The extension's hardware measurements are separate from my statistical analysis of the team's STM32L432 reference-board dataset. That analysis used permutation tests and Benjamini–Hochberg adjustment; it is not a Buck/LDO benchmark of this extension. The reference dataset remains private.

STM32L4P5 measurement example, 3.6 V condition. A waveform observation, not a claim of regulator efficiency or average energy savings.

The system operated on a physical setup. The published snapshot includes MQTT retries, but has not been re-tested during portfolio preparation. The source diagram uses CO₂ GPIO 15/16 while the code uses 4/5; TLS certificate verification and Wi-Fi recovery also need further work. These limitations are documented in the repository. No measured cost or power savings are claimed.

The BRAM core uses a sequencer and iterative division, so it is not uniformly single-cycle. Historical timing and utilization results come from separate runs and were not regenerated for this portfolio. Physical FPGA-board operation, full ISA compliance and power-efficiency rankings are not claimed. The diagrams describe the team's design, not sole authorship.

Results are reported by the team, not independently rerun. Evaluation represents 64 classes; related video frames across splits remain a possible source of optimistic scores. Test mAP@0.5:0.95 was not retained. Webcam FPS observations are not a standardized benchmark, and Raspberry Pi 5 deployment remains planned. Street photographs are withheld pending image-provenance review.