1  About the Author

Who wrote this — and why the numbers can be trusted

1.1 Lewis Matthews (Corresponding Author)

Lewis Matthews is the analyst of record for the causal-inference pipeline this manual is built on: the confounder model, the targeted-learning estimators, the GPU solver that made full-panel analysis feasible, and the negative-control and sensitivity suites. Why those numbers should be trusted is a question of track record — and his runs through the three fields this project deliberately fuses:

  • Permian Basin operations, from the inside. Five years (2016–2021) as Senior Data Scientist at CrownQuest Operating, a Midland-based operator — building statistical and machine-learning systems on production and operational data and translating them for non-technical decision-makers. The wells in this dataset are not abstractions to the author: he has sat in the rooms where injection, permitting, and curtailment decisions get made, which informs every policy translation in this book.
  • Institutional finance. A decade (2003–2013) leading a family-office advisory practice in institutional capital markets — the discipline of fiduciary-grade evidence, and the language of the operators, insurers, and capital partners this manual is written for.
  • Causal methods. Founder of Project Geminae (applying causal inference to industrial optimization problems) and corresponding author on SPE-228051, the targeted-maximum-likelihood analysis this pipeline extends. The estimators in this book are drawn from the current targeted-learning literature (van der Laan group, 2024–2025) and were implemented and validated against reference implementations before any result was reported.

Contact: .

1.2 A disclosure, made deliberately

The author spent five years inside a Permian Basin operator and maintains long-standing industry relationships — and is publishing evidence that supports limits on injection. That dual position is disclosed here because it is the right frame for reading this work: the analysis was built by someone with the operational context to get the domain details right and no incentive to overstate the seismicity risk. Where the evidence is weak or unstable, this manual says so explicitly — see the Evidence Scoreboard, which documents estimate revisions across data vintages, including an artifact we found in our own cross-validation machinery and the fix.

No funding was received from any operator, regulator, or advocacy organization for this analysis.

1.3 Project Geminae

Project Geminae is an independent research effort applying modern causal inference — targeted learning, the highly adaptive lasso, and double-machine-learning methods — to high-stakes industrial decision problems. The induced-seismicity program produces:

  • This field manual — the policy and operations translation.
  • The IRT dashboard — per-event, per-well causal attribution on live data.
  • The technical paper — the SPE-facing methods contribution (regHAL-TMLE Delta-method inference and a GPU active-set solver for full-panel highly-adaptive-lasso estimation).
  • The codebase — open at https://github.com/Project-Geminae/induced-seismicity.

1.4 A note on attribution

This manual is one part of a research program in active development. Where future editions include findings produced with external collaborators, methods reviewers, or partner institutions, the author block will reflect that. For the current edition, the authorship is Matthews.