# aj360-raw-edit: delivery report (2026-10-29)

Status: technical green, creative candidate (the owner approved the preview and picked music B; the final has not been seen by the owner yet)
File: `ice/renders/final/aj360-raw-edit_16x9_1080p.mp4`  sha256 b2c1413add0065bec28ef6337918362228acc32910e851c00c391c9d08782042  1920x1080  24 fps  13.50 s  h264 / aac 256k 48 kHz
Other deliverables: `ice/renders/final/aj360-raw-edit.ar.srt` (Arabic translation, timed to the speech), `aj360-raw-edit.en.srt` (English as spoken), `aj360-raw-edit.edl.json`, `aj360-raw-edit.transcript.json` (no speech was cut, so film time equals source time). No other ratios or cutdowns (D-09).

## The idea
The speaker's five sentences on leadership, each carried by the Al Jazeera 360 promo move that the same kind of sentence triggers in the reference, closed by the platform's end card.
Strongest frame: 0.60 to 1.20 s, the stacked words behind the head and shoulders with the red disk, as on the reference's opening.
Next version: a 9:16 reel of the same cut, or a version with the speaker named once his identity and consent are known.

## Beats
| # | Time | What it buys | Real or illustrative |
|---|---|---|---|
| B1 | 0.00 to 2.10 | Hook: "القيادة / ليست امتلاك / كل الإجابات" behind the speaker, red disk, rack focus | Real footage (raw.mp4), code-drawn type |
| B2 | 2.10 to 3.85 | Key clause in the glass capsule: "بل طرح الأسئلة الصحيحة" | Real footage, code-drawn capsule |
| B3 | 3.85 to 6.30 | "Others" and "solutions" shown as real teams at work, NASA 2026, tagged "المصدر: ناسا", caption "وتمكين الآخرين من إيجاد الحلول" | Real official footage (NASA), code-drawn caption and tag |
| B4 | 6.30 to 8.10 | Concept label in the glass panel: "هكذا نبني مستقبلًا أفضل", black and white with the speaker at full exposure; app capsule | Real footage, code-drawn panel |
| B5 | 8.10 to 10.00 | Two-part phrase beside the speaker: "معًا" / "نصنع الفارق" | Real footage, code-drawn type |
| B6 | 9.62 to 13.50 | Phone slab carries the film into the app; platform end card | Code-drawn; official logo, font and handles |

## Gates (copied from script output, measured on the file above)
| Gate | Result | Numbers | Tool and file |
|---|---|---|---|
| Cold-viewer clarity | PASS r1 (approver match below) | Viewer: Sells "Al Jazeera 360, from the corner logo and end card"; idea "Good leaders ask questions and let the team find answers, together"; turn read correctly | `ice/review/clarity/read-3-r1.md`, packet `packet-read-3/` (the 960x540 preview with music B, identical picture; the 1080p final adds only the source tag and the end card settle) |
| Brand test | Not passed: 0 of 6 masked frames named | Every masked frame "I can't tell". Not a blocking gate for this edit type (broll-overlay is not in the promotional list of `identity-tests.md` section 3); run as an extra check. Reason: on this kind of edit the platform's identity lives in its logo, name and end card, which the test covers; an ordinary viewer does not know the promo's graphic grammar | same report, `packet-read-3/masked/` |
| Competitor swap | not run | n/a for this edit type | none |
| Studio benchmark | not run | the ice-edit route has no studio benchmark (routes.md) | none |
| Film gate | YELLOW (deliverable) | fast in-shot motion 0.059 (yellow line 0.06); near-blank run 0.000 s GREEN; blue-purple share 0.148 (12.5 to 13.5 s); push-ins 8.9 per min reported | `film_gate.py`, `ice/review/gates/film_gate.txt` |
| Richness (no floor for this type) | reported | hero_share 100 %, frame_fill 16 of 16, empty_half 0 s, in_shot_motion 91.4 %, longest_still 1.08 s | same |
| Motion and freezes | 1 freeze (declared hold) | cuts 5, freezes 1 (12.04 to 13.46 s, end card), black 0, white 0, end hard black False | `motion_check.py`, `ice/review/gates/motion_check.txt` |
| Loudness | pass | I -14.1 LUFS, LRA 5.3 LU, peak -0.9 dBFS; target -14 LUFS, true peak -1.0 (two-pass loudnorm on the premaster) | `loudness.mjs`, `ice/review/gates/loudness.txt` |
| Spec | OK spec | 1920x1080 h264 24 fps 13.50 s | `probe-mp4.mjs`, `ice/review/gates/probe.txt` |
| G1 proof readable, G4 one hero | by eye | smallest text: app capsule 34 px and source tag 32 px at 1080 (readable at full screen, small on a phone, as the viewer noted); one hero per frame (the speaker, or the insert) | `ice/review/gates/final_contact.jpg` |
| Captions and sync | by eye and numbers | Arabic and English SRT timed from the word transcript, no overlaps; on-screen Arabic lands on its English words (timeline.js anchors) | `raw.words.json` |
| Template distance, fingerprint | not run | a remake of a supplied reference by request (D-11) | none |

### Clarity match (approver role, this session)
| Field | Viewer answer | Match |
|---|---|---|
| Sells | "Al Jazeera 360" | MATCH |
| Offer's job | (takeaway: none) | skipped |
| One-line idea | "Good leaders ask questions and let the team find answers, together" | MATCH |
| What happens | no misread of the turn or the hero moment | MATCH |
| Confusions | all minor; fixed: who the people are (source tag added); kept: the red develop flash at 4.5 s and the cut-off panel words at 6.5 to 7.5 s (both are the reference's own moves); the speaker is unnamed by choice (D-07) | no major |

### Yellows and holds
- Fast in-shot motion 0.059: a seated speaker in a locked studio shot; the reference's own A-roll is the same locked shot with slow digital pushes.
- Blue-purple 12.5 to 13.5 s: the platform's own navy end card, measured from the official reference (#262636 to #263F51).
- Push-ins 8.9 per min: one slow push per beat, alternating with a pull-out at the open and the end card settle.
- Declared hold: 12.04 to 13.46 s, the end card settles to rest; the reference's end card also settles (0.2 % of pixels changing in its last 3 s).

## Sound
Speech: raw.mp4 audio 0.00 to 10.00 s, uncut, same speed, last word (9.60 s) never faded. Music B: "Deep Bass Cinematic Tension Forensic Noir Investigative Pulse", Jose Luligo, Pixabay, window from 45.32 s, about 14 LU under the speech, rising after the last word. Effects: one real Pixabay recording per move (`ice/audio/SOURCES.md`). Measured, not heard by the agent; the owner listened to the preview mix.

## Run notes (auto mode: what was decided instead of asked)
- Start questions waived by the user ("ما تسألنيش أسئلة البداية وقرر أنت"); each default with its alternatives is in `ice/DIRECTION.md` D-01 to D-09.
- Owner notes applied (D-16): real library audio only, the reasons table (`ice/plan/element-reasons.md`), the speaker kept lit, finishing measured from the reference.
- Removed for lack of a trigger in the speech: neon title, crosshair, flag, "يعرض الآن" badge, programme bug.
- Downloads: official Al Jazeera 360 logo and fonts from aljazeera360.com; NASA footage from images.nasa.gov; Pixabay audio. Installs: a project `.venv` (analysis and audio packages), with the user's consent.
- Turbulent Displace is approximated by a slow wobble of the red disk; halftone, specks and flicker are on the inserts only.
- The approver and critic roles ran in this session (sequential pass, not independent), except the cold viewer, which was a separate fresh agent.

## Permissions a brand owner or rights holder would need before publication
| Item | Owner | Needed |
|---|---|---|
| Al Jazeera 360 name, logo, fonts, end card look | Al Jazeera Media Network | permission to publish under their identity (this is an unofficial practice edit) |
| The speaker in raw.mp4 | the person filmed | consent to this use, and his name if it is shown |
| NASA footage | NASA (public domain in the US) | credit "NASA" (on screen as "المصدر: ناسا"); no endorsement implied |
| Music and effects | Pixabay authors | none (Pixabay Content License; credit optional) |

## What I would fix next
1. The glass-panel words at 6.5 to 7.5 s are hard to read while they slide; a slower scroll or a second still beat would help, at some cost to fidelity.
2. The app capsule and source tag are small on a phone.
3. The person matte is per frame from u2net_human_seg; a video matting model would make the hair edge steadier behind the stacked words.
Strongest: the opening, which matches the reference's grammar on the raw's own words. Weakest: the end card, drawn in code from the reference rather than the platform's own animated card.