extend.computer
Development buildv14 · two encoder slots
20.9ms
Median of run medians
35.5 msMedian of run p95s
5,728Readable transitions · 6 runs
THE DISPLAY LAB Snapshot · 2026-10-09
Less guesswork.
More measurement.
A good extra display should feel like part of your desk. Here’s what we’ve measured, how we measured it, and what still needs work.
Direct LAN · Wi-Fi2× Retina · 1728 × 1085Same two Macs3 repeats per scene
01 / RESPONSIVENESS
From one screen to the other.
Physical screen-transition delay. Lower is better.
Duet
Reference app4.0.491 · original signed app
56.4ms
Median of run medians
66.8 msMedian of run p95s
5,743Readable transitions · 6 runs
Typical delay & slower moments
Median p950255075100 ms
p95 marks the slower end: 95% of readable transitions in a run were at or below that delay. Bars use medians across runs, not pooled events.
Every run, without averaging away the variation
| Product | Scene / repeat | Median | p95 | Transitions | Visible updates ≥ |
|---|---|---|---|---|---|
| extend.computer | motion / 1 | 25.0 ms | 37.6 ms | 938 | 50.8 / s |
| extend.computer | scroll / 1 | 20.9 ms | 33.4 ms | 973 | 49.0 / s |
| extend.computer | scroll / 2 | 20.9 ms | 33.4 ms | 953 | 49.1 / s |
| extend.computer | motion / 2 | 25.1 ms | 37.6 ms | 922 | 48.7 / s |
| extend.computer | motion / 3 | 20.9 ms | 37.5 ms | 992 | 51.0 / s |
| extend.computer | scroll / 3 | 20.9 ms | 33.4 ms | 950 | 48.7 / s |
| Duet | motion / 1 | 58.4 ms | 70.9 ms | 1022 | 51.3 / s |
| Duet | scroll / 1 | 54.2 ms | 66.7 ms | 931 | 49.2 / s |
| Duet | scroll / 2 | 54.4 ms | 66.8 ms | 878 | 46.4 / s |
| Duet | motion / 2 | 58.4 ms | 70.9 ms | 942 | 48.6 / s |
| Duet | motion / 3 | 58.4 ms | 66.8 ms | 1012 | 52.1 / s |
| Duet | scroll / 3 | 54.2 ms | 62.6 ms | 958 | 50.1 / s |
CATEGORY INDEX / PROVISIONAL
Responsiveness, on a common scale.
Best per measurement = 100; blended indices can be lower for every product. Equal weight for typical delay, p95, and every matched scene/repeat. This covers screen-transition latency only.
extend.computer100/100
Duet46/100
Both apps requested 60 FPS. Reference settings were observed before and after capture; its bitrate remains unavailable. This index covers latency only, not an overall ranking.
02 / IMAGE & INTERACTION
Clarity is more than a frame rate.
Calibrated synthetic compositor captures. Three samples per region.
Text is close. Color is mixed.
Both preserve text structure well in this fixture. Colored edges vary by region. Near-perfect structural scores do not mean identical pixels or equal physical display quality.
Text structural similarity0.9995–1.0000Both products · SSIM, higher is better
| Test region | extend · SSIM | Duet · SSIM | extend · PSNR | Duet · PSNR |
|---|---|---|---|---|
| text 10pt | 0.99995 | 0.99993 | 47.2 dB | 48.3 dB |
| text 12pt | 0.99995 | 0.99982 | 47.3 dB | 47.9 dB |
| text 14pt | 0.99994 | 0.99986 | 47.3 dB | 48.0 dB |
| text 18pt | 0.99994 | 0.99988 | 47.1 dB | 48.0 dB |
| text 24pt | 0.99986 | 0.99969 | 46.2 dB | 47.1 dB |
| colored edges 1px | 0.97441 | 0.96745 | 7.2 dB | 7.3 dB |
| colored edges 2px | 0.99921 | 0.99895 | 17.7 dB | 12.1 dB |
| colored edges 3px | 0.99853 | 0.99938 | 11.7 dB | 13.9 dB |
| colored edges 5px | 0.99921 | 0.95191 | 13.9 dB | 14.4 dB |
| grayscale gradient | 0.99420 | 0.99386 | 51.1 dB | 50.2 dB |
| color gradient | 0.99734 | 0.99732 | 46.2 dB | 40.9 dB |
SSIM compares luminance structure. PSNR compares RGB pixel error; higher is better. Text scores cluster near 1; colored-edge errors vary substantially. These metrics alone do not establish a perceptual winner.
Duet
4/4 final visual regions passedbuttondragreplacementscroll
extend.computer
4/4 final visual regions passedbuttondragreplacementscroll
Earlier accepted checks verify final button, drag, text replacement and scroll outcomes. They have not been repeated for these latest builds, and do not measure input-to-photon or transition smoothness.
03 / THE WHOLE EXPERIENCE
No overall winner. Yet.
A fast scene isn’t enough to rate a whole product.
Cursor response
Matched 10-second cursor-transition and continuous-motion checks passed. Repeated trials remain.
Input-to-photon
Synthetic input is covered. Physical hardware input is not yet measured.
Recovery & reliability
Both products resumed after a 2-second sender pause. Network loss, sleep/wake and long sessions remain.
Energy & efficiency
Process telemetry exists; matched energy evidence incomplete.
Settings coverage
Resolution, frame-rate and quality presets need a complete matched matrix.
Window & display behavior
Fullscreen, scaling and interaction outcomes tested in part.
Missing measurements stay missing. They never become zero, a pass, or a reweighted score that hides incomplete coverage.
04 / REPRODUCIBILITY
The conditions matter.
Results belong to this setup and these builds.
- source
- Apple M4 Pro · Mac16,8 · 48 GB · macOS 26.7
- receiver
- Apple M3 Max · Mac15,9 · 48 GB · macOS 27.0
- power
- Both hosts on AC power
- display
- 1728 × 1085 logical · 3456 × 2170 backing · 2× Retina
- route
- Direct LAN over Wi-Fi · en0 on both hosts
- camera
- Pixel 10a · 1920 × 1080 · 240 fps
- fixture
- AppKit · identical workload hash · 20 seconds per scene
- repeats
- 3 motion + 3 scroll runs per product
- baseline
- 5-second connected idle + 5-second warm-up per campaign
One camera. One clock.
A generated counter appears on both screens. We compare its first readable appearance in the same high-speed video. That includes capture, encoding, transport, rendering, and both physical displays.
Camera sampling and rolling shutter introduce uncertainty. Unreadable transitions are excluded; coverage stays part of the evidence.
Scoring that earns its number.
The latency index is the geometric mean of best/observed ratios for median and p95 in each matched scene/repeat, multiplied by 100. Adding products can change this relative index.
Future overall scores require every category and comparison gate to pass. Versioned weights will be declared before measurement; no overall score is calculated for this snapshot.
Measurement limits & provenance
- First readable matching counter on receiver minus source, in one camera timeline. Includes both panels and scanout.
- Unreadable transitions excluded. Visible frame rate is a lower bound, not an encoder or presentation rate. Cadence excludes stationary preparation counter zero.
- Both requested 60 FPS at matched Retina geometry. Duet FPS, resolution, transport and encoder were observed before and after the run, not continuously. Duet bitrate unavailable.
- Digital quality is median of three synthetic compositor captures against independently calibrated references. It excludes physical panels and does not score detail lost solely through viewport scaling.
- Interaction outcomes shown are earlier accepted diagnostics, not revalidated for v14 or Duet 4.0.491.
- Cursor latency, energy, presets, long-session reliability and recovery still lack complete matched comparison. Overall score withheld.
motion-scroll-extend-cadence-v2 · report SHA-256
c541a20728399081e566f838d93386ccc08f0fbf8e49960dca2b34725985a3d0motion-scroll-duet-cadence-v2 · report SHA-256
1b05d2f6b56594c714da4f84587ce7c34de93b60b5542b541b9f4a29b47ffa3bdigital-quality-extend-current · report SHA-256
269754dce761a5d541abf4fa9be52876c78a6b11f1a80b35129d04c1dc00e0f2digital-quality-duet-current · report SHA-256
fb5c4d062c713c8f5298cead32c81caf5133200910d0b5ca84b3acdfc6a7f153Shared workload SHA-256
a80b05bb39e03f9c3d0b7d9892ca170375a5c086f6499698cd064ab27326edb2