Project Artifacts: Secure Human-AI Augmentation for Equitable Disaster Recovery Support

This repository provides supplementary research project artifacts: evidential structures, encrypted aggregation metadata, and resources, to support the findings of a real-world case study on equitable disaster recovery decision-making. The project presents a decentralized privacy-preserving Human-AI augmentation framework designed to support collaboration among multiple organizations operating under strict regulatory, organizational, and data governance constraints. Secure augmentation of proprietary organizational AI decision-support systems without exposing raw sensitive data, contextual human expert heuristic judgment, and the requirement to converge towards a single global model. Privacy-preserving augmentation is supported by lattice-based Fully Homomorphic Encryption (FHE) schemes: CKKS and TFHE, for secure encrypted computation in untrusted environments. Blockchain-assisted governance mechanisms as trust-machine provide auditable coordination, integrity verification, decentralized accountability, and traceability of augmentation workflows. (a) Title: MetaData_Disaster_Recovery.json (v1.0) Description: The metadata file of SMEs affected by extreme weather events. It includes structured information on disaster events, financial metrics, credit & legal history, insurance coverage, operational indicators, government support, and fraud flags. Key metadata highlights: Disaster Events: Event Name ,Event_Date, and Met Office warning levels (Green, Yellow, Amber, Red) Sectors: Wholesale & Retail Trade, Manufacturing, Construction, Accommodation & Food Services, etc. Special inclusivity for Marginalized Small Businesses with ownership categories, marginalization multipliers, adjusted DSCR, alternative collateral, IMD postcode rules, and disaster severity adjustments. Profile design: marginalized profiles, comparator profiles, and counterfactual pairs for fairness evaluation. Evidence fields (E) with categories, value ranges, missing counts, and percentages. (b) Title: Client_Encrypted_Contributions_Reliability.json (v1.0) Description: This file contains metadata and reliability analysis results for encrypted client contributions in the Natural Disaster Recovery Support Dataset project. It reports reliability scores of 4 client organizations across three progressive augmentation stages using Fully Homomorphic Encryption (FHE). Higher values indicate more consistent and trustworthy encrypted data contributions. Stages Overview: Stage 1 (16 tasks): High reliability regime Stage 2 (24 tasks): Moderate reliability with fluctuations and detected tampering Stage 3 (24 tasks): High-complexity joint evidence Each stage includes the full reliability matrix, per-client average scores, and detailed stage descriptions. Purpose: Evaluate and monitor the trustworthiness of encrypted client contributions. (c) Title: Augmentation_Evidence_Disaster_Recovery.json (v1.0) Description: This file details the staged evidence augmentation process for the Natural Disaster Recovery Support. It defines how single and joint evidence were progressively introduced across three stages to support privacy-preserving, fair decision-making in disaster recovery lending. Stages Overview: Stage 1 (16 tasks): Single Evidence Stage 2 (24 tasks): Joint Evidence (Moderate Complexity) Stage 3 (24 tasks): Complex Joint Evidence Purpose: Document the incremental evidence augmentation strategy that enables secure, transparent, and fairness-aware model training. (d) Title: Sample_Augmentation_Task_Batch_Consent_Contribution_Governance.zip Description: Fully asynchronous governance workflow managed by blockchain Temporal delays between organizations for task awareness, voting, and contribution submission Multi-phase consent mechanism (C_REQ, C_PK, C_CTB, C_KS, C_Γ) Homomorphic encryption metadata Encrypted contribution records with SHA-256 hashes, Merkle roots, aggregation times, and transmission delays Complete audit trail with timestamps and transaction IDs This sample belongs to a series of three files covering the three augmentation stages (Stage 1 in Nov 2025, Stage 2 in Jan 2026, and Stage 3 in Mar 2026). Purpose: Examples of asynchronous consent and contribution governance by blockchain. (e) Sector_Wise_SME_Disparity_Reduction.json (v1.0) This dataset contains sector-wise disparity reduction results (ΔDis_red) for six encrypted learning techniques applied to SME decision support in natural disaster recovery scenarios. Sectors Analyzed: Wholesale & Retail, Manufacturing, Construction, Motor Repair, Accommodation & Food, Transport & Storage Profile Types: Marginalized Profiles, Comparator Profiles, Counterfactual Pairs Techniques Evaluated: ER-X with Human-AI Augmentation (TFHE) ER-X with Human-AI Augmentation (CKKS) FedProx (TFHE) FedProx (CKKS) FedAvg (TFHE) FedAvg (CKKS) (f) Title: FHE_Multivar_Function_Results.json (v1.0) Description: This file presents the noise growth and scalability analysis of Fully Homomorphic Encryption (FHE) schemes for multivariate evidential reasoning-explainer (ER-X) functions. It evaluates relative decryption error (%) under 128-bit security for CKKS and TFHE as the number of participating organizations scales from 2 to 32 (30 realizations per point). Plaintext computation serves as the zero-error baseline. Analysis Type: Relative decryption error scalability Functions Analyzed (F1–F9): Aggregated Data, Aggregated Belief, Likelihood, Basic Probability, Singleton Certainty, Non-Singleton Ambiguity, Data Decisiveness, Belief Decisiveness, and Inconsistency. Results are provided for both CKKS and TFHE across all nine encrypted functions, showing how noise grows with increasing organizations and computational depth. Purpose: To assess the practical feasibility and scalability of privacy-preserving encrypted aggregation in secure multi-party computation for disaster recovery lending decisions. (g) Title: uk_sme_disaster_recovery_storms_2022_2026.json (v1.0) Description: Records of United Kingdom Small and Medium-sized Enterprise (SME) disaster-recovery lending instances associated with major named storm events between 2022 and 2026 used in the analysis. The dataset comprises SME instances stratified by disaster-risk level (High, Medium, Low, and Very Low) and organized across major storm events and subsequent severe flooding. (h) Title: Decrypted_Decisiveness_Weight_Data_and_HumanExpert.json (v1.0) Description: This dataset contains the decrypted decisiveness weights and normalized inconsistency metrics to estimate reliability of encrypted information aggregated from multiple organizations. decisiveness_weight_data: Average decisiveness weights derived from historical organizational data across tasks.decisiveness_weight_belief: Average decisiveness weights derived from human expert judgments.normalized_inconsistency: Degree of misalignment between past data-driven decisions and current human expert assessments. These metrics quantify the reliability and credibility of encrypted contributions received from participating organizations. They are used to compute per-client reliability scores and dynamic reputation to support tamper detection and trustworthy decision-making.