SSD1333 brightness: photometer harness, four-axis driver, and measured response surface (stages 1+2)

[brickbots]

Summary

Stages 1 and 2 of the SSD1333 brightness re-characterization: the photometer harness + four-axis driver + ADR 0023 revision (stage 1), and the panel's photometrically measured response surface (stage 2) — raw journals, the sweep specs/scripts that produced them, the measured-model writeup (docs/ax/display/ssd1333-response.md), CONTEXT.md vocabulary, and the stage-3 handoff (docs/ax/display/HANDOFF-stage3-dimming-policy.md).

Stage 1 (commit feat(display): …)

• python/PiFinder/panel_photometry.py — camera-as-photometer rig (selftest/measure/sweep/monitor, period-locked exposure ladder, JSONL journals).
• ssd1333_device.py — emitted-light-space ceiling LUT + pre-charge voltage command; displays.py — four-axis set_brightness (bounded fit search, canary floor, pre-charge regime, knee curve); precharge_sweep.py — by-eye harness kept for visual artifact checks.
• ADR 0023 four-axis revision; test_ssd1333_brightness.py (12 unit tests, passing).

Stage 2 headline findings

• Contrast and master are one axis: flux depends on the two registers only through the drive product P = contrast × (master+1) (single-valued within 1.6% across independent sessions). Dark below P=4, cliff to 28.5% of reference, near-flat to P≈32, ≈√P above P≈128.
• Separability fails: pre-charge spans 32× at low drive but 1.29× at reference; cut-out edges move with operating point. The multiplicative model — and its predicted 18,000:1 span — was structurally wrong (measured whole-policy span: 233:1). Fill-fraction probe rules out APL/current-limiting: this is true register behaviour.
• Ceiling duty law (n−1)/30 confirmed to ~1% — the one axis the model got right.
• "Very dim" is reachable: 148 ADU/s at (contrast 4, master 0, ceiling 4, pre-charge 4) — ~3.8 decades within the tonal-range rule. The shipped policy's dimmest state sits 40× above the panel's floor.
• Value remap math validated at reference drive; the value-64 canary stays photometrically distinct at every measured operating point; "pre-charge glow" found under native level 1.

ssd1333-response.md §7 flags (does not fix) the ADR 0023 claims that fell; amending the ADR is stage 3's job alongside the policy refit — see the handoff.

Verification

• pytest tests/test_ssd1333_brightness.py: 12 passed. ruff check on all touched Python: clean.
• Pre-sweep selftest: all gates PASS (darkness, ladder, repeatability CV 0.71%, dark stability); sentinel drift ≤0.9% within every run; journals carry the sentinels.

🤖 Generated with Claude Code