PHILIA OS Phase 2 — 103rd DOI: D-09-Dominant Probe-Profile Shape Separation Across an Extended Five-Arm Panel, with thomas DISTANT_MULTIBASIN Secondary and Negative-Control Characterisation (Confirmatory Results)

This record reports the 103rd DOI confirmatory results of the PHILIA OS research series (Trinity AI Research Team, Konyang University). It tests whether the CV_mean probe-profile shape endpoint — driven primarily by D-09 (α=0.09) and established in the 102nd DOI (10.5281/zenodo.21426157) — reproduces chaos-family separation from a 26-member autoregressive reference panel under an extended five-arm probe panel {D-05, D-07, D-09, D-11, C-c1} and independent fresh confirmatory seeds (240–269). H-103-Primary: CONFIRMED. All three independent confirmatory batches (seeds 240–244, 245–249, 250–254) satisfied the pre-registered criterion (outside_mad ≥ 4/5 AND median margin > 0), achieving 3/3 PASS — exceeding the ≥2/3 threshold. All 15 confirmatory seeds for nose_hoover (the Primary continuity anchor) achieved outside_mad = True. All five pre-declared secondary analyses reached favourable states: chaos-family separation REPLICATED (11 non-primary substrates); legacy 4-arm panel CONTINUOUS; D-09-core (K=2) panel SUPPORTED; negative control HOLDS (henon, ikeda, lorenz96 — discrete-map and high-dimensional resistant substrates); and thomas DISTANT_MULTIBASIN elevated in 2 of 3 batches, directionally compatible with an independently identified basin structure from prior exploratory work. The Primary claim is strictly limited to the binary outside_mad direction under the locked forward-Euler (dt=0.01) integration scheme. Post-hoc numerical cross-validation (documented in the companion 102nd-DOI diagnostics) established that absolute T_mad magnitudes are integrator/dt-scheme-dependent, while the binary judgment is stable across four refined conditions (dt ≤ 0.005). A substrate-level numerical-failure disposition map is also reported: halvorsen shows a 40% divergence rate under the locked generator and is flagged DISPOSITION-UNSTABLE, requiring separate integrator diagnosis before any Tier-3 mechanistic interpretation — the divergence guard (math.isfinite check at generation time) prevented any degenerate fingerprint from contaminating primary or secondary verdicts. Governing pre-registration: 10.5281/zenodo.21472876 (v1.4; concept record originating at 10.5281/zenodo.21467129). Confirmatory script SHA256: 0de15d0cc1553ffcc601db65cce933f4949b3e4caf3661d3299674bb16558d32. Pre-registration violations to date: 1 (seeds 210–239 executed as an undeclared pilot run prior to Zenodo registration; fully disclosed in Pre-reg §0.6 and retired from confirmatory use). Measurement is life. Description, not proof. Pre-registration protects us.