Testing signal accuracy and visible quality¶

At a glance

HDMI capture can measure signal values. Controlled viewing tests can assess visibility. Neither replaces the other.

Three comparison points¶

Composer: identical decoded BL/EL planes and RPU → reconstructed samples before output quantisation.
Transport: correctly unpacked HDMI carrier → logical component samples and metadata.
Display: emitted light after display mapping, panel processing and dithering.

An operating-system screenshot is not an HDMI capture. A camera pointed at a TV includes the display, optics, exposure and the camera's own processing; it cannot isolate composer arithmetic.

Reproducible video tests¶

Record source SHA-256, container, profile, BL/EL geometry, chroma siting, RPU and presentation timestamps.
Pin app/build/commit, decoder, flags, active mode, GPU, driver, HDMI port, EDID, receiver and display firmware.
Use identical decoded inputs for mathematical comparisons. Dump values before/after EL resampling, reshaping, NLQ, composition and quantisation.
Cover identity curves, pivot boundaries, polynomial/MMR, NLQ dead zones, clipping, chroma edges, EL sizes, missing RPU, seeking and scene changes.
Separate native metadata paths from AI/adaptive modes. Record actual fallback rather than only the menu selection.
Qualify the capture path with known values. Verify carrier bytes, range and chroma remain unchanged; evaluate metadata separately.
Report exact-match fraction, mean absolute error, maximum error, signed bias, percentiles and spatial/temporal error maps per component.

Dither needs both framewise and temporal evaluation. Record pattern and phase: a shifted pattern can create substantial per-frame differences without the same average error. An oracle is a bounded reference, not automatic truth for every commercial player.

Choosing a reference¶

A commercial Blu-ray player is a comparison device, not an infallible mathematical reference. Agreement with it establishes agreement with that device in the tested mode; a mismatch requires investigation rather than an automatic failure verdict for the other player.

Define the expected result independently where possible: known decoded samples for codec tests, specified composition rules for reconstruction, and controlled source values plus metadata for display mapping. Keep the source, output mode and display path fixed. A display-adapted result need not equal the original PQ samples numerically.

Case study: HDR Dissector's content-over-MDL test¶

FIRST-HAND EXTERNAL REPORT; transcript and description reviewed, raw capture and visual method not independently audited. The HDR Dissector's Ugoos review, published 2 February 2025, includes the relevant passage at 5:37–7:44.¹

The author describes a 10,000-nit clipping pattern whose Level 1 maximum PQ metadata also represents 10,000 nits, combined with a 1,000-nit mastering-display value. He reports clipping at 1,000 nits on the Blu-ray players discussed, whereas Ugoos/CoreELEC, Shield and Apple TV preserve the dynamic information in this test. A second example uses 3,000-nit content with the 1,000-nit mastering-display value.

This is not a claim that Ugoos clips everything above 1,000 nits, nor that every Blu-ray player does. It is also not evidence that Shield or Apple TV reconstruct FEL: metadata handling and enhancement-layer reconstruction are different tests. The review predates later software playback implementations.

The useful lesson is to test the reference itself. Mastering-display metadata, scene content brightness and the receiving display's capability are different quantities. Highlight clipping loses distinctions between bright values; tone mapping may compress them into the display's range while preserving distinctions. A 10,000-nit test input does not require a TV to emit 10,000 nits.

Transcript scope¶

The section from 5:38 to 7:41 is consistent with the clipping account above and names no specific standalone Blu-ray player model or firmware. Exact affected models, settings and any firmware fix remain unresolved.

What the demonstration establishes¶

The reported clipping is evidence to investigate, not a direct measurement of its cause. The narration attributes it to a cap on dynamic brightness metadata at the mastering-display value. Without captured input and output metadata, the precise mechanism remains unverified here.

A comparison with the TV's internal app can help distinguish a simple panel brightness limit from processing differences. It does not isolate the fault to the external player: internal playback and HDMI input may use different processing and settings.

CPM's A13 build, CoreELEC-Amlogic-ng.arm-21.2-Omega_20250128030125.A13, is labelled a pre-release for wider testing. Its notes describe source-level metadata substitution, enabled by default and tested with TV-led output, plus separate Level 5 and Level 6 controls.²

This changes CoreELEC; it is not a firmware fix for the disc players in the video. The release notes establish implementation intent, not independent proof of correct output. The exact Ugoos build used in the demonstration and any later disc-player fixes remain unresolved here.

A useful retest needs exact player and TV models, firmware/builds, apps, active DV mode, hashed source files and before/after measurements. Stock Ugoos firmware, community CoreELEC builds and disc-player firmware have separate histories. The video's linked patterns have not been inspected or hashed here. Its separate HDR10/SDR capture claims must not be imported as proof of this Dolby Vision test's method.

Audio tests¶

First compare lossless decoding against known PCM. Then test routing and controlled object trajectories across speaker layouts, heights and fallback. Disable unrelated upmixing/enhancement. Align timing and gain for null tests. Finally test the 2.1 output chain with documented bass management.

Perceptual tests¶

Use randomised blind A/B or ABX with the same master, timing, picture mode, room light and seating distance. For sound, match level and speaker chain. Report participants and trials. “Not distinguishable in this test” is different from “identical”.

Current test status¶

Test	Status	Current conclusion
Local AetherEngine/CB1 execution	NOT PERFORMED	Static source review only
Local HDMI capture	NOT PERFORMED	No local accuracy ranking
DMDreview FEL/12-bit video	LINK VERIFIED; method audit pending	Relevant external lead, not a general PC verdict
HDR Dissector Avatar analysis	TITLE/AUTHOR VERIFIED; full audit pending	A topic-specific link, not an imported measurement result
HDR Dissector content-over-MDL review	TRANSCRIPT/DESCRIPTION REVIEWED	Historical reported clipping; models, firmware and raw capture audit pending
Local spatial-audio/null tests	NOT PERFORMED	Developer statements remain attributed claims

See S18–S20 and S30 for links and limits. Use templates/test-report.md. Keep large raw data outside Git and record hashes, sizes and access locations.

Reviewed 6 November 2026 · EDITORIAL · proposed protocol, not completed experiments

Evidence policy · Sources

S37, The HDR Dissector, video ID HyrA3KmcJBU. Title, author, publication date and description checked in YouTube; auto-generated English transcript reviewed at 5:37–7:44 on 6 November 2026. Player identities/builds, on-screen labels and raw results require further verification in this guide. ↩
S38, CPM releases, A13, source-level metadata controls. DEVELOPER / release notes; pre-release, tested with TV-led according to the author. Not independent hardware validation. ↩