OpenOpen
Business strategy and founder lessons

@ayoships

Decompiling the free game Block Blast, the creator claims an on-device XGBoost model scores each player's ad value and then tunes difficulty (easier, then "death algorithm") to trigger rewarded-ad revives; a reverse-engineering explainer that went viral at 20.8M plays.

20.8M plays · 14 Oct 2026

What they said

  • Block Blast: free puzzle game, App Store rating 4.9 (2.8M ratings), age 13+, chart #7 Casual. On-screen ad revenue chart: Block Blast $127M, Vita Mahjong $89.8M, Candy Crush Saga $81.1M. They say: over $100M in ads, more than Candy Crush.
  • Core claim: how well you do depends less on skill and more on how much you are worth to them in ads.
  • Inputs shown in code: phone model (EcpmFeatureConfig PHONE map such as iPad16,6: 1, iPhone17,2: 12, iPhone16,2: 23, iPhone9,4: 120), country (COUNTRY map ST:1, US:2 ... DZ:238), past ad revenue from you, even whether vibration is turned off (settings screen with Sound/BGM/Vibration).
  • All feeds an XGBoost regression model (remote config "xgboost_regression_model111.json", runModel() loops tree splits) that runs locally on the phone and predicts your ad value.
  • Depending on value, the game gets easier or harder.
  • Rule "Boost Then Kill": secretly makes it easier so you think it is the best run of your life up until your average game length.
  • If a really expensive ad is ready, it turns on you. "High Value Ad Harvest" trait (RewardHighValueAdHarvestTrait, HIGH_PRICE_MULTIPLIER = 3.5, MIN_HISTORY_AD_WATCH_COUNT = 3) confirms you have a history of watching ads, helps you beat your high score (currentScoreThreshold = currentGameHistoryHighScore + scoreDeltaStep*(t+1)), then triggers "Death Algorithm" (shouldTriggerDeath, setAlgorithmBottomList([deathAlgorithm])) that rigs the next block set so you cannot survive.
  • On loss, first thing shown is an offer to revive by watching the same ad (Revive_Proxy, E_Revive_Show, E_Advertisement_RewardOver).
  • If you revive, they add 10% to your high score secretly (AdPlusMinusStrategyTrait onAdd1ReviveSuccess: Math.floor(.1*e) ... storage classScore) so you keep coming back.
  • They run experiments on the exact minute to make you lose and measure which gets you to watch the revive ad (MABXiBaiTrait with arms 2:240, 3:360, 4:600, 5:1e3, 6:1800 seconds; reward = reviveSuccess / reviveShow).
  • Conclusion: "Block Blast isn't really a puzzle game. It's an ad harvest with a rigged game loop built to squeeze every penny out of you."
  • CTA: comment which app to decompile next. Caption: full technical explanation + code evidence will be on his website (bio) soon; submit email to get notified.

Try this

  • Creators: Copy the structure of this reverse-engineering explainer. Give every claim a visual receipt, such as highlighted code, a chart or a screen capture. That is why it reached 20.8M plays.
  • Treat the claims carefully: The reel says the game's code scores each player's ad value and tunes difficulty to trigger rewarded-ad revives. That is the creator's reading of decompiled code, and the game's maker has not confirmed it.
  • Ecommerce and app sellers: The useful lesson is that offers placed at a moment of strong emotion convert, such as post-purchase upsells and cart recovery messages.
  • Avoid dark patterns: Rigged difficulty or hidden pressure damages trust, especially for premium brands.
  • Agencies and freelancers: Run a teardown series on competitor sites, ad libraries or funnels, and gate the long write-up behind an email sign-up.

Links

  • Block Blast! (iOS game) - subject.
  • XGBoost - ML model type named.
  • Decompiled code snippets (trait classes) - shown as evidence; no repo or tool name given.
  • Website in bio (email waitlist) - full write-up not yet out.

Watch the original

Watch the full reel on Instagram by @ayoships. Follow them for more.

00:00

/

00:00

More Business strategy and founder lessons

Liked this one? Get the next

When I find a reel worth acting on, you get the short version: what it says and what to do. Free, rare, one click to leave.

Rather have it done for you? Work with Tanish