Define the research claim before collecting the study

Complete with the supervisor and, where applicable, clinical collaborators. This is not a prescribed clinical protocol.

Decision	Write before collection	Why it matters
Primary question	One testable AFO or neuromuscular hypothesis	Avoid choosing an attractive result after analysis.
Primary outcome	Exact feature, units, time window and normalization	“Muscle activity” is not a complete outcome definition.
Meaningful difference	Smallest effect worth detecting	Determines required noise, repeatability and timing performance.
Reference	Independent gait-event, angle or task reference and its uncertainty	IMU-estimated events cannot validate themselves.
Participants and unit	Participant/session/trial structure; eligibility under protocol	Thousands of samples from one person are not thousands of independent participants.
Conditions	AFO configuration, footwear, walking speed, order and rest	These can change signals independently of the intervention.
Placement	Anatomical landmarks, spacing, side, mounting axes, repeatability	Sensor changes can masquerade as physiology.
Calibration	Electrical calibration, IMU bias/alignment, EMG normalization	Physical units alone do not establish comparability.
Timing budget	Allowed alignment uncertainty for the intended outcome	200 Hz motion does not provide arbitrarily precise event timing.
Quality rules	Saturation, dropout, artifact and missing-label exclusion rules	Make decisions before examining condition effects.
Analysis	Filtering, windowing, model, repeated-measures treatment	Preserve participant independence and avoid data leakage.
Data protection	Pseudonyms, access, retention and institutional approval	Research recordings and clinical information may be sensitive.

Examples to discuss, not approved studies:

AFO comparison: normalized TA activation during independently labeled swing, with controlled speed and within-participant conditions. Requires validated event labels and reliable normalization.
Neuromuscular coordination: TA/medial-gastrocnemius coactivation under defined tasks. Requires an explicit coactivation formula and consideration of electrode placement and physiological crosstalk.
Movement monitoring: foot/shank motion features during a prescribed task. Requires mounting calibration and appropriate reference validation; do not call inferred ankle angle a direct measurement.

For machine learning, split by participant when claiming generalization to new people. Keep overlapping windows from the same trial together; otherwise train/test leakage can inflate results. Determine sample size from the chosen outcome and study design, not the 8 kHz sample count. A pilot estimates feasibility and variance; it does not establish diagnostic accuracy.

What this recorder fixes before you choose an outcome

Signal	Fixed by the build	Research consequence
EMG sampling	8,000 simultaneous frames/s; one code = 152.588 µV at the ADC input (±5 V range)	Ample for surface-EMG bandwidth; values are MyoWare RAW-output volts, not electrode-referred microvolts
EMG analog path	Two buffered 3.3 kΩ/47 nF poles (each about 1,026 Hz; combined −3 dB about 660 Hz, −1.85 dB at 500 Hz) plus the MyoWare's own RAW filtering	Report both filters; frequency features (for example median frequency) depend on them
IMU sampling	200 Hz per sensor, ±16 g and ±2000 °/s (2,048 counts per g; 16.384 counts per °/s)	Samples are 5 ms apart; event timing from motion alone has millisecond-scale uncertainty
IMU filtering	On-chip UI filter, second order, bandwidth index 4 (`GYRO_ACCEL_CONFIG0 = 0x44`). the datasheet defines it as max(400 Hz, ODR)/10, i.e. 40 Hz at 200 Hz	Suitable for segment kinematics; it attenuates sharp heel-strike impact transients. If impacts are an outcome, the ODR and filter must change and be revalidated
IMU time	Sensor clocks are independent and may be off by up to ±1.5 % at 25 °C (±2 % over temperature); FIFO ticks last 32/30 µs; host times are estimates, so use `host_time_fitted_us`. The same clock error scales gyroscope rates unless calibrated	Measure offset and drift against a reference before claiming EMG–motion timing

Bench timing-alignment check (before participants)

Choose one physical event that a reference system and the recorder both see, for example a sharp tap on the foot carrier recorded by a reference accelerometer or motion-capture marker. For EMG timing, inject a generator pulse into the synthetic-input fixture while the reference system logs the generator's sync output. Bench only; no person attached.
Repeat the event at least 20 times across a 10-minute recording.
For each event, record the reference time and the recorder time (EMG conversion_trigger_host_us, IMU host_time_estimate_us). Fit offset and drift; report the residual spread as the alignment uncertainty.
Enter the results in the session manifest's synchronization fields (offset_us, uncertainty_us, drift_test_file) and compare them with the timing budget written above.

Sample size: an illustration, not a calculation for your study

For a within-participant comparison (two AFO conditions, paired t-test, two-sided α = 0.05, power 0.80), the required number of participants depends on the standardized effect dz = mean paired difference ÷ standard deviation of the differences:

dz	Participants
0.5	34
0.8	15
1.0	10
1.2	8

Estimate the standard deviation of the differences from a pilot or prior literature, using your chosen outcome, normalization and placement procedure. Add allowance for dropouts and excluded trials, and use a design-appropriate method (for example mixed models with repeated trials) for the final analysis. Trial or sample counts do not substitute for participants.