La Tarde — on the fifteen

A is recommended. It puts a real photograph under COMMUNITY ALTAR and ARTS & CRAFTS — the carousel's own — and gives PERFORMANCES the one thing the footage pool actually holds, dance. B keeps the delivered edit intact and costs nothing new, but its second title lands on a skirt fan: COMMUNITY ALTAR over a spinning skirt says the wrong thing at the moment it says it, and no timing inside that cut fixes it, because there is no altar footage to cut to.

• The gate — go on both formats (round 5) — the marketing-director pass read every A frame at full size in 9:16 and 1:1 against the brief and the approved carousel cards and returned go on both formats, minors only. Its picture finding is closed below; its page-copy note is the mid-sweep caption now carried under frame 107 in both formats.
• Minor, both formats, the 184–215 hold — the EXIT sign is gone. A small red illuminated EXIT sign sits on the left wall for the whole hold (9:16 x 10-70, y 764-794 at frame 184; it drifts with the camera to x 95-140, y 714-744 by 214, and the 1:1 window shows the same wall 361 px higher). Of the two fixes offered, the deeper left-edge vignette was measured and rejected: at −0.8 stop the sign region still reads 141/42/55 against the wall's 114/61/81 (R−G 99.3 vs 53.2) — a dimmer red rectangle, not a wall — and the vignette is a full-height left-edge ramp on 9:16 only, so it would have to be pushed up over the dancer and duplicated for 1:1. Taken: the clean plate. The wall is NOT locked off - the sign's own box, measured on all 31 frames of the shot from the raw clip, drifts from (7,509,46,528) to (76,483,104,501) in source pixels - so a static 50x30 patch could not have held. The op carries a quadratic track (fitted to +/-1.5 px) and a 61x37 box that follows it; the fill is a per-row interpolation between the clean columns just outside the box, feathered 5 px on all four sides, so the feather ring lands on clean wall. Horizontal only: the wall carries a steep vertical gradient from the sign's own glow, which each row's own anchors reproduce, and there are only about 2 px of clean wall under the box before the seated crowd. It is applied ONCE at source resolution, before the grade, so 9:16 and 1:1 get the same retouch and both the washed and the unwashed frame carry it.

Measured against the round-4 render frame by frame: on 9:16 the only pixels that differ over the whole 31-frame hold lie inside (0, 695, 176, 809), on 1:1 inside (0, 336, 176, 448) - every other pixel of every frame is bit-identical, and the change box is clear of the kicker, the sub, both waves and every measured face box. What is left where the sign was is the wall's own glow: the region reads 9-15 points brighter than the wall around it with the same hue (R-G 66-71 against the wall's 61-65), where before it was 63-65 points brighter at R-G 145-154.

• Major, A_S_0184 (both formats re-cut; lever b) — the QC was right and the round-3 "48 px clear" was a mis-measure: on the rendered square frame the cuff's top sat at y 28 against the band's edge and the whole chalice was hidden under the band (the cup's rim is at P y ≈ 246 at 3.0 s, not the 475 written down). Lever (a) — the square focus for 1651@3.0, 0.45 → 0.36 in 0.02 steps, measured on the square frames 184 / 194 / 204 (cup and cuff clearance to the band's last inked row in their own columns, in delivered pixels, negative = under the band — the cup's rim is at P 246 / 234 / 299 on those frames, the cuff's top at P 406; the bystander's face bottom against the kicker's ink at 775, cards' 10 px margin):

No step clears: the cup needs the window at y0 ≤ ~150 (focus 0.18) while the face rule already fails from 0.43 down. Lever (b) — the in-point. The arm holds its peak from 3.1 to 3.6 s (cup top P 235–245) and comes down only from 3.7 (320), 3.8 (395), 3.867 (476), 3.9 (~530), 3.933 (556), 3.967 (609) — about 50 px a frame — so the suggested 3.1–3.6 cannot clear either. Candidates measured through render_cut's own chain (cup rim vs the last inked row of the teal wedge; face vs the kicker):

Taken: IMG_1651 from 3.9 s in both formats, square focus 0.43 (same span, same edges, kicker still on frame 92; the chalice is still up beside the head at the in-point, then the sweep). On the delivered square frames:

At 194 the sash runs diagonally (700-745) with the kicker 30 px under it; at 204 the sweep has turned her, her back is at x 600+ and the sash sits at 770-805, so the kicker's top edge lands on the sash, as it did at 204 in every previous round (negative = overlap); 184 is the proof frame. The 204 hairband entry is the headdress disc (she has turned away). Cup rows from the dark-blob detector on the delivered frame (x 800-1000, rim vs the band's last inked row in the same column); cuff, hairband, belt and face boxes read off the delivered frames at 2x with a 20 px grid. No face sits in rows 700-930 on any of 184 / 194 / 204 / 214 (checked without the overlay). The 9:16 frame 184 now shows the same moment (the chalice at the head's level, coming down) — the P framing of that shot was optional polish only, and the in-point move is 0.9 s, larger than "small": said plainly here. Both ends of the moved shot pulled at full size from the raw clip and from both renders: no signage; the defocused red exit-light glow at the far left as in every round. The face box is now measured in delivered square coordinates and asserted in make_cutplan_c (FACES).

• Minor, A_S_0107 (held at 0.10) — the kicker's beat is fixed at 92, so the only lever is the window. Measured both ways on frames 107 / 137 with the type composited: the literal move, window up 40 px (0.10 -> 0.052), moves the CONTENT down - the crown (the headdress disc at y ~600, the hair to ~760 at 0.10) drops to ~640 / ~800, under the kicker's ink at 775 by ~25 px, and the disc-to-kicker distance shrinks from 175 to 135 px. The opposite move (0.148, content up 40) takes the hair 55 px clear of the kicker but the rising chalice at the out-point (frame 137) ends 20 px from the teal wedge's last inked row (x 907), against 62 px at 0.10 - the very failure round 4 is fixing at 184, and under the 40 px rule. Neither direction is allowed and a later hold is not, so 1651@2.0 stays at 0.10 with the kicker touching the nape below the crown, 15 px under the hair; said so and left.
• Minor, A_P_0107 / A_P_0184 band sample — the sample point sits INSIDE the gold band's own soft edge, not on footage: the band plate is fully opaque there (rgba [230, 167, 67, 255]) and its gold shades from 252/183/73 at y 1740 to 216/159/64 at y 1776 before the teal at 1780, so nothing under the bands - scrim, boost or vignette (whose x-feather is 120 px; x 250 is outside it, and it is composited under the waves anyway) - can reach it. The 224/171/69 is the staged board JPEG (4:2:0 chroma at that gradient edge, teal bleeding up); the master decoded accurately reads 229/168/67 at frame 184 against the card PNG's 230/167/67, and footage frames vs card frames differ by <= 3. At the interior point (250,1750) every frame, footage or card, reads 253/183/74 against the PNG's 252/183/73. No mask needed; the table below is on every block frame, both formats.

9:16 — six band points on every block frame and the cards, from the delivered file (accurate decode), against the carousel cards' 43/175/179 · 252/183/73 · 51/56/74:

1:1

• Re-checked: zone scan on 9:16 (1533 (limit 1536)); faces at 184 / 194 / 204 / 214 on the square, asserted against the type's own ink; both ends of the moved shot from raw and render, both formats; the six band points on every block frame in both formats; 450 / 450 frames on the two masters; every A frame looked at full size. Round 5 adds: the mark's box measured on all 31 frames of the shot and asserted inside the patch's full-strength core (make_cutplan_c.EXIT_SIGN), and a whole-frame difference against the round-4 render confirming nothing outside that box moved.

• Layer order (major, both formats) — the footage block's scrim is composited under the wave plates (scrim → bands → type), as the cards are built. The six band points sampled on every frame of the block and on the cards, from the delivered files (decoded with the accurate path), against the carousel cards' 43/175/179 · 252/183/73 · 51/56/74:

9:16

1:1

• Colour tags (major) — every encode now converts its RGB sources (cards, overlay plates, graded footage) with the BT.709 matrix at limited range and tags the stream (tv,bt709,bt709,bt709,450; before: tv,bt709,unknown,unknown,450, with the card segments converted through swscale's BT.601 default). Gold type, frame 31 vs CA00P-A-1: card 255/220/2, now 253/217/31, before 255/208/20; frame 245 vs CA00P-A-3: card 252/216/2, now 251/217/1, before 255/207/0. The lossless concat is unchanged; the other codes pick the fix up at their next rebuild (render_wide.py still needs it).
• The 1:1 opener (major) — IMG_1651's in-point 1.5 → 2.0 s, framed at 0.10 (above). Photo-zone luminance at the in-point 93 (was 50) against the carousel card's 103; no lift needed. 9:16 re-checked at 92 / 107 / 137.
• 1:1 crops — the chalice raise (1651@3.0) 0.50 → 0.45: the window moves up 42 px and the blue cuffs clear the top-right teal by 48 px at frame 184. The 100–120 px asked for would put a seated bystander's face (blurred, behind the feathers, P 230–307 × 1060–1133) under the E of PERFORMANCES from 0.40 down, so the face rule caps the shift; the fingers still dip under the band at the hand's peak (~frames 190–205), as they did at 0.50. No horizontal control on the square window (full source width), so the headdress disc stays where it is.
• 9:16 polish — scrim +10% from y 1360 (60 px ramp): the ground under the headline block now reads f92 35.7 (was 41.4), f107 26.9 (was 34.0), f138 40.5 (was 51.7), f153 44.5 (was 56.9), f169 31.2 (was 39.0), f184 26.4 (was 32.6), f215 38.3 (was 48.7), f230 42.2 (was 53.8); CA00P-A-2 reads 30.5. Edge vignette on the block only: 120 px feather, −0.4 stop, from y 900, hard in/out on the block's cuts (77 / 245). Cards, grade (locked to the approved fit), bands and type untouched.
• Re-checked: zone scan on 9:16; both ends of the moved shot from raw and render, full frame, both formats; faces at kicker-on; 450 frames on both masters.

• Blocker — frame 107's butterfly replaced by IMG_1651@1.5 (feather sweep): the 50% tag, Big & Tall and the poster are gone, and the close's dancer no longer appears twice. Same spans, same edges, kicker still on frame 92.
• Major — the footage wash re-targeted to the cards' wall (above).
• Minor — the block's scrim lifted 80 px; the cards keep theirs.
• Re-checked: zone scan on 9:16, faces at kicker-on for all seven shots on both formats, both ends of the new shot from raw and from the render, 450 frames on both masters.