JCCDB v1.2 — Cryptographic Audit Hash and Macroeconomic Price Correction for Reproducible LLM-Based Construction Cost Diagnostics

Abstract

We present JCCDB v1.2, a substantive extension to the Japan Construction Cost Database (JCCDB), an open structured dataset supporting LLM-based construction cost diagnostic systems for consumer protection.

First, we introduce a cryptographic audit hash mechanism based on SHA-256 that provides reproducibility guarantees for diagnostic outputs: identical inputs always produce the same 12-character (48-bit) audit hashes printed on consumer-facing PDF reports, enabling vendor-neutral, client-side verification during contractor negotiations.

Second, we describe a War Price Coefficient (WPC) subsystem that automatically incorporates manufacturer price-change announcements into monthly price corrections, addressing the documented exploitation of material price volatility in renovation contractor fraud. The current deployed coefficient is ×1.0935 (+9.35%, Hormuz Strait/Iran-conflict baseline), computed from a weighted composite of seven CGPI series and four manufacturer announcements.

We demonstrate the production deployment of both mechanisms within HORIZON SHIELD, a Cloudflare Workers-based diagnostic service, and validate hash output stability. Theoretical collision analysis using the birthday-bound approximation indicates collision probability below 2×10⁻⁵ for realistic deployment scales within the 48-bit truncated hash space.

The mechanisms described are designed to align with the recording and human-oversight principles of EU AI Act Article 12 and Article 14 (Regulation 2024/1689, applicable to high-risk systems from August 2, 2026), and are released under CC-BY 4.0 to support independent replication.