Hardware-test Face Priority EV compensation

[erik-sutton95]

Finish hardware testing of Face Priority EV compensation. The feature is implemented; this thread is the close-out matrix, not a redesign.

Related: D-Log M mapping (Nano linearize) · roadmap index

What shipped (unreleased changelog)

• Core: FacePriorityExposure — Vision boxes on the encoded live tap, 18% gray, median of per-face medians
• Deadband 2/3 stop, max 1 third-stop per setEv (0x02/0x2E)
• Acquire 2.5 s at 0.4 s, then 1.0 s settle
• Gated to Auto expo; EV drum disabled while on; face.smiling.fill badge
• Restore previous EV (or 0.0) on disable
• Separate from Face Lock / ActiveTrack (LiveFaceAF, 0x02/0xA6 / 0x89)

Test matrix (Pocket 4 / 4 Pro, physical device)

• One face, Rec.709, indoor mixed light — EV walks to skin at middle gray and stops hunting
• Two faces, different IRE — uses median of per-face medians, not the brighter forehead
• Face leaves frame — no EV writes; restore on toggle-off
• D-Log and D-Log2 — 18% is paper IRE (D-Log2 ~30.50), not Rec.709 50
• AF-C + Face Lock together — EV still writes, gimbal lock does not fight the meter
• Manual expo — toggle hidden / no-op
• No face — no writes (log it)
• Rec start/stop mid-acquire
• Disconnect / reconnect — no stuck EV

Nano

• Confirm Face Priority is even offered (no AF; Vision can still see faces on the tap)
• Block or retune until D-Log M transfer is real (today D-Log M linearizes as Pocket D-Log)

Product nits (only if testing says so)

• Persist default-on for interview setups
• Show target IRE / face luma on the EV sheet
• Operator Setup row, not only the EV sheet

Done when

A short hardware note (even a comment on this thread) with Pocket 4 Pro pass/fail for the matrix above. File bugs as Issues if something hunts, overshoots, or writes in Manual.

[erik-sutton95]

Tracked as maintainer QA task #91.