Focusing the P3: the ring reports nothing, so read focus from the image

No sensor: five places that would show a focus reading show none

This is evidence from absence: nobody has turned the ring while logging. The focus meter's exit report does exactly that check (below).

Read focus from the image: a half-pixel blur halves the score

Drop in each score when a live P3 frame is blurred on the computer; the score's own wobble over 8 s was 0.4%.

python3 ~/git/thermal-android/tools/p3_focus_meter.py --url http://<phone>:8080

   9.7s      6549.3  ##############################  99.9% of peak  norm 0.0258
Peak smoothed score 6556.7 over 248 frames.
No unexplained metadata word changed during the run.

• Every frame is scored. Averaging frames first doesn't help: the 0.4% wobble is slow drift, not frame noise. The display smooths over about 0.3 s.
• Watch "norm" in long sessions. It divides by the scene's contrast, so a board warming up doesn't move it.
• It needs texture. The test scene had 4.5 K of spread. An evenly warm board gives the score nothing to sharpen, so focus on a contrast target.
• Its exit report is the sensor test. Turn the ring end to end during one run. If no unexplained metadata word changes, the no-sensor finding becomes a measurement.

At 45–70 mm the sharp zone is 4–12 mm deep, so fix the distance first

4.3 mm lens (the serial's lens code), 12 µm pixels, f/1.0–1.1, 2-pixel blur with diffraction included; feature = 2 pixels. The focus simulator uses 1 pixel: ±2.2 mm at 60 mm.

• No single ring setting covers 45–70 mm. The best compromise still blurs the ends by 6–7 pixels, so the mount must set the distance.
• Two pixels of blur is the floor. At f/1.0 and 10 µm light, diffraction alone spreads a point over about 2 pixels.
• The "focuses to 8 mm" claim is unmeasured. It would need 5 mm of lens travel, more than the focal length; nobody has checked it.
• The f-number is soft. f/1.0 appears only as the noise-test condition in the quick-start guide. The depths above allow for f/1.1.

Focus warm: a hot camera focuses closer

Optics constants from Riedl via the University of Arizona OPTI 521 report. Rule: set or check focus after the camera has warmed up where it will work.

The mechanical check: a fixed distance and a witness mark

The ring carries no scale or index mark, so out of the box there is nothing to read. But reviewers find it smooth, damped and free of backlash, so a mark you add stays meaningful.

1. Pick one datum and gauge it, not by eyeThe ring's front rim is the only surface you can touch; the lens sits recessed behind it. Set lens-to-board distance with the printed spacer or bracket, not a ruler.
2. Mark the ring and the body togetherOnce focused, draw a paint-pen line across the joint. Any slip later shows as a broken line.
3. Optional: make your own distance scaleFocus on targets at 45, 55 and 65 mm and mark each on the ring: an empirical scale for this unit.
4. Lock it reversiblyA split collar over ring and body, as the hub's mount plan has. Never glue or thread-lock an optical helicoid.

The procedure: focus on the bench at the real distance, then confirm in place

1. Warm up for about 10 minutesThe shutter clicks space out once the camera reaches balance. Warm it near its working temperature if you can.
2. Freeze the image pipelineFixed noise reduction (both at 0 is best), auto-FFC off for the sweep, one manual shutter just before. In the official app, X³ off: it fakes edges.
3. Put a contrast target at the exact distanceSquares of aluminium tape on something warm and matte, such as a mug of hot water or a powered board, tilted a few degrees. The foil mirrors the cool room while the surface around it glows: a sharp step in one plane.
4. Sweep through the peak, then come back to itTurn slowly until the score falls, back off, and return from the same side. Let the ring settle a moment before reading.
5. Mark, lock, mount, checkWitness mark, collar, into the gap. If the board shows texture, the score in place should match the bench.