UPDATED 2026-08-02 22:2x — frames 3, 4 and 7 reworked and re-embedded. 1, 2, 5 and 6 are untouched since 21:35. All seven ship. 3 and 4 are now a profile and a rear three-quarter rather than the same shot twice — they still measure identically, but they no longer look identical, which is the reading that matters.
PROOF — Gulf
Seven frames, all seven in the app. Generated with Adobe Firefly, 2026-08-02.
Sources in IMAGES/AIgen/ at 600×3086; what ships is downscaled to 560×2880 — the same ratio, so no crop at all.
Look at 3 and 4 together. They measure identically — same hue, same chroma, same derived color to the digit — because they are near-duplicate compositions: same gradient, same car, same reflective floor, marginally different framing. In a seven-frame rotation two adjacent near-identical frames read as the slideshow having stalled. That is a look-at-it call, not a measure-it call, which is why I built both rather than dropping one.
In rotation
Click a frame below to add or remove it from the rotation.
The sidebar crops, on purpose. The art is
600×3086 and ships at 560×2880 — a 1:5.14 strip. In a 248px sidebar it renders about 1275px tall, so a normal window shows roughly 70% of it, centered. Every frame is composed for that: the car sits between 52% and 75% of the height, inside the visible band on any screen taller than about 700px. The shipped app crops identically — a mockup that showed the whole strip would be showing you something no user ever sees.Measured
| Frame | Hue | L | Chroma | Derived | Accent contrast | Blown | |
|---|---|---|---|---|---|---|---|
| Gulf1 | 37.0° | 0.57 | 0.156 | #8C5545 | 3.20:1 | 0.65% | |
| Gulf2 | 240.3° | 0.57 | 0.178 | #097DB8 | 4.24:1 | 0.00% | orange car, blue sky |
| Gulf3 | 52.2° | 0.57 | 0.144 | #F18F4E | 8.02:1 | 0.00% | reworked — profile |
| Gulf4 | 52.2° | 0.57 | 0.143 | #F18F4E | 8.02:1 | 0.16% | reworked — rear three-quarter |
| Gulf5 | 239.8° | 0.52 | 0.142 | #1275AA | 3.79:1 | 0.00% | darkest frame |
| Gulf6 | 39.8° | 0.57 | 0.189 | #BC4E24 | 3.89:1 | 4.06% | highest chroma; 4% blown |
| Gulf7 | 240.7° | 0.57 | 0.143 | #077DB9 | 4.24:1 | 2.11% | reworked — livery on asphalt |
What the numbers do and do not say
“Accent contrast” is the derived color’s contrast, not the nav text band. It answers “could this frame carry an accent”. It does not answer “is the sidebar text readable” — that is governed by the scrim and the theme tokens, both unchanged in this rebuild, so text legibility is exactly what it was before.
The derivation is set-relative, so single-frame numbers lie. Measured alone, Gulf1 reads 37.0° and derives
#8C5545; measured in the set of seven it reads 40.0° and derives #A25134. Only the batch figures above mean anything, and they change again if a frame is added or dropped.Lightness range 0.06 across the set — a cycle travels 0.12. Clouds at 0.20 was smooth and Landscape at 0.41 visibly brightens, so this is the steadiest set in the app. Gulf5 is the one outlier at L 0.51.
Two pairs now derive near-identical colors — and it does not matter. 3 and 4 both give
#F18F4E; 2 and 7 both give roughly #087DB8. Gulf has no auto-tone. Its accents are fixed tokens — #6CACE4 in, #FF8200 out — so the per-frame derived color is a design-time reading only and drives nothing at runtime. The collisions are the same blue sky measured twice, not two frames that look alike.Still the only look-at-it call: do 3 and 4 sit too close in the rotation? They are now visibly different angles of the same car on the same floor, which is a normal thing for a set to contain. Run the strip above and see whether the crossfade between them reads as a change or as a stall.