PHILIA OS Phase 2 · 115th DOI Results Report: TextAnalyzer CH7-V5 — Structured Claim Representation and LIVE Evidence Acquisition from Base and False Claims (A/D Arm Independent Registered Exploratory Study)

This report presents the 115th DOI of the PHILIA OS research series (Trinity AI Research Team, Konyang University), corresponding to the results of the 114th registered exploratory study (pre-registration: 10.5281/zenodo.22640355). The study applies the frozen GapH3-architecture Gap Report gate (parent SHA: cc54df0e) to a 57-item CH7 fact-verification corpus under LIVE Wikipedia evidence fetch. A arm (base claim) and D arm (false claim) were executed as fully independent OS-process runs (script SHA: c54d443e…; qwen2.5:14b, temperature=0). Study type: REGISTERED EXPLORATORY. No ESTABLISHED/NOT_ESTABLISHED verdict is declared. Key findings: (1) A arm Gap Report: gap=YES 16/57, Precision 37.5%, Recall 75.0%, Invented 0; FP 10 split into retrieval-side mismatch (5) and interpretation-side (5) failure mechanisms. (2) D arm Gap Report: gap=YES 54/57; A=NO→D=YES transition in 39/57 items. (3) Evidence divergence (O4): raw hash differs in 25/57 items, entirely attributable to claim-induced retrieval-path divergence; temporal drift = 0; dual-hash design (raw + final) separated retrieval divergence from selection divergence. (4) O0: explicit false→true auto-correction = 0/57; Step1 truth-awareness observed in 27/57 D items (21 unique). (5) DERIVED-abuse structural gap confirmed by A/D arm pairing (H_005 decisive case): Stage1b cannot detect claim-anchored derivation; A arm alone concealed this. (6) L3_NEW_A D-label invalid: claim factually true under LIVE evidence (effective D arm N=56). Pre-registration analytic violations: 0. Errata: 2 (non-verdict, documented in §8). A/D arm process isolation confirmed. Related DOIs: Pre-registration 10.5281/zenodo.22640355 (114th/Zenodo 118) · Parent confirmatory results 10.5281/zenodo.22153519 (113th/Zenodo 117).