feat: persona jet-set insert reels + random 2–6 scroll gaps

[SteveSimonson]

Fixes #64

Summary

• 6 jet-set insert films from house persona avatars (Vivienne private jet Alps, Camille Dubai helipad, Noor Capri yacht, Margot Milan tarmac, Isla Maldives seaplane, Aurelia Monaco harbor)
• Registered in REELS_JETSET / full insert pool
• Links to /quiz (vibe check) with “Find your persona” CTA
• Insert protocol: unique reel ids per page; random 2–6 product gaps; day-stable seed
• Surfaces: Shop, Home, PDP similar; /reels + /watch Jet set section

Risk

Low-medium: larger static assets (~21MB video). Placement algorithm change may show more inserts than before on long shop lists.

Test plan

• Shop grid: inserts at irregular 2–6 gaps; no duplicate video on page
• Jet-set card → /quiz
• Category card still → shop room
• /reels jet set section plays all 6
• /watch Jet set filter
• tsc / CI green

[SteveSimonson]

Independent code review — PR #65

Reviewed full gh pr diff 65, focused sources (reelInserts.ts, reels.ts, ReelInsertCard.tsx, call sites), and ran npx tsc --noEmit (exit 0).

Checklist

Focus	Result
No duplicate reel ids on one page	Pass — shuffle pool, seen by id, place uniqueReels[reelIdx] without wrap/modulo
Random gaps 2–6	Pass — randInt(rand, minGap, maxGap) with defaults 2–6; lead-in + successive placement; slot = after N products
Persona reels ignore excludeCategory	Pass — !r.category || r.category !== exclude keeps jet-set in pool
REELS_JETSET → /quiz via reelHref	Pass — each jet-set sets href: '/quiz'; reelHref falls back to /quiz when no category
ReelInsertCard persona vs category CTAs	Pass — persona: badge “Persona”, kicker, “Which woman are you today?”, CTA “Find your persona”; category: “Discover” / shop CTAs; analytics persona_jetset vs cross_promo
Optional category on CategoryReel	Pass — typed optional; filter/href/CTA/analytics all guard; tsc clean
npx tsc --noEmit	Pass (exit 0)

Insert algorithm (src/lib/reelInserts.ts)

• Pool = full CATEGORY_REELS (gen1 ∪ gen2 ∪ jet-set).
• Uniqueness enforced after shuffle; no reels[i % n] reuse.
• Density: maxByDensity = floor(n / avgGap) with avgGap = (minGap+maxGap)/2, then min(unique, density, maxInserts).
• Day-stable seed retained (listName|exclude|UTC day|product ids).
• Legacy every still maps to fixed min=max gap.

Standalone simulation (40 products, exclude handbags, maxInserts 10, multiple seeds): always unique ids, all gaps ∈ [2,6], 0 handbags reels, jet-set still eligible.

Data & surfaces

• Six jet-set entries match vibe ids (luxe…gilded); assets ~20MB video total under public/brand/videos/reels/.
• /reels Jet set section + room sections; /watch jetset filter; Shop/Home/PDP pass minGap/maxGap.
• Docs in docs/amazon-insert-video.md match behavior.

Nits (non-blocking)

1. REEL_LATEST_GENERATION = 3 — gen 3 is persona-only. reelForCategory() looks for gen 3 category clips, finds none, falls back to last list item (gen 2). Correct today; slightly misleading for future category waves. Prefer keeping latest category gen separate, or document that gen 3 is jet-set only.
2. Stacking guard (slots[last] === pos → pos+1) is effectively dead while minGap >= 2. Harmless; could drop or comment.
3. No unit tests for interleaveReelInserts (uniqueness / gap bounds / exclude + persona). Worth a small pure test given density/cap changes.
4. Repo weight — ~20MB binary MP4s (called out in PR risk). Fine if intentional; R2/mint later would help clone/CI size.
5. Home density — maxInserts up to 4 (was 1–2) with 2–6 gaps is intentional but denser; watch short home rows so they don’t feel video-heavy.

Verdict rationale

All stated product requirements are implemented correctly, type-safe, and tsc-green. No functional blockers found. Nits above are polish / future-proofing only.

VERDICT: APPROVE_WITH_NITS