On the Use of AI in This Project
This project was carried out with substantial assistance from Claude, an AI assistant made by Anthropic. Some readers will reasonably want to know where that line falls, so here it is in detail rather than as a footnote.
Data reduction. The Python pipeline that parses OBD and GPS logs, detects and corrects clock offsets between recording devices, screens sensor dropouts and contaminated position fixes, and extracts qualifying steady-state windows.
Analysis. Regressions, diagnostic gates, error propagation, and the channel-provenance work that established which logged parameters are independent measurements and which are derived restatements of others.
Research. Locating and checking published third-party data — charging network distribution, generator fuel consumption, electric-vehicle towing range, factory specifications — and tracing claims back to primary sources.
Modelling. The comparative simulations on this site, including the Denver–Yellowstone trip comparison, built from stated assumptions.
Writing and construction. Drafting and structuring the text on this site, and building its HTML, CSS and navigation.
None of the data. Every mile was driven, every instrument installed, and every reading logged by me. The measured figures on this site come from an OBD interface and a GPS recorder — they are machine-logged from a moving vehicle, not generated by a language model.
None of the hardware. The umbilical, the installation and the test protocols are mine.
Not the final word on what is true. Claude proposed; I checked against the vehicle, the factory service manual, and the road.
That last point is the one that matters, and the record runs both ways.
Claude found an error in this project's analysis that had survived several revisions: that the engine power and torque channels in the data stream are not independent measurements at all, but restatements of fuel rate at a fixed assumed efficiency. That defect had silently invalidated two separate attempts to measure aerodynamic drag. Finding it cost a result I wanted and was worth more than the result.
I found errors in Claude's reasoning — an incorrect prediction that a purpose-built generator would beat a parked hybrid on fuel per kilowatt-hour, a flatly wrong statement that the trailer lacked DC fast-charging hardware, and an overstatement of my financial relationship with LightShip. Each was caught, corrected, and the correction published alongside the claim it replaced.
An AI that is never corrected is not being used carefully. The withdrawn claims collected on the Roadmap are the audit trail for both of us, and they are kept visible for exactly this reason.
See also: Who is writing this — my relationship with LightShip, stated in full. The withdrawn claims referenced above are collected on the Roadmap.
If you have questions or comments, please visit the LightShip RV Owners Club on RVForums.com and post them to this topic. I monitor that site regularly and would enjoy your feedback.