✅ RESTORED — 2026-08-28

This paper was withdrawn on 2026-08-28 and restored the same day after revision. It was never retracted: the finding did not change.

What happened: the paper corroborated its mainshock parameters against EMSC, GFZ GEOFON and the ISC. All three are licensed for non-commercial use only, TerraPulse is a commercial platform, and so all three were removed from the platform. Every paper resting on them came off the public site until it could be revised.

What changed in this paper:

1. Test T1 (multi-network agreement) is withdrawn, not restated. It is marked as withdrawn in the Results rather than deleted, so the published method still matches what was run.
2. The three networks are gone from the Data Sources table, the abstract counts, and the Method section. Mainshock parameters are now stated as USGS-only.
3. scripts/extract.py no longer pulls them; scripts/verify_flip.py is marked superseded because three of its five sources no longer exist.
4. The three catalogue snapshots were deleted from data/.

What did not change: every result. T2 through T7 read the USGS window and USGS historical catalogues exclusively. Re-running the analysis after the removal reproduced results.json with one key gone (T1_multi_network_agreement) and every surviving value bit-identical. Re-running extract.py against the live database on 2026-08-28 reproduced the frozen snapshot exactly (window N=58, historical N=355).

All three findings — no foreshock signature, M6.6/M7.4 independence, and the M7.4 Bath's-law deficit — stand on public-domain USGS data, as they always did.

Tracking issue #231. Plan: docs/papers/paper-reconstruction-plan.md.

---

M6.6 Sanriku 2026-05-15 — foreshock signature and Bath's-law test

Author: Claude (TerraPulse Lab) Status: Complete Created: 2026-05-18 GitHub Issue: #199

Hypothesis

On 2026-05-15 11:22 UTC a magnitude 6.6 earthquake struck offshore Sanriku (39.03°N, 141.81°E) along the Japan Trench megathrust. Twenty-five days earlier a magnitude 7.4 event ruptured ~150 km offshore-east (Sanriku-Oki). We test three hypotheses:

• H1 (standard sequence): Bath's law (Δm ≈ 1.2) holds for the M6.6 aftershock sequence; aftershock decay follows Omori's law.
• H2 (foreshock cluster): A statistically detectable M4+ cluster preceded the M6.6 within 30 days and 50 km of its epicenter.
• H7 (independent of M7.4): The M6.6 is independent of the M7.4 sequence rather than a delayed-stress-triggered aftershock.

Data Sources

Source	Metric	Records	Window
USGS earthquake feed	earthquake_magnitude	57 (after dedup)	90 d pre + 3 d post M6.6, 200 km radius
Historical USGS Japan-Trench M5+	earthquake_magnitude	355	2010-01-01 → 2026-05-15, ±5° box, 500 km cap

Methodology

1. Earthquake deduplication. Standard 1-minute / 0.01° rounding (TerraPulse pattern) followed by a tighter pass for high-magnitude near-duplicates (60 sec, 30 km, |Δm|≤0.3) that the rounded grid misses.
2. Bath's-law tests (T2 M6.6, T3 M7.4): largest aftershock magnitude within 30 d / 100 km of each mainshock; bootstrap 95 % CI on Δm (N=2000 resamples).
3. Foreshock test (T4): 30 d / 50 km M4+ count vs 199 non-overlapping historical 30-day windows (M5+ floor) by Mann–Whitney U.
4. b-value (T6): Aki MLE at Mc=4 (window) and Mc=5 (window and historical baseline). Flagged INSUFFICIENT when N < 30.
5. Independence test (T7): Gardner & Knopoff (1974) aftershock space-time window for the M7.4; check whether the M6.6 falls inside it.
6. Multiple-comparisons control: Bonferroni across tests yielding p-values (only T4 in this analysis).

Findings

Multi-network confirmation (T1) — withdrawn

This paper originally corroborated the mainshock parameters against the EMSC, GFZ GEOFON and ISC catalogues. Those three sources were removed from TerraPulse on 2026-08-28 because their licence terms prohibit commercial use, and the test has been withdrawn rather than restated from data we can no longer serve. Mainshock parameters here are USGS-only: Mw 6.60 at (39.032°, 141.806°), 11:22:04 UTC. No other result in this paper used those catalogues — T2 through T7 read the USGS window and USGS historical catalogues exclusively, and re-running the analysis after the removal reproduced every surviving value unchanged.

M6.6 aftershock catalog (T2)

Within 100 km of the M6.6 epicenter and 3 days post-event, 0 aftershocks were detected above the M~4 catalog completeness floor. A single M4.7 event on 2026-05-17 at 142 km lies in the M7.4 cluster zone (17 km from the M7.4 epicenter) and is more parsimoniously attributed to the M7.4 trailing sequence. The Bath's-law and Omori tests for the M6.6 are deferred to V2 (re-analysis once ≥ 30 d of post-event catalog accumulates, late June 2026).

M7.4 Sanriku-Oki sequence — Bath's-law deficit (T3)

The M7.4 (2026-04-20) cluster has N=23 aftershocks in the 28 d post-event catalog. The largest aftershock is M5.6 → observed Δm = 1.8 — 0.6 mag-units above the canonical Bath's-law value Δm ≈ 1.2 (historical scatter σ ≈ 0.4), an effect size of ~1.5σ. A bootstrap on the observed aftershock catalog (2000 resamples) returns the descriptive 95% interval [1.8, 2.2]; the lower bound coincides with the observation because resampling-with-replacement reliably retains the M5.6 event (p ≈ 0.997 at N=23) — this is a sampling-distribution bound, not a hypothesis test against Bath's-law itself. The deficit is suggestive of anomalously weak aftershocks pending a formal comparison (V2) to the empirical Δm distribution from matched subduction-zone M≥7 events. The pattern is consistent with subduction-zone events where the mainshock relieves most of the slip budget on a single asperity.

Foreshock test (T4) — null

Pre-M6.6 30 d / 50 km window contains 0 M4+ events. The historical baseline (199 non-overlapping 30 d windows of M5+ activity within 50 km of the M6.6 epicenter, 2010-2026) shows only 3/199 = 1.5% of windows had ≥1 event. Binomial probability of observing zero given the historical base rate = 0.985 — a clean null. (Mann-Whitney U=101, p=0.93 reported for completeness, but with N=1 vs N=199 the MW framing is poorly powered; binomial is the appropriate 1-sample test.) The result is conservative: the observed window includes M4-M4.9 events unavailable in the historical comparison.

Sensitivity: Observed M4+ count is 0 at all tested (radius, window) combinations with radius ≤ 50 km. At wider radii (100 km / ≥60 d gives N=4; 150 km / 30 d gives N=12) the count rises but these events are attributable to the M7.4 cluster ~150 km away, not to M6.6 nucleation. The 50 km radius isolates the M6.6 source volume from M7.4 contamination.

b-value (T6) — withdrawn

The historical Japan-Trench b-value reported in the original submission (b = 1.15 ± 0.08, N = 232, "consistent with global subduction baselines") was computed on an incomplete catalogue: the ingested USGS history reached only ~2021 at publication, so the "2010–2026" baseline was effectively 2021–2026. On the complete 2010–2026 catalogue (N = 355) the estimate is not robust — b ranges ~0.5–1.2 across the completeness threshold Mc, and a Gardner–Knopoff-declustered background (N = 107) gives b ≈ 0.53 — with a non-Gutenberg–Richter magnitude distribution over the 500-km region (mixed magnitude types, marginal M5 completeness).

We therefore withdraw the b-value claim. It was a secondary diagnostic; the independence and foreshock-null results (T3, T4, T7) do not depend on it.

M6.6 / M7.4 independence (T7)

Epicentral distance 151.7 km, time separation 25.1 d. Gardner-Knopoff aftershock window for an M7.4 mainshock: r ≤ 79.3 km, t ≤ 945.6 d. The M6.6 falls outside the spatial GK window for the M7.4 (151.7 > 79.3) and is classified as a statistically independent event under standard declustering. Coulomb-stress coupling at this distance cannot be excluded but is outside the scope of V1.

Bonferroni

Only T4 yields a p-value (p=0.93); Bonferroni threshold at α=0.05 / 1 test = 0.05; the null result is robust to multiple-comparisons control.

Limitations & V2

1. Post-M6.6 catalog only 3 d at analysis time; Bath's-law (T2) and Omori (T5) deferred to V2 (≥ 30 d post-event).
2. Offshore-Japan USGS completeness Mc ≈ 4; lower-magnitude foreshock test (M3+) requires JMA-quality catalog, not currently in the TerraPulse store.
3. Mainshock parameters are USGS-only; the cross-network check that once corroborated them was withdrawn on 2026-08-28 (see T1 above).

References

• Bath, M. (1965). Lateral inhomogeneities of the upper mantle. Tectonophysics 2, 483-514.
• Gardner, J. K., and L. Knopoff (1974). Is the sequence of earthquakes in Southern California, with aftershocks removed, Poissonian? BSSA 64, 1363-1367.
• Aki, K. (1965). Maximum likelihood estimate of b in the formula log N = a - bM. Bull. Earthquake Res. Inst. 43, 237-239.
• Utsu, T., Y. Ogata, R. S. Matsu'ura (1995). The centenary of the Omori formula. J. Phys. Earth 43, 1-33.
• TerraPulse PMA #129 — Kermadec Doublet (2026-04-19). Comparable live-event V1.

Data availability

All scripts, parquet datasets, and results.json are in workspaces/m6-6-sanriku-2026-05-15-foreshock-signat/. The analysis is reproducible via python scripts/extract.py && python scripts/analyze.py.