Pith. sign in

REVIEW 53 references

A conversational, protocol-driven server can reproduce Europe's published seismic hazard and risk numbers—475-year spectral accelerations at 73 cities within a median 5%—while letting users swap in their own ground-motion models and run the

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 09:44 UTC pith:T3V2FVPV

load-bearing objection A well-architected and honest software-integration paper that plausibly delivers the first agentic (MCP/LLM) interface to the end-to-end ESHM20/ESRM20 chain; the headline 5% replication claim is credible for most of Europe but is not yet checkable because code/data are withheld and panel selection is not described.

arxiv 2607.16249 v1 pith:T3V2FVPV submitted 2026-06-27 physics.geo-ph cs.AI

An Agentic Interface for End-to-End Probabilistic Seismic Hazard and Risk Analysis

Sreenath Vemula , Pierre Jehel , Fabrice Cotton , Filippo Gatti This is my paper
classification physics.geo-ph cs.AI
keywords Agentic AIModel Context ProtocolProbabilistic Seismic Hazard AnalysisESHM20ESRM20Ground-motion model substitutionSeismic risk assessmentConditional mean spectrum
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tries to establish that probabilistic seismic hazard and risk analysis, normally the work of specialists juggling configuration files and logic trees, can be handed to a conversational agent without losing numerical trust. Its central claim: a server exposing twenty-four typed calculation endpoints lets an LLM orchestrate the full chain—hazard curves, spectra, damage probabilities, economic loss—using the validated open-source engine and the 2020 European hazard and risk models, and reproduces the official 475-year spectra at 73 cities within a median 5%. If true, the contribution is that engineers, planners, and homeowners can run official-model calculations in minutes, substitute their own ground-motion model per tectonic region, and get every result with provenance. The paper is candid that the median hides a site-dependent residual: the three-branch ground-motion approximation underpredicts Vrancea-dominated sites like Bucharest by about 38%, so the headline accuracy metric is not uniform.

Core claim

The paper argues that the entire European probabilistic seismic hazard and risk chain can be wrapped in a protocol server that an LLM-driven agent uses as a tool catalogue. The agent plans and translates while the validated engine performs all numeric work, so the numbers are engine outputs, not language-model guesses. Using the published European models, the server reproduces official 475-year spectral accelerations at 73 cities within a median absolute log residual of 0.05 (about 5%), with the largest residuals concentrated at the Vrancea intermediate-depth zone (Bucharest 38% low, Chisinau log-residual 0.48). Users can substitute a custom ground-motion model per tectonic region, and the s

What carries the argument

Central machinery is a five-layer server exposing twenty-four typed endpoints, with a layer that translates engineering questions into model parameters, a hazard kernel, a spatial cache, and a validated numerical core running the hazard engine. The accuracy claim rests on a specific computational shortcut: the ground-motion logic tree is evaluated as a three-branch Gauss–Hermite mean of the highest-weighted branches while the source-model logic tree is evaluated deterministically on its highest-weighted subset. Supporting identities are the log-linear hazard-curve interpolation (Cornell's equation), Baker–Jayaram inter-period correlation for conditional mean spectra, the lognormal fragility

Load-bearing premise

The three-branch, highest-weighted subset of the ground-motion and source logic trees is representative enough of the full published tree that a median 5% match at 73 cities is a meaningful replication—an assumption the paper admits fails at Vrancea sites.

What would settle it

Run full Monte-Carlo sampling of the complete ESHM20 logic tree at the same 73 cities and compare the resulting 475-year spectral accelerations to the server's three-branch outputs; if the Vrancea log-residuals (0.48 at Chisinau, 0.31 at Bucharest) persist or grow at additional sites, the median-5% claim is an artifact of averaging rather than a site-level guarantee.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Non-specialists can reproduce official hazard values, curves, and spectra at any location in minutes, with source-model and ground-motion provenance attached.
  • A user-supplied empirical or machine-learning ground-motion model can be swapped in per tectonic region and propagated through the full hazard calculation, enabling rapid sensitivity comparisons against the published tree.
  • Capabilities beyond the official European web services—conditional mean spectra, deterministic scenarios, surface hazard, per-building loss, retrofit comparison, and record selection—become available in the same conversational session.
  • The layered architecture is a template that transfers to other regional hazard models and, in principle, to other hazards, since protocol, engine, regional dataset, and ground-motion logic tree can be replaced independently.
  • The benchmark quantifies accuracy: median absolute log residual of 0.05 at 475 years and 0.06 at 2,475 years across 73 cities, with high residuals concentrated at Vrancea-dominated sites.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The headline median masks a bimodal error distribution: at Vrancea-dominated sites the three-branch approximation under-predicts by roughly 38% in PGA, so users near Vrancea should treat the default outputs as lower bounds unless denser logic-tree sampling is requested.
  • The same substitution channel could be repurposed for systematic epistemic-uncertainty exploration—running full Monte-Carlo sensitivity across ground-motion branches at scale—something the paper notes as a possibility but does not demonstrate.
  • A natural testable extension: re-run the 73-city benchmark with the optional five-point Gauss–Hermite mode and quantify how much of the Vrancea and Craton residuals it removes; the paper reports spot evidence but no full-panel statistics.
  • The design's claim to generality across hazards would be tested by re-instantiating the layers on a non-seismic hazard (e.g., flood or wind), where the domain translator and endpoint catalogue would need substantial rework.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Circularity Check

0 steps flagged

No significant circularity: the central benchmark is external and the governing equations are verified against independent references.

full rationale

The paper's load-bearing replication claim is an empirical comparison against the EFEHR full-tree service, which is an external reference. The median 5% residual is a measured benchmark, not a fitted target: the server evaluates a three-branch Gauss–Hermite approximation, and the paper reports the residual honestly, including large Vrancea outliers (Chisinau 0.48, Bucharest 0.31) that worsen the median—the opposite of a fitted prediction. Governing equations (Eqs. 1–5) are verified against independent published sources (Cornell 1968; Baker & Jayaram 2008; ESRM20 spreadsheet libraries; closed-form integral oracle), and the OpenQuake engine is a third-party validated code base. The only author self-citations (Sreenath et al., 2023, 2025) appear as illustrative custom-GMM examples, not as constraints on the central replication result. The principal approximation (highest-weighted branches, 20-km discretization, finite distance cutoffs) is disclosed in Section 4.2 as the source of residual, not disguised as exact. Concerns about the representativeness of the 73-city panel or the absence of a full residual distribution are correctness/validation risks, not circularity. No derivation step reduces to its own input.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central replication claim rests on standard PSHA equations and published model files, plus several engineering approximation constants (quadrature order, grid size, distance cutoffs, branch subset). No new physical entities are postulated; the agentic server is a software artifact rather than a scientific entity. The load-bearing ad hoc choice is the three-branch quadrature used to approximate the full logic tree.

free parameters (4)
  • Three-branch Gauss-Hermite GMM quadrature = 3 nodes (optional 5-point mode)
    Default approximation to the full ESHM20 ground-motion logic tree; the paper reports it lowers median error from 9.3% to 4.7% versus the central branch, an accuracy/cost choice rather than a derived quantity.
  • Area-source discretization grid = 20 km
    Seismic sources are discretized on a 20 km grid; this approximation enters every hazard integral and contributes to the residual.
  • Tectonic-region distance cutoffs = 50 km (volcanic) to 500 km (Vrancea)
    Finite source-to-site distance truncations per tectonic region; affects hazard tails and long-distance contributions.
  • Deterministic highest-weighted branch subset = highest-weighted source and GMM branches
    Hazard is evaluated on a subset rather than Monte-Carlo of the full logic tree; authors state this is the principal source of residual.
axioms (5)
  • standard math Cornell (1968) PSHA integral as implemented in OpenQuake is the correct hazard model.
    All hazard values depend on the Cornell PSHA integral being a correct representation of seismic hazard; the paper delegates to OpenQuake without re-deriving it.
  • domain assumption ESHM20/ESRM20 published model files, weights, and EFEHR reference outputs define ground truth.
    The replication benchmark treats official EFEHR outputs as reference; if the ESHM20 model files or weights were wrong, the 5% claim would be meaningless.
  • ad hoc to paper Three-branch Gauss-Hermite mean of top-weighted GMM branches represents the full logic tree adequately at most sites.
    This is the paper's chosen approximation; it is not derived from the model and fails at Vrancea sites, as disclosed in Sections 3.2 and 4.2.
  • domain assumption Baker and Jayaram (2008) inter-period correlation model governs conditional mean spectra.
    Conditional spectra depend on this published correlation structure, imported without re-derivation.
  • domain assumption USGS Vs30 mosaic / EC8 class mapping gives reliable site amplification when no measured Vs30 exists.
    Surface spectra and risk results depend on inferred Vs30 when no measured value is supplied; the paper notes the proxy is uncertain at any single site.

pith-pipeline@v1.3.0-alltime-deepseek · 16109 in / 12556 out tokens · 104424 ms · 2026-08-02T09:44:25.340651+00:00 · methodology

0 comments
read the original abstract

Probabilistic seismic hazard and risk analyses are backbone to building codes, insurance pricing, and disaster management. Yet their open-engine pipelines remain accessible primarily to experts. We present the first agentic interface to the end-to-end probabilistic seismic hazard and risk chain via an open-source server, addressable through the Model Context Protocol (MCP). MCP wraps the OpenQuake engine and the 2020 European Seismic Hazard and Risk Models using twenty-four typed endpoints. Here, an agent is defined as a large language model (LLM) with tools. LLM is confined to the role of an orchestrator so it plans, translates, and explains, while the OpenQuake engine and custom codes compute hazard values, damage probability, and loss. Each response carries source-model, ground-motion model (GMM), and certified data provenance for transparency. Results are benchmarked against ESHM20 at seventy-three cities, the replicated 475-year spectral accelerations match official values within a median of 5 %, and a full hazard-to-loss estimate runs in minutes. The interface additionally accepts a user-supplied empirical or machine-learning GMM on any tectonic region type of the published tree, and adds additional features that existing web services omit: conditional spectra, deterministic scenarios, surface hazard, per-building loss, retrofit comparison, and record selection with waveform retrieval. The proposed design layers transfer to other regional models, and to other hazards.

Figures

Figures reproduced from arXiv: 2607.16249 by Fabrice Cotton, Filippo Gatti, Pierre Jehel, Sreenath Vemula.

Figure 1
Figure 1. Figure 1: Layered organization of the server, from the conversation down to the validated numerical core. Each layer is described by what it contributes to a query rather than by its implementation: requests travel downward, validated numbers and their provenance travel back upward, and any layer can be substituted without affecting the others. 5 [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Two user journeys through the layered stack: (a) 475-year PGA at Istanbul, at rock and at the soil of the city; (b) average annual loss for an unreinforced-masonry dwelling at Naples, reusing its cached hazard from the panel run. Values are those returned in the verification session (Supplementary Table S3). 6 [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The twenty-four endpoints, grouped by user intent. get endpoints return published ESHM20/ESRM20 content; compute endpoints run new calculations; bold boxes mark capabilities not available from the EFEHR web services. Every hazard endpoint also accepts a per-tectonic-region GMM override. 2.3 Data substrate ESHM20 contains the area-source, fault-source, and smoothed-seismicity model files in NRML, paired wit… view at source ↗
Figure 4
Figure 4. Figure 4: PGA hazard curves at representative cities, VS30 = 800 m s−1 : this framework as a three￾branch Gauss–Hermite ground-motion mean (dashed) and central branch (dashed with markers) against the EFEHR service evaluating the complete ESHM20 tree (solid). Agreement is within ∼8 % at the 475-year level for the three-branch mean except at Vrancea-dominated Bucharest, where even the three￾branch mean omits influent… view at source ↗
Figure 5
Figure 5. Figure 5: 475-year uniform hazard spectra at the panel cities. (a) Rock reference: this framework as a three-branch Gauss–Hermite ground-motion mean (dashed) and central branch (dashed with markers) against the EFEHR full-tree reference (solid). (b) Surface spectra at each city’s proxy-derived VS30, a site-amplified product the EFEHR services do not provide. 3.3 Hazard accuracy We compared the 5 %-damped acceleratio… view at source ↗
Figure 6
Figure 6. Figure 6: PGA hazard curves under two custom-GMM substitutions (VS30 = 800 m s−1 ). Istanbul: the shallow-crustal regime routed through a custom interpretable machine-learning GMM (Sreenath et al., 2025). Bucharest: the Vrancea intermediate-depth regime routed through the regional Manea et al. (2022) model. Solid: custom GMM substituted; dotted: default ESHM20 GMMs; wide translucent: EFEHR full tree; dashed: 475-yea… view at source ↗
Figure 7
Figure 7. Figure 7: condenses the workflow into one view: panels (a–c) compare the selected records with the disaggregation target at Vienna, Istanbul, and Reykjav´ık, and panel (d) overlays the target conditional mean spectrum (Eqs. 2–3), a product the existing services do not provide, with the geometric mean of the suite selected at Vienna. There the geometric mean lies within the conditional ±σln band at every period. The … view at source ↗
Figure 8
Figure 8. Figure 8: Average annual structural loss ratio at a south-eastern-European city panel for two repre￾sentative residential classes: (a) a mid-rise reinforced-concrete frame (CR-LFINF-CDM-0H4) and (b) a low-rise unreinforced-masonry building (MUR-CL99-LWAL-DNO-H2), each computed by convolving the city’s surface hazard curve with the matching ESRM20 vulnerability function at its fragility intensity measure (Eq. 5; SA(1… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 8 linked inside Pith

  1. [1]

    and Horspool, N

    Abbott, E. and Horspool, N. and Gerstenberger, M. and Huso, R. and Van Houtte, C. and McVerry, G. and Canessa, S. , title =. Earthquake Spectra , volume =. 2020 , doi =

    2020

  2. [2]

    Abrahamson, N. A. and Gregor, N. and Addo, K. , title =. Earthquake Spectra , volume =. 2016 , doi =

    2016

  3. [3]

    Baker, J. W. , title =. Journal of Structural Engineering , volume =. 2011 , doi =

    2011

  4. [4]

    Baker, J. W. and Jayaram, N. , title =. Earthquake Spectra , volume =. 2008 , doi =

    2008

  5. [5]

    Baker, J. W. and Lee, C. , title =. Journal of Earthquake Engineering , volume =. 2018 , doi =

    2018

  6. [6]

    and Orzechowski, M

    Balis, B. and Orzechowski, M. and Kica, P. and Dygas, M. and Kuszewski, M. , title =. arXiv preprint arXiv:2604.21910 , year =

    Pith/arXiv arXiv

  7. [7]

    and Bao, Y

    Chen, J. and Bao, Y. , title =. Automation in Construction , volume =

  8. [8]

    and Wang, W

    Chen, Y. and Wang, W. and Lobry, S. and Kurtz, C. , title =. arXiv preprint arXiv:2410.18792 , year =

    Pith/arXiv arXiv

  9. [9]

    Cornell, C. A. , title =. Bulletin of the Seismological Society of America , volume =

  10. [10]

    and Despotaki, V

    Crowley, H. and Despotaki, V. and Rodrigues, D. and Silva, V. and Toma-Danila, D. and Riga, E. and Karatzetzou, A. and Sousa, L. and Ozcebe, S. and Zugic, Z. and Gamba, P. , title =. Earthquake Spectra , volume =. 2020 , doi =

    2020

  11. [11]

    and Dabbeek, J

    Crowley, H. and Dabbeek, J. and Despotaki, V. and Rodrigues, D. and Martins, L. and Silva, V. and Rom. 2021 , doi =

    2021

  12. [12]

    and Crowley, H

    Dabbeek, J. and Crowley, H. and Silva, V. and Weatherill, G. and Paul, N. and Nievas, C. I. , title =. Bulletin of Earthquake Engineering , volume =. 2021 , doi =

    2021

  13. [13]

    and Giardini, D

    Danciu, L. and Giardini, D. and Weatherill, G. and Basili, R. and Nandan, S. and Rovida, A. and Beauval, C. and Bard, P.-Y. and Pagani, M. and Reyes, C. G. and Sesetyan, K. and Vilanova, S. and Cotton, F. and Wiemer, S. , title =. Natural Hazards and Earth System Sciences , volume =. 2024 , doi =

    2024

  14. [14]

    and Ngiam, J

    Errico, H. and Ngiam, J. and Sojan, S. , title =. arXiv preprint arXiv:2511.20920 , year =

    arXiv

  15. [15]

    and Lv, Z

    Feng, P. and Lv, Z. and Ye, J. and Wang, X. and Huo, X. and Yu, J. and Xu, W. and Zhang, W. and Bai, L. and He, C. and Li, W. , title =. Proceedings of the International Conference on Learning Representations , year =

  16. [16]

    and Diks, I

    Flotho, M. and Diks, I. F. and Flotho, P. and Molano, L.-A. G. and Hirsch, P. and Keller, A. , title =. Briefings in Bioinformatics , volume =. 2026 , doi =

    2026

  17. [17]

    and Liu, J

    Geng, Z. and Liu, J. and Cao, R. and Cheng, L. and Wang, H. and Cheng, M. , title =. arXiv preprint arXiv:2510.05414 , year =

    arXiv

  18. [18]

    and Nanavati, J

    Gridach, M. and Nanavati, J. and Abidine, K. Z. E. and Mendes, L. and Mack, C. , title =. arXiv preprint arXiv:2503.08979 , year =

    Pith/arXiv arXiv

  19. [19]

    , title =

    Hartung, T. , title =. Frontiers in Artificial Intelligence , volume =. 2025 , doi =

    2025

  20. [20]

    Hasan, M. M. and Li, H. and Rajbahadur, G. K. and Adams, B. and Hassan, A. E. , title =. arXiv preprint arXiv:2602.14878 , year =

    Pith/arXiv arXiv

  21. [21]

    Heath, D. C. and Wald, D. J. and Worden, C. B. and Thompson, E. M. and Smoczyk, G. M. , title =. Earthquake Spectra , volume =. 2020 , doi =

    2020

  22. [22]

    and Zhao, Y

    Hou, X. and Zhao, Y. and Wang, S. and Wang, H. , title =. arXiv preprint arXiv:2503.23278 , year =

    Pith/arXiv arXiv

  23. [23]

    and Lin, T

    Jayaram, N. and Lin, T. and Baker, J. W. , title =. Earthquake Spectra , volume =

  24. [24]

    and Li, C

    Kim, H. and Li, C. and Deng, W. and Jin, M. and Huang, W. and Lu, M. and Yuan, B. , title =. arXiv preprint arXiv:2511.20109 , year =

    arXiv

  25. [25]

    Kotha, S. R. and Weatherill, G. and Bindi, D. and Cotton, F. , title =. Bulletin of Earthquake Engineering , volume =. 2020 , doi =

    2020

  26. [26]

    Kotha, S. R. and Weatherill, G. and Bindi, D. and Cotton, F. , title =. Bulletin of Earthquake Engineering , volume =. 2022 , doi =

    2022

  27. [27]

    and Jost, A

    Kuznetsov, I. and Jost, A. A. and Pantiukhin, D. and Shapkin, B. and Jung, T. and Koldunov, N. , title =. npj Climate Action , volume =. 2025 , doi =

    2025

  28. [28]

    and Luzi, L

    Lanzano, G. and Luzi, L. , title =. Bulletin of Earthquake Engineering , volume =. 2019 , doi =

    2019

  29. [29]

    and Gatti, F

    Lehmann, F. and Gatti, F. and Clouteau, D. , title =. Journal of Computational Physics , volume =. 2025 , doi =

    2025

  30. [30]

    and Li, D

    Li, L. and Li, D. and Ou, Z. and Xu, X. and Liu, J. and Ma, Z. and Yu, R. and Deng, M. , title =. arXiv preprint arXiv:2506.06355 , year =

    Pith/arXiv arXiv

  31. [31]

    and Fu, X

    Li, Y. and Fu, X. and Verma, G. and Buitelaar, P. and Liu, M. , title =. arXiv preprint arXiv:2510.24476 , year =

    arXiv

  32. [32]

    and Zhou, Y

    Liang, H. and Zhou, Y. and Talebi-Kalaleh, M. and Mei, Q. , title =. arXiv preprint arXiv:2510.11004 , year =

    arXiv

  33. [33]

    and Ning, Y

    Lin, X. and Ning, Y. and Zhang, J. and Dong, Y. and Liu, Y. and Wu, Y. and Qi, X. and Sun, N. and Shang, Y. and Wang, K. and Cao, P. and Wang, Q. and Zou, L. and Chen, X. and Zhou, C. and Wu, J. and Zhang, P. and Wen, Q. and Pan, S. and Wang, B. and Cao, Y. and Chen, K. and Hu, S. and Guo, L. , title =. arXiv preprint arXiv:2509.18970 , year =

    arXiv

  34. [34]

    and Qu, R

    Liu, C. and Qu, R. and Liu, C. and Wang, J. and Pan, J. , title =. Automation in Construction , volume =. 2026 , doi =

    2026

  35. [35]

    Manea, E. F. and Cioflan, C. O. and Danciu, L. , title =. Earthquake Spectra , volume =. 2022 , doi =

    2022

  36. [36]

    and Silva, V

    Martins, L. and Silva, V. , title =. Bulletin of Earthquake Engineering , volume =. 2020 , doi =

    2020

  37. [37]

    and Lanzano, G

    Mascandola, C. and Lanzano, G. and Felicetta, C. and D'Amico, M. C. and Russo, E. and Sgobba, S. and Pacor, F. and Luzi, L. , title =. Earthquake Spectra (in press) , year =

  38. [38]

    and Monelli, D

    Pagani, M. and Monelli, D. and Weatherill, G. and Danciu, L. and Crowley, H. and Silva, V. and Henshaw, P. and Butler, L. and Nastasi, M. and Panzeri, L. and Simionato, M. and Vigan. Seismological Research Letters , volume =. 2014 , doi =

    2014

  39. [39]

    and Chard, R

    Pan, H. and Chard, R. and Mello, R. and Grams, C. and He, T. and Brace, A. and Price Skelly, O. and Engler, W. and Holbrook, H. and Oh, S. Y. and Gonthier, M. and Papka, M. and Blaiszik, B. and Chard, K. and Foster, I. , title =. arXiv preprint arXiv:2508.18489 , year =

    Pith/arXiv arXiv

  40. [40]

    and Shapkin, B

    Pantiukhin, D. and Shapkin, B. and Kuznetsov, I. and Jost, A. A. and Koldunov, N. , title =. Frontiers in Artificial Intelligence , volume =. 2025 , doi =

    2025

  41. [41]

    and Cotton, F

    Schneider, M. and Cotton, F. and Schweizer, P.-J. , title =. Natural Hazards and Earth System Sciences , volume =. 2023 , doi =

    2023

  42. [42]

    and Gatti, F

    Sreenath, V. and Gatti, F. and Jehel, P. , title =. JGR: Machine Learning and Computation , volume =. 2025 , doi =

    2025

  43. [43]

    and Podili, B

    Sreenath, V. and Podili, B. and Raghukanth, S. T. G. , title =. Earthquake Engineering & Structural Dynamics , volume =. 2023 , doi =

    2023

  44. [44]

    and Cotton, F

    Weatherill, G. and Cotton, F. , title =. Bulletin of Earthquake Engineering , volume =. 2020 , doi =

    2020

  45. [45]

    and Cotton, F

    Weatherill, G. and Cotton, F. and Daniel, G. and Zentner, I. and Iturrieta, P. and Bosse, C. , title =. Natural Hazards and Earth System Sciences , volume =. 2024 , doi =

    2024

  46. [46]

    and Kotha, S

    Weatherill, G. and Kotha, S. R. and Cotton, F. , title =. Bulletin of Earthquake Engineering , volume =. 2020 , doi =

    2020

  47. [47]

    and Crowley, H

    Weatherill, G. and Crowley, H. and Roull. Bulletin of Earthquake Engineering , volume =. 2023 , doi =

    2023

  48. [48]

    and Chen, Z

    Widjaja, F. and Chen, Z. and Zhou, J. , title =. arXiv preprint arXiv:2510.02139 , year =

    arXiv

  49. [49]

    , title =

    Wong, B. , title =. Nature Methods , volume =. 2011 , doi =

    2011

  50. [50]

    and Feng, C

    Xu, X. and Feng, C. and Zha, C. and He, W. and He, M. and Xiao, B. and Gao, X. , title =. Protein Science (in press; bioRxiv preprint) , year =

  51. [51]

    and Li, Z

    Yan, S. and Li, Z. and Zhu, S. and Wen, Y. and Zhang, M. and Chen, M. and Cao, J. and Hong, Y. , title =. arXiv preprint arXiv:2508.02936 , year =

    Pith/arXiv arXiv

  52. [52]

    and Wang, Z

    Zhao, D. and Wang, Z. and Wang, J. and Liao, J. A. and Li, Z. and Yang, S. , title =. SoftwareX , volume =. 2026 , doi =

    2026

  53. [53]

    and Tang, Y

    Zhu, Z. and Tang, Y. and Tang, X. and Zhang, J. and Gao, C. and Zhang, S. and Xu, H. and Guan, T. , title =. Geophysical Research Letters , volume =. 2026 , doi =

    2026