---
name: game-soft-launch-review
description: Design or audit a game soft-launch learning and exposure program across regions, cohorts, acquisition sources, FTUE, core/meta loops, retention, economy, IAP/IAA, live events, store creative, crashes/performance, support/community, global-transfer risk, and automated scale/hold/pause/withdraw gates. Use when a production-shaped game is entering bounded real-player exposure; use Game Design for the game itself and Distribution Readiness for store submission/certification.
---

# Game Soft Launch Review

Produce a **Game Soft-Launch Program** that converts bounded real-player
exposure into causal evidence and automated release decisions without reducing
the declared final game to a conventional MVP.

## Atomic boundary

Own cohort/market design, exposure stages, hypotheses, telemetry, economy/live
ops test schedule, acquisition/creative cells, pass/watch/fail predicates,
decision authority, and scale/hold/pause/withdraw evidence. Do not own the game
   design, channel submission, full marketing operating system, or provider payment
system.

Use a draft artifact ID and explicit inputs from Game Design, Economy,
Distribution, Marketing, Analytics, Payment, Refund, and Support owners. A
deterministic delivery process may seal serialized versions and digests later;
never invent a digest or treat a draft as proof.

## Agent-first invariant

Soft launch is an exposure and proof system, not permission to omit production
architecture. Build the complete declared target, automation, migrations,
observability, support, low-end, accessibility, localization framework, content
pipeline, and shutdown controls before exposure. Construct all test cells and
decision automation now; activate spend, regions, features, and concurrency
through bounded gates. Unknown data never authorizes unbounded scale.

## Workflow

1. Define traceable game, build, content, economy, and policy identities—draft
   IDs during design and observed exact values only during exposure—plus
   genre/audience/age modes, target global promise, regions/platforms,
   acquisition sources, monetization, cohort
   horizon, budget/spend caps, and irrecoverable harm boundaries.
2. Read `references/game-soft-launch-patterns.md`. Write falsifiable hypotheses
   for comprehension, delight/mastery, FTUE, core/meta loops, progression,
   social/liquidity, content, economy, IAP/IAA, performance, support, and market
   transferability.
3. Build a cohort cell matrix across install date, version/content/economy
   policy, region/locale, platform/device tier, acquisition source/creative,
   age mode, payer state, and experiment exposure. Prevent cross-cell leakage.
4. Set event/identity/consent contracts and pass/watch/fail bands for cold start,
   activation, D1/D7/D30 retention, session quality, progression, economy,
   IAP/IAA, crashes/ANRs, latency, battery/thermal, store conversion, support,
   safety, accessibility, localization, and trust.
5. Sequence deterministic simulation, agent QA, internal playtest, low-risk
   cohorts, market cells, spend cells, and scale steps. Synthetic agents expand
   coverage but never prove human comprehension, culture, usefulness, trust, or
   delight.
6. Separate product defects, cohort mix, acquisition quality, creative/listing,
   novelty, live-ops cadence, economy tuning, price/offer, and regional effects.
   Use holdouts and immutable version/policy identities.
7. Define daily automated readback, anomaly/fraud detection, support/community
   evidence, content/economy calendar, migration/compensation, and exact
   scale/hold/pause/withdraw/rollback/forward-fix actions.
8. Produce a decision record that states evidence, uncertainty, countermetrics,
   global-transfer limits, next bounded exposure, and authority.

## Source verification

Retrieve current store/test-track, regional availability, ads/IAP, child/age,
privacy/consent, reviews/community, refund, promotion, and acquisition-platform
authority for every active cell. A test market or platform permission is not a
global compliance or product-fit verdict.

## When not to use

- Use `game-design-blueprint` when the primary job is designing or repairing
  core play, progression, content, social, feel, retention, or monetization.
- Use `software-distribution-readiness` for submission assets, certification,
  reviewer access, staged release mechanics, and exact live-channel evidence.
- Use `marketing-automation-blueprint` for the complete multi-channel creative,
  organic, lifecycle, paid-spend, attribution, and shutdown operating system.
- Use `launch-readiness-review` for a cross-domain final admission verdict when
  the soft-launch learning program is already complete.

## Guardrails

- Never scale blended metrics that hide version, market, device, source,
  creative, age, payer, or experiment cohorts.
- Do not over-monetize early cells and then treat damaged retention as product
  truth; preserve baseline/no-paid-acceleration reachability.
- Purchases, durable progress, social relationships, and creations survive
  normal experiment/season/region changes; resets require explicit migration,
  compensation, support, and authority.
- One market is evidence for its cell, not global proof. Cultural/locale and
  channel differences remain hypotheses until tested.
- No autonomous system may raise spend/exposure caps or weaken quality, safety,
  consent, payment, economy, or rollback gates to chase growth.

## Output contract

Return one typed Game Soft-Launch Program with:

1. exact artifact identities, hypotheses, cells, budgets, authorities, and ruin
   boundaries;
2. exposure DAG from simulation/playtest through regions, spend, and scale;
3. cohort/identity/consent/event contract and immutable experiment policy;
4. pass/watch/fail gate matrix with confidence/uncertainty and machine actions;
5. FTUE, loop, progression, economy, IAP/IAA, live-ops, performance, support,
   community, safety, accessibility, localization, and store evidence plan;
6. anomaly/fraud, pause/withdraw, rollback/forward-fix, migration, compensation,
   and live-readback controls;
7. scale/hold/pause/withdraw decision record and next bounded exposure.

Complete only when every cell and metric maps to an action, unbounded spend is
impossible, exact versions are traceable, and global transfer claims remain
bounded by evidence.