Player economics by release

Per player, the game earns as much as ever — slightly more. What fell is how long players stay. Revenue per daily active player (ARPDAU) was $0.261 in August and $0.307 on v36; ad revenue per player is up 2.6×. Meanwhile D1 fell from 29% (v29) to 22% (v36) and D7 from 8% (v26.2) to 5% (v34). Revenue totals follow DAU, and DAU follows UA spend and retention — so this report measures averages and explains totals only in Appendix A.

ARPDAU and payer figures are all active players on those calendar days. *Retention and progression are the players who installed that build while it was the one new players received (v36 has no D7 yet; v25 and the August builds are not split out). Excludes China (third-party APK traffic, no revenue).

2a. Which build changed difficulty — v34 (25 Sep), then v36 rebalanced it

Losses per stage clear, players grouped by the build they installed. v34 raised the walls at stages 21–31, 54–56 and 77–100; v36 eased 54–56 and 77–78 but made 32–53 and 57–76 harder. Players reaching stage 60 went 10–14% (v26.2–v32) → 8% (v34) → 6% (v36).

2b. The D1 drop from 12 Oct (26% → 20%)

It is a day-1 return problem, not a difficulty or traffic problem: players reach stage 10 as often as before (40% → 40%) and the 12–17 Oct installs reweighted to the earlier traffic mix still give 20.3%. Players who played normally on day 0 (reached stage 10) came back less: 37.5% → 31.6%.

Does a non-Tier-1 D1 drop matter?

For revenue this week, a little; for where UA can scale, yes. Non-Tier-1 players are 67% of installs but 28% of lifetime revenue ($0.31 per install vs $1.56 for Tier 1). A non-Tier-1 player who returns on day 1 is worth $0.98, one who does not $0.08 — so at 531 non-Tier-1 installs a week the drop costs about $43 a week. It matters because GA ROAS Global — the only campaign near payback (Appendix B) — is 89% non-Tier-1: its ROAS falls with its D1. Tier 1 also dipped (25% → 23%), so this is likely a v36 change that hits everyone and shows most where players are least committed.

Why. Day 0 is unchanged, so the first session still works; something between session 1 and session 2 changed with v36 (96% of these installs). Candidates the build team can confirm: push notifications, offline/idle rewards on return, the reworked Battle Pass, or new purchase popups at the start of session 2. GA4 cannot separate them — v36 added no new tracking events.

3a. After the first purchase: payers keep playing, most stop buying

Of payers, 82% return in the week after their first purchase and 55% play 3+ days of it. Of those 419 engaged payers, 39% buy anything that week — 25% buy a daily, weekly, monthly or membership offer. Payers who later leave keep playing a median 7 days after their last purchase.

Why. The recurring offers work for the players who take them — they are 27% of all IAP and 322 of 802 payers have bought one — but three in four engaged payers do not. The two highest-priced weekly offers barely sell (weekly.promotion ~$34, 6 buyers; weekly.garden ~$21). We cannot tell whether the rest saw an offer and declined, or never saw one: the game logs no offer-shown event. Adding an offer impression event (offer id, price, placement) is the instrumentation that would answer this.

3b. Whales

33 players spending ≥ $100 are 33% of IAP. They buy for a median 5 days, 34% of their spend lands on install day and 76% by day 6. They progress far faster (median stage 172 by day 3 vs 62 for other payers), and 40% of those who left played a week or more after their last purchase. Only 3 installed on v34 or later.

Why. Whales buy power to progress, outrun the catalogue within about a week, and leave. The new systems (section 4) were meant to be that sink, but fewer than half of whales used any of them, typically for 1–3 days.

3c. Removing the no-ads sale pack (v32, 21 Sep)

It did not work. First 7 days per install, IAP only, weighted to the same country-tier and campaign mix:

Why. The $10.99 permanent pack was a first-48-hours buy for price-sensitive new players. Without it, the $29.99 pack sold 10 → 5 times (nobody traded up) and some moved to the $10.99 30-day pack, which refunds at 32%. Other IAP rose over the same weeks, but with the v34/v36 packs, not from former no-ads buyers. Restore the $10.99 permanent pack.

4a. New systems: Alchemy, Astrology, Treasure, Loot Village

At most a small effect on how long existing players stay, and almost no revenue. Few eligible players adopt them, and those who do use them for 1–2 days.

Last column: players already past the unlock stage when the system shipped, active days in the 14 days after vs the 14 days before, next to the same ratio for the two weeks before that (natural decay). A system that held players would push the first number above the second: Alchemy and Astrology sit slightly above (+0.07), Treasure and Loot Village level — at most a small effect. Players who use a system are more active afterwards than matched non-users (6–11 vs 2–6 active days), but they are the engaged players to begin with, so that is not evidence of effect.

Why. They are side activities without a reason to return daily or to spend: uptake is 14–38% of eligible players, median use is 1–2 days, and they sell almost nothing (Loot Village storage $196, Alchemy $114, Astrology $16, Treasure $0). Loot Village also launched in v25, the build that broke the purchase flow and the Research tutorial, and its 14% uptake at an early unlock (~stage 26) suggests players do not find it. To work as a sink they need a daily loop (timers, daily rewards worth collecting), progression that matters for the main game, and an IAP hook — plus discovery for Village.

4b. Live events

Event and pass packs earned $1,392 (7.3% of IAP), $857 of it the Tiny Guide pass. Against non-participants who played and spent the same beforehand: Lucky Machine did not raise spending during the event, but participants were more active afterwards (8.1 vs 5.6 days in the next 14; 32 players). The Flameborn pass window shows $2.12 more spend per participant (28 players).

Why. Events reach the engaged 10–20% and act as retention for them; they do not create new spenders. Judge Dice and Fishing on participant retention and pack conversion among stage ≥ 101 players.

1. Find the v36 session-2 change behind the D1 drop (notifications, return rewards, Battle Pass, popups). Retention is what fell; this is the newest loss.
2. Review the v34/v36 difficulty walls at stages 21–31 and 54–78: reach to stage 60 halved since v29.
3. Log offer impressions, then fix what engaged payers are shown in week 1–2: 61% of them buy nothing that week.
4. Give the new systems a daily loop and an IAP hook (and surface Loot Village); today they are tried once and dropped, including by whales.
5. Restore the $10.99 permanent no-ads pack and fix the 72-hour auto-refund of unacknowledged purchases.
6. UA: keep spend on GA ROAS Global; Tier 1 at $5–11 CPI cannot pay back at current per-player numbers; repair the Facebook revenue postback before judging FB.

Weeks are Monday–Sunday, matching the publisher report. One source per number: IAP = Play net (Play’s order ledger; the publisher’s “after tax” is within ±15% of it every week), ads = the MAX figure in the publisher report (GA4 records only 20–45% of it), spend = AppsFlyer (identical to the publisher report).

Why our weekly report (+68%) and the publisher’s (−27%) disagreed for mid-October: ours ran Friday–Thursday, theirs Monday–Sunday. The v36 launch spike on 9–11 Oct ($240, $172, $280 IAP per day) fell in our current week and in their previous week. Both were right; all reports now use Monday–Sunday.

D7 ROAS is 35% (AppsFlyer, all paid cohorts): an average paid install costs $1.75 and returns about $0.61 by day 7, and later days add little (1.19× by D30, 1.23× by D60). Per campaign, GA4 behaviour over AppsFlyer installs:

Why. ARPPU ($20–26) is fine; retention and payer rate are too low for the price. GA ROAS Global is the only campaign near payback; Tier 1 pays more often but costs 4.8× as much. Facebook shows $0 IAP in AppsFlyer while GA4 sees its payers — the AppsFlyer revenue postback for Facebook is broken.

• All figures exclude China. GA4 through 18 Oct (the export stopped 19 Oct 13:18 UTC); Play orders through 18 Oct; AppsFlyer through 20 Oct.
• Install-build cohorts compare builds on their own first week; same-date comparisons are used where two builds overlapped. v33 is excluded (138 installs, no stage losses or purchases logged, 4.9% crash rate).
• Matched comparisons (events, systems) pair each participant with non-participants in the same activity and spend bracket beforehand; small groups are directional.
• Scripts: notebooks/rev1021/{analysis,money_weeks,deep_why,followups}.py; SQL dashboard/rev1021_sql/*.sql; builder dashboard/build_truth_report.py.