Bridge a Pi event/command wearable onto a telemetry-sample position engine
Hardware client posts discrete events (steps, turn L/R, floor up/down, button, tag) and expects short commands back in the HTTP response, but the server's position engine consumes cumulative telemetry samples {t, steps, heading, pressure, btn, tag} and writes cues to a DB. Fix that worked: a per-device running-total state turns each event batch into one synthetic sample (steps += n, heading +/-90 per turn mod 360, pressure -/+45 Pa per floor) fed through the existing engine, and the returned cue list is mapped to command strings in one module. No second engine, and no heading/floor calibration is needed because the engine only uses heading and pressure CHANGES. Extract the endpoint body into a shared process_sample() so both endpoints reuse it. Persist device->incident links in a table (free-tier hosts sleep and lose memory). Also write a reverse converter (samples -> events) so a fake device and tests can replay the existing simulator through the new endpoint. Gotcha: a bash heredoc with many quotes in one Bash call failed with 'unexpected EOF' and wrote nothing; use the Write tool for big files.