TrailFix: Browser-Based GPS Post-Processing Without the $1,000 Upgrade

The Bad Elf Flex Mini Standard and Extreme use identical hardware — same GNSS chip, same dual-frequency L1/L5 antenna. The $1,000 price difference is a firmware unlock for RTK and improved SBAS. For hobby use like trail mapping and dirt bike routes, I wanted to see how close I could get with post-processing and free reference data. The answer: pretty close.

Try TrailFix →

The Processing Pipeline

TrailFix runs entirely in the browser — no server, no uploads, no accounts. You drop a GPX or NMEA file and it runs through four stages:

Speed filter removes GPS spikes. Consumer GNSS receivers occasionally report positions that jump hundreds of meters between readings. The filter discards any point that implies movement above a configurable threshold (default 50 mph for trail use, adjustable up to ~93 mph for faster activities). This catches the worst outliers before they corrupt downstream processing.

Stop detection and averaging finds clusters of points within 5 meters of each other — places where you paused on the trail. Instead of keeping dozens of scattered readings at each stop, it averages them into a single high-confidence point. More samples at the same location means better accuracy, and this is where consumer GPS data actually has an advantage: all that redundant data at stops becomes useful.

Kalman smoothing reduces jitter on moving segments. The filter estimates the true position by weighing each GPS reading against a predicted position based on the previous trajectory. The strength is adjustable — higher values produce smoother tracks with less noise, lower values preserve more of the original detail.

CORS differential correction is the big accuracy gain. NOAA operates a network of Continuously Operating Reference Stations with precisely known positions. By comparing what a CORS station observed versus where it actually is, you get a correction vector that applies to nearby receivers. TrailFix downloads RINEX observation data directly from NOAA’s AWS mirror, parses it, and applies code-phase corrections scaled by your distance from the station.

CORS Accuracy by Distance

The correction quality degrades with baseline distance, but the results are solid within 50 km:

< 20 km from station: ~0.3–0.7 m accuracy
< 50 km: ~0.5–1.0 m
< 100 km: ~0.7–1.2 m

The app ships with ~18 hardcoded CORS stations across WA/OR/ID/BC, sorted by distance from your track. Click download, drop the .gz file (auto-decompressed via pako), and apply.

Visualization

Three views for inspecting results:

Canvas map — lightweight 2D rendering showing raw track (gray), smoothed track (green), removed spikes (orange), and CORS-corrected points (blue). Supports pan and zoom.
Satellite map — full MapLibre GL or Mapbox GL integration. Free ESRI satellite tiles work out of the box, or paste a Mapbox token for their imagery.
Points table — every point with coordinates, raw vs. improved position, and status flags.

Export

Processed tracks export to three formats, all named with the original filename plus a _trailfix suffix:

KML for Google Earth
GPX for other GPS software
GeoJSON for GIS tools and web maps

Why a Single HTML File

The entire app is one 68 KB HTML file. No build step, no npm, no framework. You can download it, put it on a USB drive, and open it offline (the only network requests are optional: CORS data from NOAA and map tiles). External dependencies are just pako for gzip decompression and the map libraries, loaded from CDNs on demand.

This was a deliberate choice. GPS tools shouldn’t require an internet connection to function, and the people most likely to use this — outdoor hobbyists mapping trails — are the ones most likely to have intermittent connectivity.

What’s Next

The processing pipeline has room to grow:

Multi-pass averaging — ride the same trail multiple times, average across all passes for even better accuracy
Elevation profiles — visualize climb and descent along the track
Speed and distance analytics — leverage GPX timestamps for pace, moving time, and segment breakdowns
PWA support — service worker for true offline use with cached map tiles

The source is on GitHub and the live app is at adavenport.dev/trailfix.