Pre-registration
Hypotheses, classification rules, test scripts, data snapshots and pass/fail criteria are written, hashed and publicly registered first. Any change afterward is a logged deviation, never a silent edit.
Exauren · Applied Research for Complex Environments
Does the public record already know where the next serious injury will happen?
I run a research program on public enforcement records, beginning with U.S. occupational safety. OSHA documents hundreds of thousands of workplace hazards a year: a missing guardrail, an unshored trench, exposed conductors, an unguarded machine. Each is energy that has not yet been released. When an employer has been documented with one, is a serious injury or death from that same energy more likely to follow, beyond what size, industry and prior injuries already predict?
Pre-registered
Energy-Specific Recurrence
The lead study tests whether documented uncontrolled high-energy exposure at an employer (identified through a prior OSHA citation referencing a standard that establishes an energy threshold, or through a prior serious injury or fatality) predicts subsequent serious injury or fatality involving the same energy class, beyond the predictive contribution of prior injury history, establishment size and industry.
| Registration | Open Science Framework, t6gds, public since 15 October 2026 |
|---|---|
| Design | Retrospective employer-level cohort on national OSHA inspection, citation, accident-investigation and Severe Injury Report records, 2005–2026; five energy families (fall, struck-by, caught-in, electrical, trench); placebo citations and negative controls; out-of-time validation |
| Data | Public records only, acquired as a single-day snapshot whose hashes are registered. No worker or victim identifiers enter any analysis. |
| Status | Registered analyses complete; manuscript in preparation for the American Journal of Industrial Medicine |
| Extension | A NIOSH exploratory research proposal in preparation, asking what further predictive information the public record holds but does not expose: how much energy was present, what the inspector judged, and what control was installed |
| Investigator | Michael Beauchamp, ORCID 0009-0009-6039-3970 |
How the work is done
Written down before anyone looks.
Reproducibility
Analyses run in a pinned computational environment on the registered snapshot. A rerun reproduces the registered output byte for byte.
Symmetry
Results are reported against the registration, whether supported, inconclusive or withdrawn. A hypothesis that fails its test is withdrawn in print.
The program measures and reports. It does not adjudicate, and it does not identify individuals. Compliance and enforcement determinations remain with the institutions authorized to make them.
Selected portfolio
Academic work samples
These projects were completed as coursework in the MIT Professional Education Applied Data Science program. They are examples of model development and evaluation through public notebooks and recorded outputs, separate from the research program.
Computer vision · Image classification
Malaria Detection
Classification of parasitized and uninfected red-blood-cell images. An academic exploration of image preprocessing, neural-network models and evaluation; not a clinical diagnostic system.
Deep learning · Digit recognition
SVHN Recognition
Digit classification using Street View House Numbers image data. The notebook explores dense and convolutional neural networks and records differing model outcomes.
Background
An unusual route to the question
I was a PLC programmer before business programming became my career. I spent most of a decade and a half building analytics and data infrastructure inside a Fortune 500 supply-chain organization, with earlier data engineering on healthcare claims and higher-education systems. Along the way I studied as much mathematics as a working life allowed.
What that background gives the research is an engineer's habit of building the measurement before trusting the number.
Professional experience on LinkedIn →- M.S., Predictive AnalyticsNorthwestern University
- B.S., MathematicsNortheastern State University
- Applied Data Science Program CertificateMIT Professional Education
Collaborate
Conversations the registered work has opened.
Occupational epidemiologists and biostatisticians; energy-based safety researchers; safety-technology providers who want their methods tested rather than assumed; institutions that sponsor this kind of work. Knoxville, Tennessee.
- Research inquiries
- michael@exauren.ai
- Registration
- osf.io/t6gds
- ORCID
- 0009-0009-6039-3970