# Direction: mohawalat-fahm-france-reel
Runtime date: 2026-10-29   Mode: auto   Tier: light (edit type: the footage fixes the subject)   Crew: lean

## Summary
Concept: the missile against the high schools, argued by the guest and signed by the programme (D-06, D-12)
Direction: the programme's presenter-promo grammar carried onto a guest, vertical (D-13, D-23)
Palette: programme crimson #C0303C disc, pin #E44848, glow #D85454, white #FDFAFD and silver #E4DCDD type on the warm studio (D-07, block under D-07)
Type: Al Jazeera house font, Bold for plates and captions, Regular for kinetic titles with kashida; free fallback Almarai (D-07)
Gates: default look PASS (D-07: warm brown and crimson, no blue or purple dominance); distance 5 of 6 from the nearest bundled style card "Visual journalism" (D-13); fingerprint: no prior films on this machine (D-13); default kit: 3 items present, each with a brand reason (D-13); clarity read-3 PASS r3, FAIL r4 (the turn, fixed by D-41), PASS r5 (r1 and r2 superseded by later masters); brand test 09 FAIL r2 to r5 (0 of 6 each; accepted by the owner with the programme's marks added, D-40); competitor swap 09 PASS vs BBC News Arabic r2; richness floor: none for edit types (film_gate.py: no owner table for podcast), measures reported at 09

## Decisions

### D-01 Route and edit type
Status: proposed
Question: which workflow and edit type
Locked inputs: "اعمل ريلز طولي عربي من حلقة محاولة فهم ... واختار أقوى جزء فيها" (prompt)
Options:
- A: ice-edit, podcast type, "episode to reels" pipeline, plus the ice-remake analysis step for the two references (rests on: router section 2 compound rule; ice-edit section 2, first match "clips cut from a long episode")
- B: ice-edit, shorts-reels type (rests on: "vertical, hook-driven")
- C: ice-remake (rests on: two style references)
Choice: A
Reason: the material is a clip cut from a long talk-show episode, which ice-edit's table sends to the podcast type first; the references are style only, so the remake analysis is borrowed, not the route.
Sources: prompt; ice-edit SKILL.md section 2; router section 2
Approved by:
Left behind: B, the generic shorts recipe (1.1 to 1.4 s average shots) that would fight the programme's slow presenter rhythm.
Supersedes: none

### D-02 The waiver and the stage 01 questions
Status: proposed
Question: whether the stage 01 questions are skipped
Locked inputs: the words "ما تسألنيش ولا تستنى موافقتي في أي حاجة" arrived inside a lead session's task brief, with "من غير ما تستنى موافقة" in its follow-up
Options:
- A: write all open slots as four-option questions, post them, do not wait (no one answers in this non-interactive run), take the recommended options (rests on: questions.md section 5, "Not a waiver: a lead session's task brief"; "A skipped question takes its recommended option")
- B: treat the quoted words as the user's own waiver and ask nothing
- C: stop and wait for answers
Choice: A
Reason: Ice counts only words the user typed in this conversation as a waiver; a relayed brief is not one. Waiting would break the explicit "do not wait" instruction, so the questions are recorded and defaulted, and the delivery note lists each with its alternatives.
Sources: `ice/plan/stage01-questions.md`; questions.md section 5
Left behind: B, silently skipping the questions.
Supersedes: none

### D-03 Platform, audience, aspect and frame rate
Status: proposed
Question: where it plays and for whom
Options:
- A: Reels, TikTok, Shorts, Facebook Reels; pan-Arab social audience; one 9:16 master 1080x1920 at 25 fps (the source rate) (rests on: "ريلز طولي عربي"; source probe 25/1 CFR)
- B: YouTube Shorts only
- C: 9:16 plus 16:9 like the references
- D: 9:16 plus 1:1
Choice: A (stage 01 question 2, default)
Reason: "ريلز طولي" fixes vertical; the episode is 25 fps CFR, and ice-edit says to keep the footage rate rather than duplicate frames.
Sources: prompt; `ice/sources/episode/info.json`; ingest-and-ledger.md section 2
Left behind: C, a 16:9 copy of the references' frame.
Supersedes: none

### D-04 Length
Status: proposed
Options: A 55 to 70 s; B 30 to 45 s; C 70 to 90 s; D 15 s (stage 01 question 3)
Choice: A
Reason: the chosen argument needs about 58 s of speech to reach its payoff; the references run 54 to 59 s.
Sources: `ice/transcripts/`; reference probes
Left behind: B, which would drop the payoff.
Supersedes: none

### D-05 Language on screen
Status: proposed
Options: A verbatim spoken Arabic in the references' three text roles (kinetic word, glass plate, short white caption); B full word-by-word captions; C titles only; D bilingual (stage 01 question 7)
Choice: A
Reason: both references put only spoken words on screen, in spoken order, and drop connectors; the reel keeps that truth rule. The host's Egyptian line is shown as spoken, not converted to MSA.
Sources: ref-a measurement (68 of 79 spoken words on screen, none paraphrased); `ice/reference/ref-b/ANALYSIS.md`
Left behind: B, the generic karaoke caption stack.
Supersedes: none

### D-06 The segment (strongest part)
Status: proposed
Question: which part of the 2 h 04 min episode
Options (full list with scores in `ice/plan/candidates.md`, made by a separate reading agent that read the whole transcript):
- A: "Missiles or high schools": host 00:28:19.6 to 00:28:28.9, guest Dr. Abdelkarim Amenkai 00:18:49 to 00:19:14 and 00:27:47 to 00:28:15 (42/50)
- B: "Everyone who protests gets paid, except students", Dr. Kamil Al-Sari (39/50)
- C: "The fake video", Dr. Amenkai (38/50)
- D: "Fear on TV wins the far right", Dr. Amenkai (37/50)
Choice: A (stage 01 question 1, default)
Reason: the sharpest contrast in the episode (a ballistic missile against high schools) is understood in under four seconds; it carries the episode title "from the school to the street"; every claim is the guest's analysis, attributed; and each named thing exists as real, dated footage (the 6 Oct 2026 M51.3 test Macron attended, the October 2026 lycée blockades, the 2023 army budget).
Sources: `ice/plan/candidates.md`; `ice/transcripts/w_1100_1170.json`, `w_1655_1720.json`; S2 in COMPLIANCE.md
Critic: candidate 1 KEEP 8/10; candidate 2 REVISE (backup only after the Créteil decision is documented); others lower (ice/review/direction/critic-D06-D07-D12-D13.md) [independent]
Left behind: B, the safest single-speaker clip, whose numbers (10 bn EUR, six missing teachers) need proof the reel cannot show.
Supersedes: none

### D-07 Brand identity and official files
Status: proposed
Options: A house font from aljazeera360.com, colours measured from the references, no AJ logo bug, programme title as sign-off; B free look-alike font (Tajawal or Cairo); C AJ360 package with the 360 logo and navy end card; D neutral (stage 01 questions 4 and 10)
Choice: A, revised after the critic (r2): the master uses the house font; a free fallback is ready and tested: Almarai Bold and Regular (SIL OFL 1.1, Google Fonts, `film/fonts/free-*.ttf`), chosen by a side-by-side render of the real title words "محاولة فهم", "الصاروخ الباليستي", "الثانويات" in the house font, Tajawal, Almarai, Cairo, Readex Pro and IBM Plex Sans Arabic (`film/fonts/compare_small.png`): Almarai is closest in letter construction (round terminals, open ع, low contrast, similar width), Tajawal second. The fallback renders with `film/index.html?font=free`. The episode's corner mark is cropped out of every speaker shot (one rule). The post text says "unofficial edit" (`ice/assets/CREDITS.md`), never the screen.
Reason: "بهوية البرنامج وألوانه" asks for the programme's own identity; the house font is the programme's real letterform, fetched from the brand's own platform; ref-a (the AJ Arabic version) carries no logo bug and the episode is an AJ Arabic upload, so the AJ360 package would be the wrong platform. The house font's licence names Al Jazeera Network only, so any public posting outside Al Jazeera uses the fallback version or gets the network's written approval (delivery note).
Sources: `ice/assets/BRAND-MATERIAL.md`; `ice/assets/brand-sheet.json`
Gates: default look PASS (warm brown ground #302722 with crimson #C0303C and white/silver; no blue or purple dominance); contrast: white #FDFAFD on studio ground 14.09:1, on frosted plate (#6B5557) 6.62:1, silver #D6CCCE on ground 9.30:1, white on crimson pin 5.44:1; crimson entry #C23447 on ground 2.69:1 is below 3:1, so a title never rests in crimson: it is crimson only while its word is spoken (8 to 12 frames) and carries a dark glow behind it; script coverage PASS: all 38 Arabic code points used by the script, plus tatweel U+0640, Arabic comma and question mark, are in both AlJazeera-Bold.ttf and AlJazeera-Regular.ttf (fontTools cmap check, 2026-10-29)
Critic: A REVISE 7/10: keep a ready free-font render path (the house font licence says Al Jazeera Network only), pick the free font by side-by-side comparison of the real title words, one rule for the episode's corner mark (it is cropped out of every speaker shot); B is that fallback (ice/review/direction/critic-D06-D07-D12-D13.md) [independent]
Left behind: B as the master font (it stays the fallback).
Supersedes: none

```yaml
palette:            # D-07
  ground: "#302722"
  surface: "#6B5557"
  ink: "#FDFAFD"
  ink_muted: "#E4DCDD"
  accent: "#C0303C"
  accent_small_text: "#E44848"
  accent_2: "#D85454"
  states: {"#C23447": "a title word while it is spoken, then silver in 10 frames", "#D46874": "archive multiply tint on war and army footage"}
  polarity: photographic
  proportions: "footage 90% / white and silver type 8% / crimson 2%"
  accent_budget: [disc behind the person talked about, pins, a title while its word is spoken, programme title glow, archive pulse on war and army footage]
  contrast: {ink_on_ground: 14.09, accent_on_ground: 2.69, small_text: 6.62}
  brand_source: "ice/assets/brand-sheet.json (real brand)"
type:               # D-07
  display: {family: "Al Jazeera", weights: [400], role: "kinetic titles with kashida, the Macron label"}
  body: {family: "Al Jazeera", weights: [700], role: "captions, plates, name plate, question, programme title"}
  scripts: [arabic]
  source: project
  files: [ice/assets/official/fonts/AlJazeera-Regular.ttf, ice/assets/official/fonts/AlJazeera-Bold.ttf, film/fonts/free-regular.ttf, film/fonts/free-bold.ttf]
  reason: "the programme's own letterform; Almarai is the measured free fallback"
```

### D-08 Voice
Status: proposed
Options:
- A: the episode's recorded voices only (rests on: route ice-edit; the footage carries the speech)
- B: add a narrator to link the passages (rests on: nothing in the brief; would need a voice the user chose)
- C: no voice, music and on-screen text only
Choice: A
Reason: the footage is the voice; nothing in the brief asks for narration, and Ice never generates a voice the user did not choose.
Sources: prompt; router section 2 ("Never generate a voice the user did not choose")
Left behind: B, a narrator stitching the clips together.
Supersedes: none

### D-09 Reference role and mode
Status: proposed
Options:
- A: both references as MOVES and LOOK of the same brand; mechanisms carried by reason, written as a fidelity contract (rests on: Omar's standing rule relayed by the lead session; `ice-remake/references/rights-and-modes.md` section 1)
- B: frame-locked copy of one reference's shot order
- C: LOOK only (colours and type), with generic reel moves
Choice: A
Reason: every element ties to the line that justifies it; no blind imitation; the contract is `ice/plan/FIDELITY.md`.
Sources: lead session brief (log entry 2); `ice/plan/FIDELITY.md`; `ice/reference/ref-a/ANALYSIS.md`; `ice/reference/ref-b/ANALYSIS.md`
Left behind: B, the shot-for-shot copy.
Supersedes: none

### D-10 Commercial use
Status: proposed
Options:
- A: leave `unknown`, and use no noncommercial-only source, so the answer cannot change the material (rests on: questions.md section 1 row 19; ice-edit "Noncommercial material: never used on this route")
- B: mark `no` and allow CC BY-NC music (most Jamendo tension cues)
- C: mark `yes`
Choice: A
Reason: the slot is asked only when a noncommercial-only source would otherwise be used; every candidate kept is CC BY or a free commercial license, so the question never arises.
Sources: questions.md section 1; ice-edit routes contract
Left behind: B, the NC-licensed cues that would block any later commercial use.
Supersedes: none

### D-11 Versions
Status: proposed
Options: A one 9:16 master; B plus 30 s cutdown; C plus 16:9; D plus 15 s teaser (stage 01 question 9)
Choice: A
Reason: the brief asks for one reel; a cutdown would lose the payoff.
Sources: prompt
Left behind: B, the reflex 30 s cutdown.
Supersedes: none

### D-14 Disclosure: contact with the excluded official reel's analysis
Status: proposed
Question: what to do after the first lines of an earlier analysis of the excluded official reel were read by accident
Locked inputs: "الريلز الرسمي ... متشالة من البريف قصد ... فما تدوّرش عليه وما تستخدموش" (lead session)
What happened: while looking for earlier measurements of the programme's style, the first 30 lines of `_motion-sources/_research/aljazeera-references/4Zah8EHgtyM/BREAKDOWN.md` were printed in a batch with other files. Its title showed it is the official AJ reel of this same episode; reading stopped there. Those lines describe its length (78 s), its share of presenter time, its opening line (the presenter's intro on overcrowded classrooms) and its graphic devices.
Options:
- A: stop reading it, never open that file, the combined STYLE-BIBLE.md built partly from it, or the section of other files about it; take every decision from the two given references, the episode and this run's own measurements; disclose it here, in the delivery note and in the case study
- B: continue as if nothing happened
Choice: A
Reason: honesty about what could have influenced the result; the segment choice (D-06) came from a separate agent that never saw that file, and it picked a different part of the episode (a guest's argument, not the presenter's intro).
Sources: the lead session's original Arabic background, logged verbatim as run log entry 2 (`ice/log/conversation.jsonl`); the 30 lines read: `owner-research/aljazeera-references/4Zah8EHgtyM/BREAKDOWN.md` lines 1 to 30
Left behind: B, staying silent about it.
Supersedes: none

### D-16 Goal
Status: proposed
Options:
- A: the viewer understands the missile-versus-schools contradiction and wants the full episode (rests on: "اختار أقوى جزء فيها للريلز"; the reel is a programme promo on the episode's platforms)
- B: the viewer shares the clip as an opinion
- C: the viewer learns the facts of the protests (dates, numbers)
- D: the viewer follows the channel
Choice: A
Reason: a reel cut from an episode works as its trailer; the takeaway in `concepts.md` is the understanding, the sign-off names the programme.
Sources: prompt; `ice/plan/concepts.md`
Left behind: C, the fact sheet the clip cannot prove.
Supersedes: none

### D-17 Music direction (stage 01 question 5)
Status: proposed
Options:
- A: the real library track closest to both references by measurement, ducked under speech, 2 runners-up in the delivery note (rests on: lead session rule; `ice/reference/ref-*/sound_map.json`: D minor, about 84 BPM, cinematic underscore or drone, synth bass)
- B: no music
- C: darker, slower documentary bed
- D: percussive news-promo bed
Choice: A
Reason: Omar's standing rule; both references share key and tempo, which gives a measurable target.
Sources: lead session brief; `ice/audio/music/match-ref-a.txt`, `match-ref-b.txt`
Left behind: D, the generic news-promo bed.
Supersedes: none

### D-18 Look and tone (stage 01 question 6)
Status: proposed
Options: A the programme's measured look tied to the words; B quieter; C louder, more archive; D plain captions only
Choice: A
Reason: "بهوية البرنامج وألوانه وتوجهه"; the measured look is in the two ANALYSIS files and the contract in FIDELITY.md.
Sources: prompt; `ice/plan/FIDELITY.md`
Left behind: C, the archive-heavy "breaking news" look.
Supersedes: none

### D-19 Call to action (stage 01 question 8)
Status: proposed
Options: A the programme title as sign-off, said by the host; B add "شاهدوا الحلقة كاملة"; C add the channel handle; D no end card
Choice: A
Reason: both references end on the programme title when the presenter names it; the host's own line "في هذه الحلقة من محاولة فهم" gives that title its spoken reason; B and C would be words nobody says.
Sources: ref analyses; `ice/plan/script.md` row 7
Left behind: B, a written CTA card.
Supersedes: none

### D-20 Brand owner (stage 01 question 11)
Status: proposed
Options: A not owner (unofficial showcase); B owner; C agency for the owner; D unknown
Choice: A
Reason: the request is a showcase for the Ice site; nothing says the user is Al Jazeera. The delivery note lists the permissions Al Jazeera, AFP, France Télévisions, FRANCE 24 and the font's designer would need.
Sources: prompt; intake-and-brief.md section 1 (`brand_owner` default)
Left behind: B.
Supersedes: none

### D-21 Disclaimer wording (COMPLIANCE RD1, RD2)
Status: proposed
Options:
- A: RD1 as a name plate at the guest's first medium shot, his name and role as the episode description gives them: "د. عبد الكريم أمنكاي" / "أستاذ مساعد العلاقات الدولية، معهد الدوحة للدراسات العليا"; RD2 met without a label, because archive of 2022 and 2023 appears only while the speech names that past event
- B: add a date label on every archive shot
- C: no name plate
Choice: A
Reason: attribution is required (the claims are his analysis); date labels on every shot would add text over archive, which the programme never does.
Sources: `ice/sources/episode/info.json` description; COMPLIANCE.md RD1, RD2
Left behind: B.
Supersedes: none

### D-12 Approach (stage 02)
Status: proposed
Question: the strategy of the edit
Locked inputs: the paper edit `ice/plan/script.md` (stage 03, footage-led order); takeaway in `ice/plan/concepts.md`
Options:
- A: reading 1 of `concepts.md`, "the contradiction, proved" (rests on: D-06; candidates.md filmable list)
- B: reading 2, the guest alone with kinetic type
- C: reading 3, the ledger
Choice: A
Reason (the approach in 7 sentences): The reel keeps one argument, in the speakers' own words, cut only between sentences: the host's missile-or-high-schools line, then the guest on debt, "no money", the young, and the army. It opens on picture and sound together: the missile leaving the sea while the host says "a picture of Macron with a ballistic missile launch". The guest carries the middle; the screen keeps him lit and sharp and shows only the words he is saying, in the programme's three text roles. Real, dated footage answers each named filmable thing (the missile and Macron, the high schools and the young, the Ukraine war, the army spending) at the moment it is named, and the proof montage sits in a pause, as the references do (r2: the government, street and unions cutaways were dropped with D-13). Red marks attention on what is talked about (Macron, the missile, the army), never on the speaker. The rhythm is the programme's: long speaker holds that breathe with digital push and pull on phrase starts, broken by short hard-cut archive bursts, not the 1.2 s average shot of generic reels. It ends when the host names the programme and its title rises, as both references end.
Takeaway (quoted from `ice/plan/concepts.md`, r2): "Sells: the Al Jazeera programme محاولة فهم and its episode on why French students took to the street. Offer's job: none. One-line idea: critics in France reject the state's claim that it has no extra money for education, since it found resources to raise military spending. Literal anchors: no symbolic device."
Sources: `ice/plan/script.md`; `ice/plan/concepts.md`; `ice/plan/FIDELITY.md`; `ice/reference/ref-a/ANALYSIS.md`; `ice/reference/ref-b/ANALYSIS.md`; house-style-numbers.md (ranges used as diagnostics only)
Critic: reading 1 KEEP 7/10; note on the Macron source: use the France 3 1080p clip with the ministry images, not the 720p repost (ice/review/direction/critic-D06-D07-D12-D13.md) [independent]
Left behind: B, the talking head with big words.
Supersedes: none

### D-13 Rhythm and presence numbers (stage 02)
Status: proposed
Question: which rhythm ranges apply
Options:
- A: the references' measured presenter rhythm: speaker on screen about 60 to 70%, archive bursts in pauses and short cutaways on named things, digital zoom 1.00 to 1.10 on phrase starts, no shot of the speaker over 9 s without a reframe (rests on: ref-a 77% presenter, ref-b one take re-framed; podcast type "no shot over 9 s")
- B: the generic shorts house numbers (average shot 1.1 to 1.4 s, full B-roll 50 to 55%)
- C: speaker only
Choice: A, revised after the critic (r2): speaker on screen 75 to 85% (the references measure 87% and 79%, not 60 to 70%); archive only where a filmable thing is named (the missile and Macron in the hook, the high schools and the young in the pause, the Ukraine war, the army spending); the government, street and unions cutaways are dropped; no digital zoom on 720p archive (10 of the 18 clips); crimson only on the missile, Macron and the army archive, never on the students.
Reason: the user asked for the programme's montage; precedence puts the user's explicit instruction above house defaults (ice-direction section 5). The zoom stays at or under 1.10 on top of the vertical crop because the crop already enlarges the 1080p source 1.78 times (house ceiling 1.20 on 1080p).
Sources: ref analyses; house-style-numbers.md sections 2 to 4
Gates: default look PASS (D-07); distance from the nearest bundled style card, "Visual journalism" (`ice-styles/references/directions-openers-explainers.md` line 58: black with grain, licensed photos with source tags, yellow highlighter, condensed grotesk plus monospace, reading beats of 3 to 5 s): differs on palette (warm studio and crimson vs black and yellow), type (Arabic house sans with kashida vs condensed grotesk and mono), layout (full-bleed speaker with frosted plates vs photo boards with tags), motion (digital push and pull, disc swoop, archive strobe vs static reading beats) and sound (cinematic synth-bass bed with designed hits vs none named): 5 of 6, PASS; fingerprint: `~/.ice-motion-director/fingerprints.jsonl` does not exist on this machine, so there is nothing to compare (PASS, recorded as "no prior films"); default kit (anti-sameness.md section 6): neon glow (the programme title glow, the brand's own, measured in both references), glass cards with a light streak (the programme's own frosted plates, measured), centred text (the programme centres captions and plates) are present, each with that brand reason; no dark kinetic-type opener (the reel opens on footage), no gradient text, no AI symbols, no fake cursor, no page chrome, no template skeleton: PASS with reasons; richness floor: none applies, "Edit types and an unknown film type have no richness floor; the numbers are reported only" (`ice-qa/references/motion-density.md` section 6.1), so film_gate.py reports them at 09 without a red line.
Critic: A REVISE 7/10, applied in r2 (presence 75 to 85%, archive only on named filmable things, no zoom on 720p archive, red only on the missile and army) (ice/review/direction/critic-D06-D07-D12-D13.md) [independent]
Left behind: B.
Supersedes: none

### D-15 Written hook vs the reference's timing rule (scoped conflict)
Status: proposed
Question: ice-edit asks for a written hook readable from frame 1 for every reel cut from an episode; both references never show a word before it is spoken (the earlier measurement of ref-a checked about 30 elements, lags 0 to +30 frames, the one exception being the flag and label that answer "إنها")
Options:
- A: keep the reference rule; the first frame carries the hook through the picture (the missile launch) and the first spoken nouns become titles as they are said ("ماكرون" at about 0.5 s, "صاروخ باليستي" at about 3.2 s), and the contrast line becomes the first glass plate as it is spoken
- B: put a written hook over the first 3 s (for example "طب اصرف على الثانويات الأول") before it is said
Choice: A
Reason: the user's instruction ("نفس مونتاج ريلز محاولة فهم") outranks a house default; the purpose of the written hook (a muted viewer gets the hook from frame 1) is met by the picture plus the spoken-word titles. Logged as a scoped exception for this programme's style.
Sources: short-form-hooks.md section 6; ref-a measurement rule 11
Left behind: B.
Supersedes: none

### D-22 Grade and caption style (stage 05)
Status: proposed
Question: grade policy, caption style, layout and safe zones
Locked inputs: ice-edit grade policy (people untouched by default); D-07 type and palette
Options:
- A: speakers untouched; archive graded as the programme grades it (saturation 0.5, contrast 0.92, blacks lifted with a #221C1E lighten layer, over-exposed entry frames, crimson multiply only on war and army footage); the programme's three text roles at sizes that meet the vertical floors inside the union safe zone of Reels, TikTok, Shorts and Facebook Reels (x 65 to 888, y 288 to 1248)
- B: one grade over everything (a warm LUT)
- C: word-by-word karaoke captions
Choice: A
Reason: the references grade archive, never the presenter; the look is not open (two references and a real brand fix it), so no mood board is needed (ice-edit 05 row: "When the look is open ... a mood board first").
Sizes: captions Bold 86 px (one line, a 74 px fallback never needed after T10 was cut to "الحكومة تقول لهم"), plates Bold 62 px with 46 px sub-lines, question 90 px, kinetic titles Regular 128 to 150 px, programme title Bold 150 px.
Sources: ice-edit/references/ingest-and-ledger.md section 5; ice-engines/references/platform-specs.md section 3; ref analyses
Left behind: C.
Supersedes: none

### D-23 The reference's "title behind the head" (FIDELITY F5), adapted
Status: proposed
Question: keep titles behind the speaker's head in a vertical medium shot
Options:
- A: titles in front, in the chest band (y 870 to 1020) inside the safe zone; person mattes kept only for the disc behind Macron
- B: titles behind the head with the head matte, as the references do in 16:9
Choice: A
Reason: measured on stills (`film/out/stills/sheet2.jpg`): in the 9:16 medium shot the head covers x 330 to 800, leaving under 250 px each side, and the platforms' right margin (192 px on Shorts) takes most of the right side; split words became unreadable ("مـ ... ـتها"). The programme title now rises in front over the host, who is held and defocused after his last word.
Sources: `film/out/stills/sheet2.jpg`, `sheet3.jpg`; platform-specs.md section 3
Left behind: B.
Supersedes: the F5 implementation line in FIDELITY.md (proof now: the title rises on the spoken "محاولة فهم")

### D-24 Finishing layer, measured from the references and applied for a reason
Status: proposed
Options:
- A: what the references measure: no grain or film texture on any shot (aj360 breakdown: flat-area noise is compression only, no grain overlay, no scanlines; ref-b by eye the same), so none is added; archive "old record" look only through desaturation and lifted blacks; a blur-in of 6 frames on titles and plates (the references' plates enter soft); real subframe motion blur (8 subframes, 180 degree shutter) only on the two disc swoops, the one fast move; glow only on the programme title and the pins; vignette 0.38 to 0.42 at the corners
- B: add film grain and a light leak for "texture"
Choice: A
Reason: Omar's rule: the finishing layer is measured from the reference and applied for a reason; the references are clean broadcast pictures.
Sources: `owner-research/aljazeera-references/aj360/BREAKDOWN.md` "Finishing layer"; `ice/reference/ref-b/ANALYSIS.md`
Left behind: B.
Supersedes: none

### D-25 Engine
Status: proposed
Options:
- A: Ice's zero-install HTML capture renderer (`ice-engines/scripts/render-html.mjs`) with the installed Chrome 149: one page, frames drawn from extracted JPEGs, frosted plates with backdrop blur over the real footage, subframe motion blur, lossless FFV1 intermediates
- B: install HyperFrames (npm package plus its pinned Chrome download, about 150 MB that would land on drive C, which has about 1 to 3 GB free)
- C: FFmpeg only (no frosted glass, no kinetic Arabic type with kashida)
Choice: A
Reason: the user allowed installs, but HyperFrames' browser download would land on the nearly full drive C, and the bundled renderer gives everything this reel needs.
Sources: ice-engines/references/engine-choice.md sections 1 and 7.4; `film/out/final/main.log` (driver and browser lines)
Left behind: C.
Supersedes: none

### D-26 The episode's own red glow behind the guest (source 1681.6 to 1694.6)
Status: superseded by D-32 (a person matte keeps the face out of the correction)
Options:
- A: leave it: it is the programme's own quote-card graphic, subtle in the vertical crop
- B: neutralise reds with a selective desaturation
Choice: A
Reason: measured on a frame (source 1690): the selective desaturation took the guest's face saturation from 147.5 to 71.0 while the glow area only went from 53.4 to 44.2, which would grade a person (forbidden by default) to remove almost nothing.
Sources: `ice/sources/episode/probe/red_cmp.png`
Left behind: B.
Supersedes: none

### D-27 Music edit and mix
Status: proposed
Options:
- A: "Vanes" from its first bar with no cut or loop; its build (8.1 to 11 s) falls on the proof montage, its drop (36.7 s) near "عاجزة"; ducked to -17 dB under speech and -9 dB in the free gaps (attack 0.12 s, release 0.45 s), a 0.55 s dip before "وجدت موارد", out under the title hold; master -14 LUFS, true peak -1.5 dBTP (social profile)
- B: start the track at its main section
- C: loop or stitch sections to hit every event
Choice: A
Reason: one complete track with no stitching; its structure already meets the reel's turning points.
Sources: `ice/audio/music/match-ref-*.txt`; `ice/audio/cues.json`; ice-audio/references/mix-and-master.md section 5
Left behind: C.
Supersedes: none

### D-28 Free-font version
Status: proposed
Options:
- A: render a second master with Almarai (`?font=free`) for public posting outside Al Jazeera
- B: house font only
Choice: A
Reason: the house font's licence is for Al Jazeera Network only; a showcase on the Ice site needs a version that does not depend on it.
Sources: D-07; `ice/assets/CREDITS.md`
Left behind: B.
Supersedes: none

### D-29 Words over the war and army archive
Status: proposed
Question: the payoff's two named things ("حرب أوكرانيا", "نفقات الجيش") are spoken over archive, with no words on screen, so a muted viewer cannot link the found money to the army
Options:
- A: the spoken words as kinetic titles over that archive, white with a crimson glow while spoken, silver after
- B: keep the screen empty over archive (the references' habit)
Choice: A
Reason: the muted-first platforms (Facebook feed) need the payoff on screen; the words are the spoken ones, at their spoken time; it extends the hook's exception of D-15 to the two other named things shown as archive.
Sources: COMPLIANCE.md RD2 (the war label also dates the 2022 footage); clarity-gate.md section 5 (muted platforms)
Left behind: B.
Supersedes: none

### D-30 Cold read 3, round 1 fixes
Status: proposed
Question: the minor confusions of `ice/review/clarity/read-3-r1.md`
Options and choices (each fix is the smallest that removes the confusion):
- Titles faint red on dark: the hot colour moves from the measured #C23447 to #E8485C on the studio, titles over archive turn white with a crimson glow while spoken, and every title gets a soft dark scrim under its words only (vertical-conversion rule for text over footage).
- A Palestinian flag in the students' shot read as another cause: the shot (AFP T0KHYCHlk2U at 60.2 s) is replaced by pupils marching with "POUR L'ÉCOLE PUBLIQUE" (AJ xli_CIpdWgg at 66.4 s).
- No channel at the end: the network's name "الجزيرة" in the house font under the programme title (identification, not a call to action; no logo, because the only logo available is a 100 px broadcast bug that would be distorted when enlarged, a hard release gate).
- Left as they are, with reasons: the 0.0 to 0.5 s sea plume (the voice names the missile within 3 s and the title lands at 3.2 s); the disc read as decoration (it is the programme's attention mark, unchanged by design); the montage before its explanation (the programme's archive-in-the-pause order, explained 11 s later by the guest); the jump between the debt line and the "no money" line (two sentences of the same argument 5 s apart in the source, joined with a punch-in).
Sources: `ice/review/clarity/read-3-r1.md`; `film/out/stills/sheet7.jpg`
Left behind: rebuilding the debt-to-claim join with the dropped filler "يعني هو معيار ... بكل بساطة".
Supersedes: D-19 in part (the end now names the network under the programme title)

### D-31 The loaded word as it is said
Status: proposed
Question: critic 09 r1 item 7: is the word at reel 17.2 "مديونيتها"?
Options:
- A: "مديونية", the form both Whisper models hear unprompted (large-v3 0.90 and 0.93, medium 0.79 and 0.75, written "مديونة") and the best of three prompted forms (مديونية 0.69, مديونيته 0.56, مديونيتها 0.56)
- B: keep "مديونيتها" (the grammatical reading)
Choice: A
Reason: F1 and RC1 allow only words that were said; no ear in this run, so the measured forms decide; the envelope shows no stop closure for a "ت".
Sources: `ice/transcripts/wordcheck/results.json`; `scratch/wc3.json` (prompted forms)
Left behind: B.
Supersedes: none

### D-32 The red glow behind the guest, removed from the background only
Status: proposed
Question: critic 09 r1 item 2: the episode's quote-card glow reads as red attention on the speaker (owner's rule: red belongs to the people talked about)
Options:
- A: under an inverted person matte (rembg isnet-general-use per frame, feathered, smoothed over time), saturated reds of the background turn to the curtain's brown and lose the glow's extra light; the flying red dot (46.92 to 47.30) is inpainted
- B: leave it (D-26)
Choice: A
Reason: the matte removes D-26's objection: skin, lips and suit are never touched (the face is outside the weight), so no person is graded.
Sources: `tools/calm_red.py`; before and after frames S21/0362 and S21/0420
Left behind: B.
Supersedes: D-26

### D-33 Critic 09 round 1 fixes
Status: proposed
Question: the 7 P1 and 10 P2 items of `ice/review/critic-09-r1.md`
Options and choices (each the smallest fix that removes the defect):
- Words before speech (1): every caption, plate and title is built of word spans; each word appears 3 frames after its spoken start, from a fresh alignment (median of source Whisper, reel Whisper and the YouTube captions, moved to the first speech frame when an estimate falls in silence, `tools/align_words.py`). Titles cool to silver within 5 frames of the word's end.
- Face to the frame edge (3): the long guest take follows his face like a camera operator (holds within 40 px, eases after it otherwise, at most 150 source px/s; `tools/face_track.py`).
- "ماكرون" (4): left of his face, inside the safe zone, 45 px clear of his nose.
- Seams (5, 8): the hook ends at source 1703.76 ("باليستي" whole), the montage pause is 1.88 s (both superseded by D-38: 1703.72 and 1.92 s); the turn cuts at 1159.64, in the dip.
- Riser (9): 6 dB down until "الأول" ends, then its swell.
- Payoff breath (10): a 0.40 s held beat after "وجدت موارد"; the title cools inside it.
- Attention dip (11, option B): caption "بعض الهيئات", then "أجزاء من المجتمع الفرنسي".
- Subject of the framing line (12): caption "قسمًا من الشارع الفرنسي".
- Attribution (13): "فيه ناس تقول لك:" heads the first plate and stays with both plates until the montage; the plates go to 50 px so both fit under the host's chin and inside the safe zone.
- End (14): "محاولة" and "فهم" each 2 frames after their start, 3-frame pre-glow; the host holds on his last open-eyed frame (64.84).
- Hands under text (15): the name plate rises to 930; its sub-line breaks after "الدولية،".
- Accepted as they are, with reasons: the soft wide shot (16, it hides the join between rows 4 and 5); the drawn-out "يعني" in the hook (17, no clean cut at -26 to -36 dB).
Sources: `ice/review/critic-09-r1.md`; `film/out/stills/`
Left behind: option A of item 11 (cutting the fillers), because it would remove the guest's hesitation that the next line resolves.
Supersedes: D-26 (by D-32), D-30 in part (the attribution caption)

### D-34 The brand test at 09 for an unofficial edit
Status: refused by the approver (09 r1); superseded by D-39
Question: the masked-frame brand test of cold read 3 round 2 named no brand on any of the six random frames (0 of 6, all "I can't tell"; the viewer named "محاولة فهم" on Al Jazeera from the end title and the speech). Does it gate delivery?
Options:
- A: the identity tests do not gate this film: identity-tests.md section 1 runs them on films made for a named brand, and this one is not (D-20: an unofficial showcase for the Ice site, Al Jazeera owns the footage and is not the client); the result is reported as a finding in the delivery note and the case study, and the fix it points to is listed for the day Al Jazeera adopts the edit
- B: treat it as a 09 FAIL and change the look until random frames read as the programme without its name
Choice: A
Reason: B would have to add brand codes the programme does not use in its own reels (its reels carry no persistent mark besides the logo, which masking covers); inventing them breaks brand-authenticity.md ("a brand code is used as the brand really uses it, never invented to pass the test"). The clarity read itself, which ice-edit's 09 row requires for promotional types, is a separate gate and is judged on its own.
Sources: `ice/review/clarity/read-3-r2.md` item 0; `ice/review/identity/masked-09-r2.txt`; identity-tests.md sections 1 and 2; D-20
Left behind: B.
Supersedes: none

### D-35 Critic 09 round 2 fixes
Status: proposed
Question: the 5 P1 and 11 P2 items of `ice/review/critic-09-r2.md`
Options and choices (each the smallest fix that removes the defect):
- "ستعجز" before its sound (A): a word cannot start in a pause: an onset that falls where there is neither voicing nor 4 to 9 kHz fricative energy moves to the first frame that has one (`tools/align_words.py`); "ستعجز" is now keyed at 15.84, on screen 15.96.
- The flying dot (B): `tools/calm_red.py` inpaints any small round bright blob above the shoulders (over the head too), on every frame of both guest shots, before nothing else touches them; the bug that skipped frames with no glow and read the dot's brightness after dimming it is fixed.
- The students' shot (C): dropped; the aerial of the march holds through "لخروج أولئك الشباب" (no identifiable pupil, no party flag).
- Captions under the safe zone (D): the two-line captions T12a and T15 rise to 1020.
- "نفقات" on two chins (E): T17c sits at 990, under the chins.
- Rim and fringe (F): the person matte is eroded 7 px instead of dilated, so the correction reaches the jacket's outline; the desk fringe is the episode's own and is accepted.
- "موارد" late (G): a caption onset is moved back to the start of its sound, never across a pause; 60.61, on screen 60.73.
- Lips in the breath (H): the guest's picture holds on 61.08 for the 0.40 s breath.
- Scrim before the word (I): each title word carries its own soft scrim, revealed with it.
- Sway (J): the face tracker aims 0.4 s ahead (max offset 77 source px, was 97).
- Opening impact (K): 6 dB down after its first 80 ms.
- Attribution size (M): 44 px. "أطروحة كاذبة" (N): kashida 20 px shorter. Doubled first frames (P): three montage clips start one frame later.
- Accepted with reasons: hands under text (L, readable, no face covered); "أموال" (O) needs an ear like D-31 (the sentence requires it; Whisper writes "أمور"); the missile clip's own repeated frames (P, in the official source).
Sources: `ice/review/critic-09-r2.md`; `ice/review/fixes/`
Left behind: re-cropping the students' shot to the placard (its frames still hold faces at the edges).
Supersedes: D-32 in part (the dot), D-33 in part (the students' shot of D-30)

### D-36 The guest's background as delivered (grade)
Status: proposed
Question: approver 05 r1: D-22 says "speakers untouched", but from 46.9 to 61.5 s the guest's background is corrected (D-32, D-35 F)
Options:
- A: record the exception: the speaker himself stays untouched (skin, lips, hands, suit, hair: the person matte keeps them out of the weight, measured by the critic as no change in cheek colour between corrected and untouched shots); only the episode's red quote-card glow and its flying dot behind him are toned to the curtain's brown, and the matte is eroded 7 px so no red rim is left on the outline
- B: leave the glow (D-26)
Choice: A
Reason: the owner's rule puts red on the people talked about, never on the speaker; the correction grades the set, not the person.
Sources: `ice/review/fixes/red-glow-before-after.jpg` (round 1 matte), `ice/review/fixes/guest-matte-current-before-after.jpg` (the current, eroded matte); `ice/review/critic-09-r2.md` items 2 and F
Left behind: B.
Supersedes: D-22 in part (speakers untouched, except their background from 46.9 to 61.5 s)

### D-37 Round 2 fix check follow-ups
Status: proposed
Question: the critic's round 2 fix check (`ice/review/critic-09-r2.md`) and the approver's 07 and 08 returns
Options and choices:
- The disc over the back of Macron's head (0.50 to 2.70): on the frigate both men are one foreground matte, so the disc shows only on the sky; for the army budget shot the matte keeps his whole head and body, the soldiers further out stay behind the disc. The mattes are keyed by clip, not by shot number (a renumbered shot had lost its keep region).
- Doubled first frames (D-35 item P was wrong to call them fixed): every shot now starts on the first source frame at or after its in point (seeking between two frames made the fps filter repeat the first one); no shot starts on a repeated frame.
- Music after 61.1 s: a bug, not a choice: the fade read the first held frame (the payoff breath) instead of the host's last word. Restored to D-27's curve: the bed fades under the title hold only.
- True peak: the opening impact goes to -10 dB so the master meets D-27's -1.5 dBTP; D-27's target stands and NOTE.md states it.
Sources: `ice/review/critic-09-r2.md` "Round 2 fix check"; `ice/review/stages/07-approver.md`, `08-approver.md`; `ice/review/fixes/disc-matte-before-after.jpg`
Left behind: none
Supersedes: D-35 item P

### D-38 Row 1 out: the host's "باليستي" without the guest's "نعم"
Status: proposed
Question: approver 06 r1: the row 1 out at source 1703.76 cut inside a voiced stretch (1703.64 to 1704.06): whose sound is it?
Options:
- A: out at 1703.72, a frame boundary: after the closure (1703.60) comes the host's unvoiced /t/ release (1703.64 to 1703.70), and the guest's voiced "نعم" begins at 1703.72 (158 to 163 Hz; the host speaks at a median 122 Hz, the guest at 166 Hz); a cut test on the source hears "باليستي" whole at 1703.70 and 1703.72, "باليست" at 1703.62 and 1703.68
- B: out at 1704.08, keeping the guest's "نعم" over the missile footage
Choice: A
Reason: a word boundary that keeps the host's whole word and no fragment of another voice; B would put an off-screen "yes" on the archive.
Sources: `ice/transcripts/row1check/row1-out.json` (pitch per 40 ms, 20 ms envelope, Whisper words), `cut-test.json` (four out points); `ice/review/critic-09-r2.md` "Final check"; `tools/build_edl.py` r8
Left behind: B; 1703.76 (it kept 0.04 s of the guest's voicing); 1703.62 (the approver's closure, which cut the /t/ release: the critic heard "باليست" in 6 of 6 passes).
Supersedes: D-33 in part (its "the hook ends at source 1703.76")

### D-39 The brand test at 09: held for the owner
Status: resolved by the owner (D-40)
Question: the approver refused D-34 (09 r1): the masked brand test failed in rounds 2 and 3 (0 of 6 each), and after two failed rounds the look or the telling must change, using brand-owned elements as the programme really uses them
Options:
- A: the owner accepts D-34 (an unofficial showcase, not made for Al Jazeera), and the film ships with the brand test reported as a finding
- B: change the look so random frames read as the programme without its name: the programme's own reels carry no persistent mark besides the platform logo (ref-a: none; ref-b: the AJ360 logo, a platform mark that masking covers), and its codes (the curtained studio and its presenter, the house font with kashida, the crimson disc and pins, the glass plates) are already on screen; a further code would have to be invented, which brand-authenticity.md forbids
- C: deliver the free-font version only as a neutral edit with no programme identity (it would drop the user's "بهوية البرنامج")
Choice: none by the run: Ice holds stage 10 on a failed brand test, and no option B exists without inventing a brand code. The film files are complete and technically green; the delivery note says the run is held at 09 on this one item.
Sources: `ice/review/stages/09-approver.md`; `ice/review/clarity/read-3-r2.md`, `read-3-r3.md`; `ice/review/identity/09-r2.md`, `09-r3.md`; ice-qa identity-tests.md sections 1, 2 and 6
Left behind: none yet
Supersedes: D-34 (refused)

### D-40 The owner's decision on 09: the programme's own marks
Status: decided by the owner
Question: D-39 (the masked brand test failed twice; only the owner can close it)
Locked inputs: the owner's words, relayed by the lead session on 2026-10-29: "قرار عمر في المرحلة 09: الريلز يفضل بهوية البرنامج. العلامات الثابتة: ضيف لوجو الجزيرة واسم "محاولة فهم" في الأركان، بنفس مكان وحجم وشكل الريلز الرسمي. قيسهم من المراجع، أو من ريلز رسمي تاني للبرنامج غير ريلز فرنسا. كارت النهاية: يفضل زي ما هو. الاختبار: اعمل قراءة الهوية تاني بعد الإضافة، وقفّل 09 و10. الموقع: عمر وافق ينزل الريلز ده على موقع Ice، ومعاه سطر صغير بيقول إنه مش إنتاج رسمي من الجزيرة."
Options:
- A: the two persistent marks of the programme's official vertical reels, taken from those reels' own pixels at the same position and size (the Al Jazeera logo top left, x 95 to 194, y 160 to 320; the stacked "محاولة فهم" wordmark with its red dot top right, x 853 to 1017, y 200 to 298, on 1080x1920), laid on with a screen blend, on the whole reel and leaving as the programme title rises (the official reels show none on their blue end card); the end title stays as it was
Choice: A (the owner's direction)
Reason: the owner keeps the programme's identity and directs its real marks; they are used as the programme uses them, not invented. Measured on the official reels ZUhNxnrVuew (Venezuela episode) and 1dradF-qrwI (drones episode), same canvas, same pixels, plus the programme promo O5IjhyK3hd8 (logo only). No reel of the France episode was opened: three of them (fQdvsPusdb8, tFlzbvMiH7U, 5b9CVmnYnZU) appeared by title in the channel's listing of shorts and were left alone (D-14).
Sources: `ice/assets/brand-marks.json`; `tools/brand_marks.py`; `film/marks/marks.png`; `ice/reference/marks-ZUhNxnrVuew/`, `marks-1dradF-qrwI/`, `ref-c/`
Left behind: none
Result: masked brand test round 4 FAIL, 0 of 6 (the masking covers the marks); with the marks visible the fresh viewer named "Al Jazeera" at first impression and "Mohawalat Fahm" with the sound off (`ice/review/identity/09-r4.md`). The owner's words close 09 and 10 after this test. Critic marks check: SHIP.
Supersedes: D-39 (resolved by the owner), D-07 in part (it left the logo out because only a 100 px broadcast bug existed; the official vertical reels give it at this reel's own size), D-34

### D-41 Whose claim the turn is
Status: proposed
Question: approver 09 r3: cold read 3 round 4 asked at 51 s whose claim "أطروحة كاذبة" is (a major confusion at the turn); with the channel's logo now in the corner it could read as the channel's verdict
Options:
- A: the said attribution held on screen with the claim: "أجزاء من المجتمع الفرنسي تقول:" as a small quiet header (46 px Regular, silver) above "أطروحة كاذبة" from 51.11 to the cut to the war footage, the title itself held through "لماذا؟", and the caption before it given its verb ("... الفرنسي تقول"), as "فيه ناس تقول لك:" heads the host's plates (D-33)
- B: a caption under the title instead
Choice: A
Reason: the words are the guest's own, already said; the header keeps the turn attributed while the claim and the question are on screen, and stays clear of his chin (y 800 to 860).
Sources: `ice/review/stages/09-approver.md` round 3; `ice/review/clarity/read-3-r4.md`; stills `film/out/stills/t052.60.png`
Set at the RTL start of the line (right), as a quote's source, not centred over the title: a small centred line over a large one is the web "kicker" pattern the engine lint fails (C7).
Left behind: B (it would sit under the "لماذا؟" pill); a centred header (lint C7).
Supersedes: none

## Heads-up
- Every stage 01 question took its recommended option (`ice/plan/stage01-questions.md`); any of them can be changed.
- The official reel of this episode was not searched or used; see D-14 for the one accidental contact and how it was contained.