Horizon or Daybreak

Down to two. A is Cara's HORIZON, reproduced from her file frame by frame — her timings, her roll, her retraction. B is the same reveal on a rule that crosses the frame. Same lockup, same 7.007s, same silence. All 336 frames pass eight arithmetic checks.

Candidates

Horizon

Cara's, reproduced

The rule draws in dim and distant, firms to full as it completes, and the wordmark rises out from behind it with a roll that unwinds as it clears. Then it sets: the word sinks back and the rule retracts the way it came.

0.06srule draws, dim

2.16sfirms to full

4.04sword cleared, level

6.14ssets, rule retracts

Measured off her file frame by frame: the rule at 66% alpha while drawing, the roll 7.0 deg unwinding to 0, and an exit that is the entrance run backwards — which is why it needs no fade. Her proportions, over 7.007s.

Daybreak

riff on Horizon

Her vertical reveal on the horizontal spine. The rule crosses the frame edge to edge and the wordmark rises out from behind it as it passes.

0.25srule enters at frame edge

1.00sword begins to rise

3.64scleared and level

5.90sfade

The two ideas do one job instead of two. Roll reduced to 4.2 deg because the horizontal travel is already carrying the eye. Ends on a fade rather than a retraction, so it butt-cuts straight onto frame 1 of the feature.

What these are, exactly

Raster	1920 x 1080 (review)	3840 x 2160 for delivery — matches the master exactly
Frame rate	23.976 fps	24000/1001, the master's own rate. 168 frames = 7.007s
Duration	7.007 s	Inside the 5–10s the distributor-card clause allows
Ground	true black, 0	NOT Cara's #0A0A0A — see the note below
Audio	none	Silent by design — see the note on the splice
First / last frame	uniform black	Verified on all five. Butt-cuts at both ends

Before this gets cut in

Silent still needs a silent track

No music, as asked. But at the splice the ident needs 7.007s of actual silence in the master's channel layout, not a missing audio stream — drop the stream and every subsequent sound cue slips forward by the length of the ident. Silence is a thing you author, not a thing you omit.

The master is HDR; these are not, yet

The 66 GB master is PQ / BT.2020 10-bit. Splicing an sRGB card into it without conversion will make the bone read wrong. The delivery render converts to smpte2084 / bt2020 with the bone mapped near 100 nits — not the 1000-nit mastering peak. A bone-white card at peak in a dark cinema is painful to look at.

Reel 1, not composited into the picture

Whichever wins, it should reach the DCP house as its own asset to be built as reel 1 of the CPL — not burned into the feature's picture essence. Same result on screen, and it keeps the film untouched.

v1 was square, and that was the only real problem with it

All ten of Cara's originals are 3840 × 3840, because the mark is drawn on a 1080 × 1080 canvas. Nothing was wrong with the craft — the frame is simply the wrong shape for a cinema screen. They also ran at 24.000 fps against a 23.976 master. v3 keeps her letterforms and her lockups and rebuilds the frame around them.

The clause still has to land

§9 of the draft agreement currently says the marker “is not embedded in any master” and leaves [before] bracketed. Cara wrote the correct language on 29 July; it never made it into the draft. That, plus a written OK from Sam on the logo card specifically, wants to be settled before the house is paid.

Why the ground is black and not Cara's ink

Her mark files use #0A0A0A, which is right for web and print: on a white page a dead black looks like a hole. A cinema is the opposite case. The auditorium is dark, and a full-frame #0A0A0A card reads as a lifted black — the screen sits there visibly glowing grey for seven seconds and then drops when the feature's true black arrives. The ident has to live in the same black as the film's leader, so these are built on 0. It is the one place the mark files were not followed, and it is deliberate. The clips above are true black; this page's background is her ink, so you can see the difference at the edge of each picture.

Three defects caught this round

None of these were visible to me on playback. All three came out of the verifier.

The horizon mask was doing nothing at all

An SVG trap: clip-path and transform on the same element. The element's transform establishes the coordinate system its own clip resolves in, so the clip travels with the element and can never clip it. The wordmark was sitting 90px below the rule on frame 1 and being drawn anyway. Echo had the same bug quietly — its offset copies dragged their wipe along with them. Fixed by putting the clip on an outer group with no transform.

The roll lifted the word over the horizon early

Hiding the wordmark behind the rule is not just a matter of the lockup's height: a 7° roll about the left end lifts the far end by width x sin(angle), which at 540px is 66px — enough to make it peek over before it exists. The hide distance now accounts for the roll.

Horizon does not settle when the others do

Keeping Cara's proportions puts its hold at 0.576–0.876 of the clip against 0.520–0.842 for the rest. Tested against a shared window it reported a false failure, so each variant is now verified on its own timing rather than a constant.

Verification

336 frames, eight checks each: frame count, uniform black at both ends, byte-identical hold, clear right margin on every frame, settled wordmark inside title safe, no single-frame outlier, no snap into the hold, and a fade that never brightens. Earlier passes caught two defects this way that the eye missed entirely — a full-frame one-code-value flicker through every fade, caused by compositing an ink rectangle over an ink ground, and sub-ground pixels on the glyph diagonals. Both are gone, and the ground is now true black rather than Cara's #0A0A0A, which on a cinema screen reads as a lifted black.