P1 Discovery — v4

Acts 2 and 3 are now the actual 3D map, driven as a camera rig. You were right that what we built looks better than what I was substituting for it. Eight cinematic takes off the live register: an orbit at 67° pitch, a dive from regional altitude into the towers, a low pass at 80° for parallax, the real filter panel narrowing the field, the real colour-by-delivery-record mode, and a pull-back for the close.

The reveal is generated from the data, not faked. I rebuilt the map's own height expression with a growth factor and animate it 0 → 1, so every registered tower in Gurgaon physically rises out of flat ground on the reveal line.

Shots now sit inside beats, not one per beat. v3 held one look per narration line — 14 shots over 88 s, a 6.3 s average. There are now 27, and outside the three uncut demonstrations the average is 3.2 s. Shots are weighted by each beat's tension, so the cut rate accelerates through the agitation and slows at the reveal. The pivot is a single long hold; the reveal, the narrowing and the shortlist are never cut into, because their counters need to read.

Three new frames. Cutting faster with the old set would just have repeated the same pictures, so act 1 gained a broker sliding one brochure forward while five sit ignored, a couple sitting tired in traffic under half-built towers, and a phone at midnight.

The performance had pauses and dynamics. I was pouring a wall-to-wall music bed into every pause and then limiting the peaks off. Music now plays in blocks with real rests — including full silence under "Until now" — and the limiter only catches peaks.

I also tested recasting, and it would have made things worse. I generated the same passage on nine voices and measured pitch movement in semitones. Lovekesh scores 6.8; every alternative scored 3.8–5.2. He is the most dynamic voice available to us. Judge it yourself — but the measurement says the casting was never the problem.

Direction was partly to blame, and that is fixed. [short pause] is not a real v3 tag — it was being ignored. Rewriting the pauses as paragraph breaks took the take from 26% to 36% silence and 6.3 to 7.2 semitones, with no settings change.

Rise → silence → hit → hold. A 2.6 s riser leads in, the music cuts to nothing, a sub-heavy impact lands on the word, the city stands up, and the bed swells back above its earlier level. Both stingers are synthesised for this cut.