Hardware bring-up round (v1.57+): the 7-item flight-readiness register

[avrabe]

The v1.45→v1.56 round closed PX4 capability parity in simulation (see docs/PX4-PARITY-ASSESSMENT.md, refreshed in #156). What remains is flight-proven maturity — the boundary between "provably correct in SITL" and "flown." This epic tracks the 7 register items; each is an assignable slice others can pick up independently.

1. Real driver bodies on silicon — beyond the 4 we have (falcon-imu-icm42688, falcon-gnss-ubx, falcon-esc-dshot, falcon-baromag mock-bus). Real register transactions on a board.
2. Sensor calibration — accel/gyro/mag/baro calibration routines on a real board (not SITL-injected).
3. Per-airframe flight tuning — the SE(3)/ADRC math is proven; gains for a real frame are not. Cascade co-tuning (position loop is bandwidth-limited by attitude — see project memory).
4. WCET leaf measurement — the compositional WCET proof is done; cycle counts on the target MCU are not measured.
5. Physical HITL transport — UART/USB/UDP end-to-end (falcon-hitl exists as a seam).
6. libm qualification — transcendental math is on the flight path; centralize behind a math seam (relay-math) before qualifying.
7. First flight — the irreducible field test. Falsifies (or confirms) everything above.

Plus the integration project: meld→loom→synth→gale onto the board. The v1.56 component bundle (falcon-components-v1.56.tar.gz) is the input to that hand-off.

Sequencing note: these are NOT all serial — items 1/2/5/6 can proceed in parallel; 3 depends on 1+2; 4 depends on 5; 7 depends on everything. Pick a slice, open a child issue, ship it through the feature loop as its own release.

🤖 Generated with Claude Code

[avrabe]

Validity check (2026-06-13) — register confirmed accurate, item 6 advanced.

Re-validated all 7 items against the current tree:

• Items 1, 2, 3, 5, 7 (drivers-on-silicon, calibration, per-airframe tuning, physical HITL transport, first flight) — still accurate, genuinely hardware, correctly postponed.
• Item 4 (WCET leaf measurement) — still open; the compositional proof stands, cycle counts on the MCU remain unmeasured.
• Item 6 (libm qualification) — the SEAM HALF IS NOW DONE (feat(math): centralize libm across the full flight path via the relay-math seam (#157) #164, v1.57 round): relay-math centralizes all flight-path transcendentals and is the sole libm consumer across cargo and bazel (a stray libm:: in a flight crate no longer compiles). So item 6 narrows to discharging the qualification at the one seam site (proven polynomial core / CMSIS-DSP / CORDIC / measured conformance) — the software prerequisite is complete; the qualification itself is the remaining work.

Integration note: the meld→loom→synth→gale hand-off input is now falcon-components-v1.57.tar.gz (the bundle is regenerated per release by the #163 release job).

Epic stays open and postponed per the current direction (software/P3 arc first). No item is stale.