Rule: GROUNDED = 3+ independent shippers; STRONG = 2 (or 1 with internal data); 1 = hypothesis. Radar data is the falsification layer — principles failing data get downgraded, noted in record.
GROUNDED: (1) Simple mechanics, tons of depth — every session new. (2) Strong core fantasy sells the click (kid-sleep "what if" test). (3) Clippability drives organic growth (design moments creators clip; watch what they clip). (4) Ideas abundant, execution/talent is the bottleneck.
STRONG: (5) Demand signals over vibes. (6) Console+mobile UX is a ceiling, not a detail (Tizzy internal data ×2). (7) Onboarding = emotional investment in 90s, not tutorial completion (benchmarks in tizzy note).
HYPOTHESIS: (8) Socialization in core loop = flywheel (Tizzy ×2). (9) Iteration volume wins (patches daily, 9 tutorials). WATCH: (10) live-ops event formats spike CCU ("admin abuse" → 60k). (11) Thumbnail = promise of feeling, mobile-readable (data test pending vs our 523-thumbnail dataset).
Open contradictions (our strategic questions): focus vs portfolio (his focus; our factory — resolve wide-early/obsessive-after-launch); talent irreducible vs process compresses (our factory's bet); embedding-discovered taxonomy vs verb/thematic grammar (run both, reconcile).
Usage: GROUNDED rows are defaults in specs — violating them needs explicit justification. STRONG/HYPOTHESIS rows get data verdicts as Radar matures.