מסחריות >70: "שירים מוכרים ומסחריים בלבד — להיטים שכולם מכירים"
מסחריות <30: "חובה שירים נישתיים, עמוקים ופחות מוכרים! אסור להיטים! תבחר B-sides, אמנים עצמאיים, שירים שרוב האנשים לא מכירים. זה הדבר הכי חשוב!"
מסחריות 30-70: "תמהיל — חצי שירים מוכרים וחצי פחות מוכרים"
עברית (כל ערך): "בדיוק {X}% מהשירים חייבים להיות בעברית ו-{100-X}% בלועזית. זה חובה!"
ווקלי >70: "העדפה לשירים ווקליים עם שירה בולטת"
ווקלי <30: "העדפה לשירים אינסטרומנטליים או עם מינימום מילים"
אנרגיה >70: "אנרגיה גבוהה — שירים קצביים, תוססים, אפטמפו"
אנרגיה <30: "אנרגיה נמוכה — שירים רגועים, איטיים, נינוחים, chill"
familiarity <40 (אזהרה כפולה): "⚠️ תזכורת: הלקוח ביקש שירים נישתיים ופחות מוכרים. אל תביא להיטים! תחפש אמנים עצמאיים, אלבומים פחות מוכרים, שירים עמוקים."
🧬 DNA עסקי → injection לפרומפט
"🏷️ סיווג עסק: {label} DNA מוזיקלי: {vibe} ז׳אנרים מומלצים: {baseGenres} להימנע מ: {avoid} טווח BPM: {bpmRange[0]}-{bpmRange[1]} אנרגיה בסיסית: {energy} אמנים מייצגים: {refArtists + hebrewRef}"
⚡ V5 TURBO ARCHITECTURE
Priority Hierarchy: 1. Hard constraints (real tracks, max 2/artist, language) 2. Slider overrides → 5-tier system 3. Business intent signals 4. Music persona style 5. BIZ_DNA base 6. Soft preferences
Music Personas: Curator · Entertainer · Tastemaker Ambient Designer · Eclectic DJ Auto-detected from sliders + DNA
5-Tier Sliders: Each slider: 0-20 | 21-40 | 41-60 | 61-80 | 81-100 + Cross-effects between sliders e.g. high energy + low familiarity = underground dance
Flow Engine: Smooth transitions (BPM/key/genre/mood) Energy micro-arcs per time slot Genre clustering (2-3 then switch) Hebrew spread (never >2 in a row) Surprise injection every 6-8 tracks
Anti-AI Layer: 12% surprise picks · 15% discovery tracks No obvious genre representatives Era spread across decades Avoid algorithmic playlist clichés
Enhanced Sonic Attributes: warmth · brightness · density · groove texture · moodProfile · grooveLevel
Conflict Resolution Engine: commercial + luxury → premium deep cuts low energy + bar → lounge social-low high hebrew + global → Israeli-global sound high discovery + short dwell → familiar genre, new artist All conflicts detected and resolved explicitly
Persona as Behavior System: riskCurve · familiarityCurve · transitionStyle genreFlexibility · tempoBehavior emotionalConsistency · discoveryStrategy
🌊 V7 LIVE ARCHITECTURE
Music = Control System · Crowd Simulation · Delta Thinking
Principle Zero: MUSIC IS A CONTROL SYSTEM What is happening → What should happen → How music moves it Every song is an ACTION, not content
Crowd Simulation Engine: crowdSize · density · energyLevel · mood engagement · socialState · turnoverRate alcoholLevel · timeOfStay · noiseLevel · trend → Not just state — WHERE it's going
Delta Thinking: Each song scored by IMPACT not FIT energyDelta · moodShift · engagementChange riskLevel · momentumEffect fit ≠ enough. impact = critical.
Decision Loop (per track): 1. Read crowd state (or simulation) 2. Compare to target 3. Calculate delta 4. Select candidates by impact 5. Choose track 6. Update crowd state 7. Check failure modes 8. Repeat
Momentum Model: rising → take risks, push forward stable → maintain, gentle evolution falling → stabilize with safe tracks peak → ride wave, plan cooldown recovery → safe choices, gradual rebuild
Failure Mode Prevention: energy_spike · energy_flat · vibe_loop panic_commercial · flow_break hebrew_cluster · momentum_crash Each detected and fixed in real-time
Control Strategy: escalationPlan · recoveryPlan · peakDesign fallbackLogic · energyCeilings · riskProfile Music manages crowd, not entertains it
⚙️ V7.5 OMEGA ARCHITECTURE
Decision State Engine · Triple-Gate · Budget System
12 Fundamental Principles: 1. Decision State (7 vars per track) 2. Compare→Select (8 candidates, discard 50%+) 3. Triple-Gate Rejection 4. Flow as State (not post-hoc) 5. Budget System 6. Mandatory Track Role 7. Forced Tradeoff 8. Self-Compression (5-7 principles) 9. Mini-Critic 10. Operational Rules Only 11. Relative Judgment 12. Balance Only When Needed
3-Phase Pipeline: PHASE 1: Self-Compression → Compress all context to 5-7 core principles → Initialize budget system (risk, familiarity, surprise, vocal_density) → Set initial state (energy_baseline, target_mood, language_quotas)
PHASE 2: Generate + Compare + Select (per position) → 8 candidates per position → Triple-gate: brand/slider fit → flow/energy fit → better than alternatives → Forced tradeoff declaration → Mandatory role assignment → Mini-critic (too predictable? too intense? repeating?) → If 2+ flags → reject
PHASE 3: Validation Pass → Max energy delta 0.3 between adjacent → Max 3 vocal-heavy in a row → Max 2 same-language in a row → Saturation: max 2 same genre in 4-track window → Fix violations by swapping/repositioning
Decision State (7 Variables): target_energy · target_familiarity · target_role language_state · risk_budget · brand_guard · avoid_now → Updated after EVERY track selection
Budget System: risk_budget (0-100): How much "surprise" left to spend familiarity_budget (0-100): Mainstream allowance remaining surprise_pressure (0-100): Need for unexpected picks vocal_density_budget (0-100): Vocal track allowance → Budgets deplete as playlist builds → Forces diversity naturally
Triple-Gate Rejection: Gate 1: Brand/Language/Slider fit? → REJECT if fails Gate 2: Flow/Energy/Repetition fit? → REJECT if fails Gate 3: Better than alternatives? → REJECT if not best → Minimum 50% rejection rate ensures quality
Track Roles (Mandatory): opener · connector · lift · settle deepener · peak-touch · recovery → Every track MUST have a declared role → Prevents "filler" tracks
Forced Tradeoff: For each track, declare what was SACRIFICED: "interesting vs accessible" "branded vs flowing" "surprising vs stable" → Makes decisions explicit, prevents lazy choices
🔍 צפייה בגרסאות מוח:
📋 כללים קשיחים (בכל פרומפט)
"מקסימום 2 שירים לאמן" "כל השירים חייבים להיות אמיתיים וקיימים ב-Spotify" "חלק שווה morning/afternoon/evening/night (~12 שירים)" "DNA = נקודת מוצא, סליידרים = override"