Full review · Sample · 1.0.8 and 1.1.0 · Sep 1–21, 2026

Sample 1.0.8 and 1.1.0, Sep 1–20: full review

New players leave at the first purchase in the tutorial, and paid UA does not pay back at any bid

What changed since v001. The data window extends to Sep 21, the ROAS line in 3.3 was recounted. Everything else as before.

1Summary

New players of Sep 1–20 by the build of their first launch. Every number is made up: the example shows what each block looks like.

Key metrics by version

Metric1.0.81.1.0All players
Players in the sample11894212
Players, estimate ×10≈ 1,180≈ 940≈ 2,120
Retention D131.2%n = 10935.1%n = 7732.8%n = 186
Retention D312.4%n = 9714.0%n = 5713.0%n = 154
Retention D75.1%n = 79—n = 05.1%n = 79
Active time per player, min11.8median 5.312.6median 5.412.2median 5.3
Sessions per player2.90median 2.03.10median 2.03.00median 2.0
Average session, min4.1median 1.94.1median 2.04.1median 1.9
Interstitials per player3.403.603.50
Rewarded per player0.600.900.70
ARPU (ads), $$0.0120$0.0150$0.0133
Ad revenue in the sample, $$1.42$1.41$2.83
Ad revenue, estimate ×10, $≈ $14.20≈ $14.10≈ $28.30

Cohort: devices whose first event is in Sep 1–20, 2026 (v_players); columns: version of the first event. D1/D3/D7: an event on day 1/3/7 after the first, over devices whose day is over (n under the value). Active time: gaps between events inside a session, each capped at 10 min; a session is an AppMetrica session_id. Ads: ad_revenue_events, revenue in USD as sent. A 10% device sample: 'estimate' rows = sample × 10; rates and per-player values stay as they are.

Short answers

What did 1.1.0 change

More players finish the first match, no more reach the second. 74% vs 62%, second match 45% vs 52% (2.3).

Where do we lose new players

At the first purchase in the tutorial and in the first 30 seconds of play. 21.2% and 11.3% leave (2.2).

A second paragraph inside an answer works too.

Does paid UA pay back

No. A player costs $0.40 and brings $0.13 (3.3).

2Product metrics

Everything about the game except money: installs and retention, drop-off, the first matches, what players do before they leave.

2.1 New players and D1 by day

OrganicPaidUnattributed
from Sep 15: new build (example) 0 10 20 30 40 50 Sep 1 — Organic 26; Paid 0; Unattributed 2 · total 28 28 Sep 2 — Organic 27; Paid 0; Unattributed 1 · total 28 28 Sep 3 — Organic 25; Paid 0; Unattributed 3 · total 28 28 Sep 4 — Organic 29; Paid 0; Unattributed 2 · total 31 31 Sep 5 — Organic 24; Paid 0; Unattributed 2 · total 26 26 Sep 6 — Organic 28; Paid 0; Unattributed 1 · total 29 29 Sep 7 — Organic 27; Paid 0; Unattributed 2 · total 29 29 Sep 8 — Organic 26; Paid 0; Unattributed 3 · total 29 29 Sep 9 — Organic 30; Paid 0; Unattributed 1 · total 31 31 Sep 10 — Organic 25; Paid 0; Unattributed 2 · total 27 27 Sep 11 — Organic 27; Paid 0; Unattributed 2 · total 29 29 Sep 12 — Organic 26; Paid 0; Unattributed 1 · total 27 27 Sep 13 — Organic 28; Paid 0; Unattributed 3 · total 31 31 Sep 14 — Organic 25; Paid 0; Unattributed 2 · total 27 27 Sep 15 — Organic 11; Paid 0; Unattributed 1 · total 12 12 Sep 16 — Organic 10; Paid 0; Unattributed 1 · total 11 11 Sep 17 — Organic 9; Paid 7; Unattributed 4 · total 20 20 Sep 18 — Organic 12; Paid 20; Unattributed 5 · total 37 37 Sep 19 — Organic 10; Paid 14; Unattributed 3 · total 27 27 Sep 20 — Organic 11; Paid 12; Unattributed 2 · total 25 25 Sep 21 — Organic 9; Paid 31; Unattributed 7 · total 47 47 Sep 1 Sep 3 Sep 5 Sep 7 Sep 9 Sep 11 Sep 13 Sep 15 Sep 17 Sep 19 Sep 21 September, day of month · made-up data
From Sep 15 organic drops about 2.5×, paid starts on Sep 17. Hover a bar for the split.
Build 1.0.8Build 1.1.0
switch 0% 10% 20% 30% 40% 50% UA starts Sep 1 — Build 1.0.8 31% Sep 2 — Build 1.0.8 30% Sep 3 — Build 1.0.8 32% Sep 4 — Build 1.0.8 29% Sep 5 — Build 1.0.8 30% Sep 6 — Build 1.0.8 31% Sep 7 — Build 1.0.8 28% Sep 8 — Build 1.0.8 30% Sep 9 — Build 1.0.8 29% Sep 10 — Build 1.0.8 31% Sep 11 — Build 1.0.8 30% Sep 12 — Build 1.0.8 29% Sep 13 — Build 1.0.8 30% Sep 14 — Build 1.0.8 31%; Build 1.1.0 30% Sep 15 — Build 1.1.0 33% Sep 16 — Build 1.1.0 35% Sep 17 — Build 1.1.0 34% Sep 18 — Build 1.1.0 36% Sep 19 — Build 1.1.0 35% Sep 20 — Build 1.1.0 37% Sep 21 — Build 1.1.0 36% 36% Sep 1 Sep 3 Sep 5 Sep 7 Sep 9 Sep 11 Sep 13 Sep 15 Sep 17 Sep 19 Sep 21 share who came back the next day · made-up data
line: two series with a gap (None), markers, the value at the end of the line, a band and a vertical mark.

Preliminary. The last seven days are still arriving: D1 for Sep 19–20 can grow. A .note sits next to the number it qualifies.

2.2 Drop-off by steps and by time

StepReachedLostLost, % of previousReached, % of cohortReached, estimateStep time, s: median · p90 · max
Cohort: first event212——100.0%≈ 2,120—
Loading: SDK20752.4%97.6%≈ 2,0700.4 · 1.1 · 38.0
Loading: ads188199.2%88.7%≈ 1,8807.8 · 24.0 · 3,899
First scene18442.1%86.8%≈ 1,84010.7 · 31.0 · 1,357
Tutorial: first step17952.7%84.4%≈ 1,790—
Tutorial: first purchase1413821.2%66.5%≈ 1,410—
Active time, m:ssReachedLostLost, % of previousReached, % of cohortReached, estimate
Cohort: first event212——100.0%≈ 2,120
0:301882411.3%88.7%≈ 1,880
1:00171179.0%80.7%≈ 1,710
1:30160116.4%75.5%≈ 1,600
2:0015195.6%71.2%≈ 1,510
2:3014296.0%67.0%≈ 1,420
3:0013664.2%64.2%≈ 1,360
3:3012975.1%60.8%≈ 1,290
4:0012543.1%59.0%≈ 1,250
4:3011964.8%56.1%≈ 1,190
5:0011097.6%51.9%≈ 1,100
5:3010643.6%50.0%≈ 1,060
6:0010154.7%47.6%≈ 1,010
6:309833.0%46.2%≈ 980
7:009533.1%44.8%≈ 950
7:309144.2%42.9%≈ 910
8:008833.3%41.5%≈ 880
8:308533.4%40.1%≈ 850
9:008322.4%39.2%≈ 830
9:308033.6%37.7%≈ 800
10:007822.5%36.8%≈ 780

Cohort: 212 devices whose first event is in Sep 1–20, 2026 (v_players). A step counts when the device reached it or a later one. Active time: gaps between events inside a session, each capped at 10 min; every session of the device. A 10% device sample: 'estimate' rows = sample × 10; rates and per-player values stay as they are.

0% 2.5% 5.0% 7.5% 10.0% 12.5% 0:30: 11.3% 1:00: 9.0% 1:30: 6.4% 2:00: 5.6% 2:30: 6.0% 3:00: 4.2% 3:30: 5.1% 4:00: 3.1% 4:30: 4.8% 5:00: 7.6% 5:30: 3.6% 6:00: 4.7% 6:30: 3.0% 7:00: 3.1% 7:30: 4.2% 8:00: 3.3% 8:30: 3.4% 9:00: 2.4% 9:30: 3.6% 10:00: 2.5% 0:30 1:30 2:30 3:30 4:30 5:30 6:30 7:30 8:30 9:30 active time, m:ss · share lost of those who reached the interval · n = 212
Most players leave in the first 30 seconds of active time; after that the loss is steady, 3–6% per interval.

Interpretation. The first purchase in the tutorial needs money a new player does not have yet: 21.2% never get past it. This is a reading of the numbers, not a fact — H1 tests it.

2.3 The first matches

Hero purchase · N in sessionDevices% of cohortMedian, m:ssMean, m:ss
118084.9%0:422:20
213061.3%2:254:42
39544.8%4:228:09
47033.0%6:0411:24
55124.1%8:0015:08
Theft · N in sessionDevices% of cohortMedian, m:ssMean, m:ss
115070.8%1:333:50
211252.8%3:316:51
38339.2%5:1710:45
45827.4%6:5016:23
54119.3%8:0918:41

Cohort: 212 devices whose first event is in Sep 1–20, 2026 (v_players). The Nth action is the Nth time within one session; time is since the session's first event (wall clock, includes backgrounded time). A 10% device sample: device counts are the sample (× 10 for an estimate); shares stand as they are.

Build 1.0.8 (neutral .s-old)Build 1.1.0
0% 10% 20% 30% 40% 0 · Build 1.0.8: 16.2% 16.2% 0 · Build 1.1.0: 20.8% 20.8% 1 · Build 1.0.8: 32.4% 32.4% 1 · Build 1.1.0: 34.0% 34.0% 2 · Build 1.0.8: 13.2% 13.2% 2 · Build 1.1.0: 9.4% 9.4% 3–5 · Build 1.0.8: 19.1% 19.1% 3–5 · Build 1.1.0: 20.8% 20.8% 6–10 · Build 1.0.8: 13.2% 13.2% 6–10 · Build 1.1.0: 9.4% 9.4% 11+ · Build 1.0.8: 5.9% 5.9% 11+ · Build 1.1.0: 5.7% 5.7% 0 1 2 3–5 6–10 11+ matches started over the lifetime · share of new players · made-up data
grouped_bars: values above the bars when they fit. The distributions nearly match.
New players1.0.81.1.0Difference
Players11894−24
Finished the first match62%74%+12 pp
Started a second match52%45%−7 pp
Playtime, median5.3 min5.4 min+0.1
Sessions in total342291−51

.tnote under a table: cohorts, window, filters. td.n numbers, td.ok / td.bad a verdict, tr.total a total.

Interpretation. 1.1.0 moved the loss from the first match to the step to the second one. 7 pp on 94 players is a direction, not a proven effect.

2.4 Leaving after negative events

EventDevices with itBackgrounded within 20 sBack within 20 sNothing else within 20 sNever came back
Hero stolen12046.7%56 devices53.6%30 devices20.8%25 devices5.0%6 devices
Got hit15040.7%61 devices55.7%34 devices14.7%22 devices1.3%2 devices
Interstitial started13164.9%85 devices50.6%43 devices22.9%30 devices0.8%1 devices
Hero purchase (neutral)18022.8%41 devices73.2%30 devices0.0%0 devices0.0%0 devices

Cohort: 212 devices whose first event is in Sep 1–20, 2026 (v_players). Shares are of devices that had the event at least once; backgrounded = a pause event (app_pause) within 20 s, back = of those backgrounded; nothing else = no other event within 20 s; never came back = no event at all after it. A 10% device sample: device counts are the sample (× 10 for an estimate); shares stand as they are.

Interpretation. After a theft 46.7% background the game — twice as often as after a neutral purchase (22.8%).

0% 5% 10% 15% 20% 25% Backgrounded the game within 20 s of a theft: 25%Backgrounded the game within 20 s of a th… 25% Left to the menu mid-match: 16%Left to the menu mid-match 16% Restarted after losing a race: 9%Restarted after losing a race 9% Doubled a reward with an ad and left: 6%Doubled a reward with an ad and left 6% Pressed nothing: the last event is a level load, long and without a progress bar: 3%Pressed nothing: the last event is a level l… 3% share of new players who left · a long label is clipped, the tooltip has the full text
Every fourth player who left backgrounded the game within 20 seconds of a theft.

3Revenue metrics

Ads by format, placement and network; revenue concentration; paid acquisition.

3.1 Ads by placement

PlacementImpressions in the sampleImpressions, estimatePer playereCPM, $
Interstitial after a match612≈ 6,1202.93.10
Rewarded: double the reward118≈ 1,1800.66.40
Rewarded: second chance31≈ 3100.15.90
Total761≈ 7,6103.63.72

≈ an estimate from a 10% device sample (× 10), italic .est. Shares and per-player values stand as they are.

Network ANetwork BNetwork CNetwork DNetwork ENetwork F
0% 10% 20% 30% 40% week 1 · Network A: 34% 34% week 1 · Network B: 22% 22% week 1 · Network C: 15% 15% week 1 · Network D: 12% 12% week 1 · Network E: 10% 10% week 1 · Network F: 7% 7% week 2 · Network A: 31% 31% week 2 · Network B: 24% 24% week 2 · Network C: 14% 14% week 2 · Network D: 13% 13% week 2 · Network E: 11% 11% week 2 · Network F: 7% 7% week 3 · Network A: 29% 29% week 3 · Network B: 21% 21% week 3 · Network C: 18% 18% week 3 · Network D: 12% 12% week 3 · Network E: 12% 12% week 3 · Network F: 8% 8% week 1 week 2 week 3 six series: the whole categorical palette .s1–.s6 in order · made-up data
Six categorical colours in order. There is never a seventh series: the rest folds into "other".

3.2 Revenue sits with a few players

Top fiveEveryone else
$0 $0.25 $0.50 $0.75 $1.00 $1.25 #1 player 1: $1.17 #2 player 2: $0.97 #3 player 3: $0.76 #4 player 4: $0.26 #5 player 5: $0.23 #6 player 6: $0.19 #7 player 7: $0.16 #8 player 8: $0.14 #9 player 9: $0.12 #10 player 10: $0.11 #11 player 11: $0.10 #12 player 12: $0.09 #13 player 13: $0.08 #14 player 14: $0.07 #15 player 15: $0.06 #16 player 16: $0.06 #17 player 17: $0.05 #18 player 18: $0.05 #19 player 19: $0.04 #20 player 20: $0.04 #21 player 21: $0.04 #22 player 22: $0.04 #23 player 23: $0.03 #24 player 24: $0.03 #25 player 25: $0.03 #26 player 26: $0.02 #27 player 27: $0.02 #28 player 28: $0.02 #29 player 29: $0.02 #30 player 30: $0.01 #31 player 31: $0.01 #32 player 32: $0.01 #33 player 33: $0.01 #34 player 34: $0.00 #35 player 35: $0.00 #36 player 36: $0.00 #37 player 37: $0.00 #38 player 38: $0.00 #39 player 39: $0.00 mean $0.13 median $0.04 top 5 = 67% of total 1 5 10 15 20 25 30 35 39 made-up players, ad revenue, descending
Five players pull the mean up; the median is several times lower.

3.3 Paid acquisition

Break-even revenue per player (example): cost $0.40, ROAS 100% → need $0.40 per player (now $0.13, ×3.1) cost $0.40, ROAS 50% → need $0.20 (×1.5)
In profit (.s-good)At a loss (.s-bad)
−$4 −$2 $0 $2 $4 Sep 12: −$3.20 −$3.20 Sep 13: −$1.50 −$1.50 Sep 14: $0.40 $0.40 Sep 15: $2.10 $2.10 Sep 16: −$0.80 −$0.80 Sep 17: $1.60 $1.60 Sep 18: $3.30 $3.30 Sep 19: −$2.40 −$2.40 Sep 20: $0.90 $0.90 Sep 21: $1.20 $1.20 Sep 12 Sep 13 Sep 14 Sep 15 Sep 16 Sep 17 Sep 18 Sep 19 Sep 20 Sep 21 revenue minus spend per day · made-up data
bar with negative values: a bar grows down from zero, only the data end is rounded.

Console · all countries · Sep 17–20

A made-up console screenshot: five blue bars on a white card

What the screenshot shows and what it does not.

Console · US · Sep 17–20

A made-up screenshot: orange bars

Console · US · Sep 21

A made-up screenshot: green bars

4Technical data

Technical problems and analytics defects. .cards: .hot loss, .adc ads, .okc good.

38%of events

Duplicated income

One payout arrives as two events: economy totals are doubled.

346errors

False ad timeouts

342 of them are real impressions in the same second.

60FPS

Performance is fine

An .okc card.

Telemetry details

A card without a number or a modifier; AdImpression as code.

5Countries

The country split in one place.

CountryPlayersD1Playtime, medianARPU, $Share of revenue
US8138.3%6.1 min0.03171%
Brazil6431.1%5.0 min0.0049%
India6727.4%4.6 min0.0025%
All21232.8%5.3 min0.013100%

6Hypotheses

What we think is going on. Each hypothesis rests on numbers from blocks 2–5 and says how to test it.

H1 confidence high

The first purchase in the tutorial is too expensive

Rests on: 21.2% of those who reach the "first purchase" step leave there (2.2).

How to test: give starting money for the purchase; confirmed if the loss at that step falls below 10% at n ≥ 150.

H2 confidence medium

After the first match the player is not offered a next step

Rests on: 45% start a second match vs 52% (2.3); players leave right after the results screen (2.4).

How to test: the share starting a second match ≥ 60% in a new build, same countries.

H3 confidence low

Paid UA in the US can pay back only through rewarded ads

Rests on: rewarded eCPM 6.40 vs 3.10 for interstitials (3.1), a player costs $0.40 (3.3).

How to test: 7-day ARPU ≥ $0.40 for a paid cohort after rewarded placements are added.

7Recommendations

Ordered by payoff. .acts numbers them itself.

Give starting money for the first purchase

Grant the price of the first purchase when the tutorial starts.

Expected effect: the loss at the "first purchase" step from 21% to ~10%, +20 of 212 players reach the game.

How to measure: ga.py dropoff on the new build a week later, n ≥ 150.

→ H1

A "Next match" button on the results screen

The main button starts the next match; the garage comes second.

Expected effect: second-match starts from 45% to 60%.

How to measure: the share starting a second match among those who finished the first, new players.

→ H2

Buy traffic to measure, not to earn

A small budget per build, one country.

Expected effect: 40 new players per build for $16 — builds compared without organic noise.

How to measure: the key metrics of block 1 by version.

→ H3

Data I need

Method