 78%; jejak perhitungan ikan pada tingkat antarmuka tidak dapat diverifikasi. Serangkaian uji deterministik mengonfirmasi perilaku algoritma rekonstruksi untuk kasus kosong, nol, satu, bukti tunggal, perubahan urutan, dan input tidak valid. Audit juga menemukan teks mitigasi lama yang tidak aman dan asal-usul bobot aturan yang tidak memadai. Arsitektur revisi memisahkan skrining dari konfirmasi laboratorium serta menambahkan versi aturan, snapshot input, jejak perhitungan, dan catatan rujukan. Skor CF tetap merupakan ukuran konsistensi terhadap aturan dan bukan kadar formalin, probabilitas kontaminasi, akurasi prediktif, atau pengganti pengujian kimia tervalidasi.