Practitioner playbook — Tizzy (producer, top-100 game, $119k/mo)

Decide rubric (game picks, pre-prototype)

1. One-line mechanic (the verb) 2. Depth from that one mechanic (replayability) 3. Curiosity hook — what unlocks tomorrow 4. Social flex element 5. Strong core fantasy — kid-sleep test: phrase it as the "what if" a 7-year-old imagines at bedtime 6. Clippable moment for creators 7. Console+mobile UX pass. Fail 3+ = no prototype. Additions from "five guarantees" video: 8. 90-second test (threat/stakes felt in first 90s; one mechanic at a time; minimal UI) 9. 10x-more-fun-with-friends (social in core loop by design) 10. Every-session-new (depth operationalized) 11. Mobile friction budget (tap count of core loop = CCU ceiling predictor — he estimates his own game capped at 60k vs 100-150k CCU purely from mobile tap friction).

Onboarding benchmarks (his internal data)

New-user first-session retention: good ≈ 95% @30s, 87% @1min, 65% @5min; bad = 20–30%. Funnels: track micro-steps (open shop / bought / close shop separately), slice by device. Cutscene cut bounce 22%→15%. Tutorial completion rate alone is vanity.

Dress playbook (thumbnails)

Thumbnail = promise of a FEELING (power/anticipation/outplaying), readable at mobile size. Rubric: scene communicates fantasy; emotion readable; mechanic teased; 1–2 characters max; high color contrast; simple background; text never covers key objects. Concepts: before/after, first-person, two-character conflict, character+action. Test 2–3 thumbnails/week; PTR decays with impressions (expected). He researches by VERB (searches "hack" on creator exchange) — validates verb/thematic classifier grammar.

Radar validation set (labeled calls to test against)

My Giant Sandwich (called 200→20k CCU), airport game (50→30k CCU). Anti-call: Clean the Library (blowing up, he can't explain why). Exemplars: Sell Lemons (onboarding), Build a Plane (community longevity), Escape Tsunami (level-design onboarding), 99 Nights (social loop).

Promise signals (beyond CCU, computable from public API)

Session length (CCU×1440÷daily visits), favorites/visits ratio, rising daily CCU floor (retention), like ratio within genre, badge depth, age-normalized velocity (CCU/day since creation), visit velocity leading CCU (external traffic), update cadence. Filter fake movement first: flat plateaus, 4am-US activity, high CCU + ~zero session length.