Evidence A — Fixed PTW Benchmark vs. IDA-PTW Adaptive

Evidence A — Benchmark Fixed PTW vs. IDA-PTW Adaptif

IDA-PTW Adaptive matches Fixed 10 s composite accuracy while providing sub-3 s latency for the dominant (>94%) low-intensity event class. At Sa(0.3s) — the most safety-critical structural period — IDA-PTW outperforms Fixed 10 s by +0.9 p.p.

IDA-PTW Adaptif mencapai akurasi komposit setara Fixed 10 s sambil memberikan latensi di bawah 3 s untuk kelas event intensitas-rendah yang dominan (>94%). Pada Sa(0,3 s) — periode struktural paling kritis untuk keselamatan — IDA-PTW mengungguli Fixed 10 s sebesar +0,9 p.p.

* IDA-PTW evaluated on the N=2,747 stratified subset (traces with valid 3s+5s+10s windows). Fixed PTW rows on N=25,058. Source: benchmark_results_fixed.csv, benchmark_results_ida.csv.

* IDA-PTW dievaluasi pada subset stratified N=2.747 (trace dengan window 3s+5s+10s valid). Baris Fixed PTW pada N=25.058. Sumber: benchmark_results_fixed.csv, benchmark_results_ida.csv.

Evidence B — Information Ceiling Analysis

Evidence B — Analisis Information Ceiling

The theoretical upper bound achievable on this dataset; adding ~88.9 s of pre-P ambient noise yields only +0.93% R², empirically confirming that pre-P information is negligible for spectral prediction.

Batas atas teoretis yang dapat dicapai pada dataset ini; menambahkan ~88,9 s derau ambient pra-P hanya memberikan +0,93% R², secara empiris membuktikan bahwa informasi pra-P dapat diabaikan untuk prediksi spektral.

† Post-P Full-Wave and Total MiniSEED values are quoted from experimental runs c7a50193 and c3399cac respectively; raw per-period CSVs not retained in current reports/. See cross-check for status.

† Nilai Post-P Full-Wave dan Total MiniSEED dikutip dari eksperimental run c7a50193 dan c3399cac; CSV per-periode raw tidak tersimpan di reports/ saat ini. Lihat cross-check untuk status.

Evidence C — Magnitude Saturation Test (High-PGA Subset)

Evidence C — Uji Magnitude Saturation (Subset PGA-Tinggi)

Forcing high-intensity events (PGA ≥ 0.1 gal, N=1,204) into fixed 3-second windows severely degrades long-period spectral accuracy — the saturation effect that IDA-PTW's 10 s Damaging-class window is designed to prevent.

Memaksakan event intensitas-tinggi (PGA ≥ 0,1 gal, N=1.204) ke window tetap 3 detik menurunkan akurasi spektral periode-panjang secara signifikan — efek saturasi yang dirancang untuk dicegah oleh window 10 s kelas Damaging pada IDA-PTW.

Source: saturation_test_results.csv. Relative gain reported against the Fixed 3 s baseline for direct physical interpretation.

Sumber: saturation_test_results.csv. Gain relatif dilaporkan terhadap baseline Fixed 3 s untuk interpretasi fisis langsung.

Evidence D — P-Arrival Pick Sensitivity

Evidence D — Sensitivitas Pick P-Arrival

Fixed 10 s PTW under synthetic P-pick perturbations ±2 s. The pipeline's dominance of cumulative features (CAV, CVAD, Arias Intensity) confers robustness to realistic picking errors — maximum degradation is 0.45%, well below the median error of any operational P-picker.

Fixed 10 s PTW di bawah perturbasi P-pick sintetis ±2 s. Dominasi fitur kumulatif (CAV, CVAD, Arias Intensity) membuat pipeline robust terhadap error picking realistis — degradasi maksimum 0,45%, jauh di bawah median error P-picker operasional mana pun.

Source: p_arrival_sensitivity.csv. Ground-truth (0.0 s) picks confirmed optimal, validating dataset pick quality.

Sumber: p_arrival_sensitivity.csv. Pick ground-truth (0,0 s) terbukti optimal, memvalidasi kualitas pick di dataset.

Synthesis

Sintesis

The four evidences establish a coherent validation hierarchy:

Keempat evidence membentuk hierarki validasi yang koheren:

• Evidence A quantifies the operational performance at ~R² = 0.832, parity with Fixed 10 s at 1/3 the latency for >94% of events.
• Evidence A mengukur performa operasional pada ~R² = 0,832, setara Fixed 10 s dengan 1/3 latensi untuk >94% event.
• Evidence B demonstrates the 0.115 R² headroom to the ~341 s post-P ceiling is irreducible aleatory uncertainty (site/path variability) not recoverable from P-wave information alone.
• Evidence B membuktikan bahwa headroom 0,115 R² ke ceiling post-P ~341 s merupakan aleatory uncertainty yang tidak bisa direduksi (variabilitas site/path) dan tidak dapat dipulihkan hanya dari informasi P-wave.
• Evidence C proves that the 10 s adaptive allocation for high-intensity events prevents the +7% R² loss that fixed 3 s windows incur through magnitude saturation.
• Evidence C membuktikan bahwa alokasi adaptif 10 s untuk event intensitas-tinggi mencegah hilangnya +7% R² yang terjadi pada fixed 3 s karena magnitude saturation.
• Evidence D confirms operational robustness to P-pick uncertainty, validating deployability on automated picking pipelines (GPD, PhaseNet, EQTransformer).
• Evidence D mengonfirmasi robustness operasional terhadap ketidakpastian P-pick, memvalidasi deployability pada pipeline picking otomatis (GPD, PhaseNet, EQTransformer).

Appendix Preview (DOCX rendered as images)

Pratinjau Lampiran (DOCX ditampilkan sebagai gambar)

Page 1 — Evidence A + B tables

Halaman 1 — Tabel Evidence A + B

Page 2 — Evidence D + Synthesis + Data Provenance

Halaman 2 — Evidence D + Sintesis + Data Provenance