PROM v5.3 Validation Report

This report was measured on the production solver server on September 8, 2026, using PROM 5.3 / prom-rs/0.4.0. It replaces the older report as the current benchmark. The propeller and motor results were calculated again on the installed production engine; they were not copied from the 5.0 report.

The capture covers 75 propellers at 9,842 operating points and 98 motor models across two bench datasets. The public motor envelope below contains 95 T-Motor models. Aggregate results from the separate bench dataset are shown separately. All 9,842 propeller points converged.

What was compared

For each propeller, the model receives its catalog geometry, the measured RPM and the wind-tunnel airspeed. We compare predicted thrust and shaft power with the measurements at those same conditions. This capture follows the curated catalog path, including the stored geometry normalization used by the production model.

For motors, measured shaft speed and torque are prescribed, and the native model predicts battery-side current. Full-throttle rows also compare predicted maximum shaft speed with measured RPM. The benchmark uses production defaults for an unspecified ESC: six-step control and medium timing. The benchmark's motor and ESC parameters were checked against the production component builder before scoring.

These are component benchmarks. They do not measure the accuracy of a coupled powertrain, battery discharge, temperature prediction, aircraft flight time or an autopilot mission.

Reading the results

• MAE is the average absolute difference. It is not a maximum-error bound.
• RMSE gives greater weight to larger errors.
• Bias shows the direction of the average error. Positive means over-prediction.
• R² describes how much of the measured variation the prediction explains. It is not a percentage-accuracy score.
• Propeller errors use dimensionless thrust and power coefficients. The summary page shows mean absolute error, the 95th-percentile absolute error, performance curves, parity and cumulative error distributions. Plot statistics include all selected points, including zero and negative coefficients. Every operating point has equal weight. Percentiles use the nearest-rank rule.
• Motor current and RPM errors are percentages. Efficiency errors are percentage points.

The original test groups are kept separate for comparison with earlier reports. This is a regression benchmark, not a newly blinded validation study. These datasets have been examined during model development. “Separate test group” and “Performance test group” refer to their original partitions, not a claim that the current model has never been evaluated against them. The “Development group” is included below so the full campaign remains visible.

Validated envelope

These ranges describe the measured components and conditions in this campaign. They do not establish accuracy for every combination within those ranges or for unmeasured aircraft configurations.

Benchmark results

Propeller: performance test group

APC Thin Electric propellers from the original performance test partition.

Propeller: separate test group

APC Sport and Slow Flyer propellers from the original separate test partition.

Propeller: development group

Motor: performance test group

Motor: separate bench dataset

Motor: development group

Per-case tables

Propellers

Motors

Known limitations

The propeller differences vary substantially by family and operating condition. The summary's family and condition filters show that variation; its full-dataset statistics include all original test and development groups. The tables here retain those groups separately. Individual errors can be larger than the mean or the 95th percentile. The new model does not reproduce the earlier report's error figures.

Motor current error averages 7.2% in the performance test group and 9.6% in the separate bench dataset. Full-throttle speed error averages 6.1% and 14.8%, respectively. The larger error on the separate dataset is relevant when considering a motor outside the better-represented group.

All propeller measurements in this campaign use axial inflow. This report does not establish measured accuracy for edgewise flight, tilted inflow, rotor interaction, coaxial stacks or steep descent. The model can evaluate additional configurations, but calculation support and measured validation are different things.

Battery discharge, heating, complete powertrain operation and aircraft missions are not part of this measured dataset. These component statistics must not be presented as flight-time error bars or an aircraft-level accuracy guarantee.

User-supplied and custom components

The error figures above are for the known catalog components in this campaign. They do not apply to user-supplied geometry or specifications. The curated geometry path is part of this capture; a custom propeller does not automatically inherit that provenance. Check the accuracy of your inputs and measure the final setup before relying on an absolute performance prediction.

Capture provenance

The run used the installed production wheel, the selected consistent BEM with the original ncrit=9 section policy, and the production catalog geometry path. Its application commit was e07405697a1d237acf171cd90b0261c2e7b6d9cb. The artifact records the native binary and wheel hashes, model and geometry hashes, corpus hashes, source hashes and capture timestamps.

The propeller aggregate errors were independently recomputed from the saved measurement/prediction pairs on the solver server. All 18 propeller metric rows matched the published rounding. The capture also passed checks for engine identity, finite predictions, sample counts, convergence accounting, motor parameter agreement and the per-case publication rules.

The PROM 5.0 report remains available as historical evidence. Its numerical limitations are not substituted for the measurements in this report.

Data sources

• UIUC Propeller Data Site, Volume 1 — APC Thin Electric / Slow Flyer / Sport — Brandt, J. B. and Selig, M. S., 'Propeller Performance Data at Low Reynolds Numbers', AIAA 2011-1255, January 2011; Brandt 2005 UIUC M.S. thesis. (revision: e60b6ff97f78)
• UIUC Propeller Data Site, Volume 4 — APC Thin Electric — Dantsker, O. D., Caccamo, M., Deters, R. W. and Selig, M. S., 'Performance Testing of APC Electric Fixed-Blade UAV Propellers', AIAA 2022-4020, June 2022. (revision: e4c6b24c399e)
• T-Motor published bench dataset (curated corpus) — Manufacturer-published product-page test tables, curated and energy-consistency filtered. (revision: bcd9bcaba479)
• Independent third-party motor bench — Commercial laboratory dataset; source withheld under its terms of use. Used for aggregate metrics only. (revision: cb506f184793)

Motor benchmarks use measured electrical and mechanical performance from the committed T-Motor corpus and the separate motor-bench dataset. Their scoring conditions and test groups are kept distinct above.