A sudden chill crawled down the corridor of the Felt Mansion at 12:02 a.m. My 590 nm-converted Nikon began a thirty-second exposure while the pocket-sized FLIR traced ghost-white heat around a locked door. Twenty seconds in, the Trifield meter chirped and the Arduino data logger stamped a 6.8 mG spike on the EMF channel. Whether that shiver along my spine meant anything more than nerves would depend on what the data said once every sensor’s timestamp lined up.

Parts One and Two of this series explained why infrared matters for paranormal work and promised a testable workflow. This third installment delivers a complete experiment you can pack in a messenger bag and run in any location, moving you from anecdotal “orbs” to correlated, time-synced evidence.

First, the gear. A 590 nm-converted mirrorless or DSLR body such as a used Nikon Z fc handles high-resolution stills; a clip-in 720 nm filter gives a control wavelength you can swap in hourly. A pocket FLIR ONE Edge Pro provides surface-temperature video, while a Trifield TF2 EMF meter captures electromagnetic changes. An inexpensive Zoom H1n recorder running 96 kHz WAV handles audio anomalies and clap-board syncs. The experiment’s heartbeat is an Arduino Nano 33 IoT fitted with a GPS module and MicroSD card. By feeding the GPS PPS pulse to the Nano, every logged value—EMF, ambient temperature, camera trigger—shares millisecond-level Universal Time. One 20 000 mAh USB-C power bank keeps everything running for a full-night vigil, all strapped to a cold-shoe rail atop the camera. If you need to economize, swap the FLIR for a $45 AMG8833 thermal sensor on an ESP32-Cam and replace the Trifield with a DIY Hall-sensor probe; the workflow still works for under $200.

Rig assembly is straightforward: mount the cold-shoe rail on the camera’s hot-shoe, clamp mini-ballheads for the FLIR and EMF meter, Velcro the Arduino under the rail with the GPS antenna clear of metal, and keep all cables short to minimize radio-frequency noise. The Nano logs every sensor reading while flashing a 3.3-volt trigger pulse into the Nikon’s remote jack, stamping each frame with a clear “shutter” flag in the CSV file.

Before any investigation, capture a dark frame with the lens cap on and a flat-field shot of the twilight sky to iron out sensor quirks. Next, collect a fifteen-minute baseline in an empty control room: run the Nikon timelapse, the Trifield, and the FLIR while noting humidity, barometric pressure, and outside temperature. These baselines are crucial for separating genuine anomalies from gear noise.

The field procedure repeats every twenty minutes. Spend five minutes on a locked-down IR timelapse, two minutes panning the FLIR, and one minute with the Trifield sitting motionless; then use the remaining twelve minutes to move to the next room, hydrate, and make notes. If your location has a safe control room, keep a duplicate rig running there with nobody inside. Hourly, swap from the 590 nm setup to the 720 nm filter to catch wavelength-specific effects.

Back home, import RAW files into Lightroom, apply a 590 nm profile, and export mid-resolution JPEGs. Use FLIR Tools to convert radiometric video into per-frame CSV data, and pull the Nano’s log_20250617.csv from the SD card. A short Python notebook can merge the three streams with Pandas, compute z-scores, and flag any moment when an EMF spike and an IR-brightness jump coincide within a one-second window. Plotting EMF z-scores against IR z-scores instantly shows whether anything unusual clusters in the top-right quadrant—your best candidates for follow-up.

When interpreting results, look for events that appear on two or more sensors at once; single-sensor spikes are usually environmental artifacts or operator movement. Watch, too, for regular patterns every twenty or thirty minutes, which often trace back to HVAC cycling. As always, keep a skeptic’s hat on: absence of evidence is not evidence of absence, but sensor artifacts certainly are not phantoms.

Safety and ethics matter more than any data set. Obtain written permission for every site, mind asbestos and lead paint in abandoned buildings, and never trespass. Respect property owners and avoid antagonistic tactics that make ghost-hunting videos go viral but erode credibility.

If your budget is tight, an ESP32-Cam with an AMG8833 thermal array, a used Canon Rebel lacking an IR-cut filter, and a DIY EMF probe bring the entire experiment under two hundred dollars, sacrificing some resolution but preserving the correlation principle.

I have bundled a field-data CSV template, the full Arduino sketch, and the Jupyter notebook into a starter kit you can download here: Download field-data starter kit. Run your own vigil, crunch the numbers, and share your plots—tag @EovaldiArtScience or drop a link in the comments. Together we can push paranormal investigation from legend toward evidence-based storytelling.

Happy hunting, and may your sensors make your spine tingle for scientific reasons.