Validation
What is measured, what is targeted, and what is still unknown.
This page separates engineering goals from measured evidence. Early TWR runs are reported here with their hardware, firmware, geometry, and limitations.
July 2026 baseline
What the current runs show
| Distance | Valid samples | Mean error | P95 absolute error | Valid rate | Packet success |
|---|---|---|---|---|---|
| 1 m | 3,125 | +4.9 mm | 28.0 mm | 52.08 Hz | 99.71% |
| 3 m | 9,245 | −57.2 mm | 77.9 mm | ~51.4 Hz | 97.70% |
| 5 m | 9,256 | +54.8 mm | 77.3 mm | ~51.4 Hz | 97.88% |
Conditions: one hardware pair, indoor clear line of sight, same antenna orientation, tape-measured ground truth, and one responder calibration preserved across the sweep. The 1 m row is the calibration-point run; 3 m and 5 m pool three approximately one-minute runs each.
Firmware: the original twr_init-2 / twr_resp-2 50 Hz profile. P95 absolute error means 95% of valid samples were within the stated magnitude of ground truth.
Profile experiments: longer preamble and SFD configurations raised packet delivery to roughly 99–100% at 40 Hz, but increased static P95 error at 5 m. The current twr-5 release is an experimental diagnostic profile, not a new benchmark result.
These short runs are preliminary engineering evidence, not a product guarantee or a completed public benchmark. Raw datasets will be linked after the test metadata and naming are normalized.
V2 engineering goals
Targets under test
Targets are acceptance goals, not guarantees. Final results must identify hardware revision, firmware revision, environment, geometry, and sample count.
Reproducible TWR test
Static ranging protocol
- 01
Fix the geometry
Mount both tags at the same height with clear line of sight, consistent antenna orientation, and no movement during each run.
- 02
Measure ground truth
Use a laser distance meter between documented reference points. Record the environment and nearby reflective surfaces.
- 03
Calibrate once
Calibrate the pair at a known distance, then preserve the same offset across the complete distance sweep.
- 04
Collect repeated runs
Use the browser Static Test at 1 m, 3 m, 5 m, 10 m, and 15 m. Capture at least three runs and 500 valid samples at each distance.
- 05
Export untouched CSV
Keep the original timestamped export. It contains measured distance, truth, signed error, RMSE, deviation, and percentile errors.
- 06
Report failures too
Publish dropped exchanges, obstructions, outliers, firmware identifiers, and anything that changed between runs.
Publication contract
What every result must include
- Raw, unedited CSV export
- True and measured distance
- Mean signed error and RMSE
- Standard deviation and P95 absolute error
- Valid sample count and dropped exchanges
- Board, firmware, placement, and environment notes
A result is not promoted to the product page until the source data and test conditions are linked here.