Lucky Machine, hero packs and heroes per player

Tiny Knightfall · Android, non-China · GA4 export through 2026-10-26 · built 2026-10-27

Answers

Lucky Machine was not a measurable net improvement in ARPU, ARPPU or retention. For stage ≥ 31 players, ARPDAU went $0.28 → $0.33 → $0.29 (before / during / after). The during-vs-before gap is $0.05 (90% CI −$0.13 … $0.24), and it is two or three large buyers: payers were 27 / 27 / 26. Veterans came back at the normal rate, and players who reached stage 31 during the event retained no better (D7 16.5%, vs 23.8% before and 18.6% after; the lower 'before' gap follows the late-September traffic change). Its own revenue was $91.72 from 8 pack purchases (6.3% of IAP in its window); the two ticket packs sold 0.
Bring it back, but as a measured test, not as a proven earner, and don't build 2–3 more early events yet. The stage-31 gate did reach players: 40–49% of stage 31–100 players spun. They did not pay, though: every pack dollar came from stage ≥ 101, and early reachers' retention didn't move. An early event is cheap engagement, not revenue. Re-run Lucky Machine with a remote-config holdout and the two defects fixed, and judge the early-event idea on that result.
New hero packs: the Nun pack works, the rest are small, and none of them grew the payer base. The Nun pack ($9.99) sold to 23 buyers / $248 in 5 weeks, the best new pack. The other four together made $136. 1 of 31 buyers made their first-ever purchase on a hero pack; buyers are deep (median stage ~330) existing payers, and ARPDAU and payers per 1,000 DAU did not rise after launch. So they mostly re-route existing payers' money.
Heroes per player: almost everyone has only the starter. Estimated, because no event logs a hero. 68.7% of players who reach stage 31 own only Pom (mean 1.45 of 16); payers average 2.3, mostly because a first purchase grants Knight. The free routes barely land: Volt (5 ads) reaches 18.7% of stage-31 players and Sai (7-day login) 2.6%. Each pack hero is owned by under 2%.

1. Was Lucky Machine a net improvement in ARPU, ARPPU or retention?

No. The 10 event days look a little better on ARPDAU, but the difference is inside the noise, payer counts are flat, and the retention measures are flat or lower. The event added $91.72 of its own pack revenue.

Stage ≥ 31 players (the event's gate), all builds

Stage ≥ 31 players	Players	Player-days	ARPDAU (IAP + ads)	ARPDAU excl. top payer	Payers	Payer rate	ARPPU (10 days)	Top payer $
Before (09-06 – 09-15)	655	2058	$0.28	$0.23	27	4.1%	$20.82	$91
Lucky Machine (09-16 – 09-25)	586	1876	$0.33	$0.28	27	4.6%	$22.15	$90
After (09-26 – 10-05)	502	1440	$0.29	$0.23	26	5.2%	$14.23	$76

During − before: ARPDAU $0.05 (90% CI −$0.13 … $0.24); payer rate +0.5 pt (90% CI -1.4 pt … +2.3 pt). Bootstrap over players. 09-28 (export gap) excluded.

Daily ARPDAU of stage ≥ 31 players, 7-day trailing mean

Retention of existing players: followed from one 10-day block into the next

Followed from → into	Players (stage ≥ 31, ≥ 3 days active)	Came back at all	Activity kept (days ratio)	Payer rate before	Payer rate after	IAP per player after	… excl. top payer
08-07 → 08-17	388	62.9%	0.54	17.8%	4.1%	$0.32	$0.25
08-17 → 08-27	401	59.1%	0.53	18.0%	5.2%	$1.46	$1.16
08-27 → 09-06	393	62.3%	0.53	13.7%	3.1%	$0.44	$0.33
09-06 → 09-16 — event	323	65.0%	0.57	10.8%	4.3%	$1.30	$1.02
09-16 → 09-26	294	65.0%	0.55	9.2%	4.8%	$0.80	$0.54
09-26 → 10-06	209	71.8%	0.61	12.0%	5.7%	$1.47	$1.20

Same rule for every row: players at stage ≥ 31 who were active ≥ 3 days in the first block, all builds (intent to treat). "Activity kept" = share of days active in the second block ÷ share in the first. The event row sits inside the range of the other rows on every column.

Retention of players who reached stage 31 (the moment the event appeared)

Players reaching stage 31	Players	D1	D3	D7	IAP per player, first 7 days	Payer rate, 7 days	Spun the machine
Before — reached 31 on 09-02 … 09-14	517	49.5%	34.2%	23.8%	$2.33	8.3%	1.0%
During — reached 31 on 09-16 … 09-24, event build	278	47.1%	28.4%	16.5%	$1.29	5.8%	51.4%
After — reached 31 on 09-29 … 10-07	199	50.3%	33.7%	18.6%	$1.87	7.0%	0.0%

D7 during − before: -7.2 pt (90% CI -12.1 pt … -2.6 pt). Half of the "during" group spun the machine, yet D1–D7 were not higher. The dip vs "before" matches the traffic change of late September (UA campaigns ended; D1 fell game-wide for 25–30 Sep installs), so it isn't evidence the event hurt; "after" is on the same traffic and is level with "during". Day N = active on day N (D3/D7 also accept N+1 so the 09-28 gap doesn't count as churn).

Who played and who paid, by stage

Highest stage during the event (event build)	Players	Spun at least once	Median spins (spinners)	Pack buyers	Pack $	Reached 10,000★ (Leo)
31–50	95	40.0%	28	0	$0.00	0
51–100	181	48.6%	32	0	$0.00	0
101–200	88	31.8%	34	1	$9.99	3
201+	181	30.4%	16	6	$69.11	4

217 players spun (6,915 spins). Ticket packs (x10 $4.99, x120 $49.99): 0 purchases. 7 players reached the 10,000★ milestone, which grants Leo, the same SS hero the $19.99 Divine Cat pack sells (that pack has sold 2).

Why the event couldn't show much. (1) Reach: on day 1 only 11.2% of players were on an event build (v28/29/32/33/34), rising to 80.4% by day 10. (2) Two defects reported on 10-02: ad tickets granted without an ad (87% of ad-ticket grants have no ad), and a check-in double claim (41 players). Free tickets compete with the ticket packs. (3) Spin prizes and pack grants are not logged, so the economy can't be audited. (4) There was no holdout. Every comparison here is before/after, so a +20% effect would be invisible at this scale (~27 payers per 10 days).

Re-run recommendation

Re-run it: it is already built and costs little. Ship it on one build that is ≥ 90% of DAU before the start date.
Fix the ad-ticket grant (ticket only after ad_impression_MAX on rewareded_mini_event) and the check-in double claim.
Hold out 20% of eligible players by remote config (event hidden). That turns the next run into an answer: ARPDAU, payer rate and D7 for event vs holdout.
Log spin prizes and pack grants (earn with the SKU as position, as every other pack does).
Reconsider giving away Leo at 10,000★ while the Divine Cat pack sells him at $19.99, or make the milestone reachable only with paid tickets.

2. Should we add 2–3 more events triggered this early?

Not yet. The stage-31 gate brought in players (40–49% of stage 31–100 players spun vs ~30% at 101+), but none of them paid for the event and their retention didn't improve. Two or three more events of the same kind would add engagement we can't show is worth anything. What would change this: the Lucky Machine re-run with a holdout showing a D7 or payer-rate gain for stage 31–100 players.

Gate reach (unlock report, v34 installs, by day 3): stage 30 = 14% of installs, stage 65 = 5.5%, stage 101 ≈ 3%. An early gate reaches about 4× the players, but they are the ones who don't pay yet.
Where the money is: every Lucky Machine pack dollar came from stage ≥ 101 players. The 10-14 sizing found stage ≥ 101 players are ~25% of DAU and 61% of IAP.
Dice Adventure and Fishing (planned, gate 101) already target the paying group. Run them as planned, each with the same holdout and grant logging, rather than adding new early events.
If the goal is early retention, the problem is elsewhere: the v36/v37 difficulty wall at stage 34 (losses 10% → 40–45%) and the D1 cliff around stage 24. An event at stage 31 sits right in front of that wall.

3. How effective are the new hero packs?

The Nun pack is the one that works (23 buyers, $248, ~$51/week). Hero's Growth, SS Hero + Toys, Soulstone and the Divine Cat pack together made $136. They have not grown revenue overall: buyers are existing, deep payers, and game-wide ARPDAU and payers per 1,000 DAU are flat across the launch weeks.

Hero packs (non-China, through 10-26)

Pack	Live since	Buyers	Revenue	$ per week live	Was buyer's first purchase	Median stage	Median prior spend	Tapped (logged)
Lord of Hell — The Nun ($9.99)	09-23	23	$248.26	$51.11	4.3%	326	$52.35	12
Divine Cat — Leo ($19.99)	09-22	2	$44.08	$8.82	0.0%	343	$124.41	9
SS Hero + Toys — Candle ($4.99)	10-11	6	$31.34	$13.71	0.0%	392	$65.88	3
Hero's Growth, 24 h per hero ($9.99)	10-11	4	$43.88	$19.20	0.0%	314	$18.69	15
Hero Soulstone ($4.99)	10-05	3	$17.03	$5.42	0.0%	673	$20.98	1
Selective Hero ($24.99, older)	08-29	14	$399.16	$47.36	14.3%	110	$95.48	28
Lich fragments ($9.99, older)	08-26	10	$109.57	$12.37	0.0%	256	$108.77	0
Lord of Hell fragments ($9.99, older)	08-29	10	$105.98	$12.57	0.0%	360	$134.19	0

"Tapped (logged)" undercounts. purchase_click isn't fired on every path (the Nun pack sells from its own popup), so tap→buy rates are not meaningful. "Was buyer's first purchase" = no earlier Play-billed purchase on that device.

Game-wide, by week

Week (Mon)	Player-days	ARPDAU (IAP + ads)	ARPDAU excl. new hero packs	Payers per 1,000 DAU-days	New hero packs $	Share of IAP
08-31	6190	$0.24	$0.24	10.7	$0	0.0%
09-07	6264	$0.21	$0.21	9.1	$0	0.0%
09-14	5040	$0.25	$0.25	9.9	$0	0.0%
09-21	4395	$0.21	$0.18	10.7	$156	19.6%
09-28 ¹	3472	$0.18	$0.18	7.8	$33	6.8%
10-05	3346	$0.38	$0.35	12.3	$105	9.7%
10-12	2640	$0.23	$0.22	11.7	$32	6.5%
10-19 ²	2879	$0.21	$0.20	11.8	$38	7.4%

¹ includes the 09-28 export gap. ² includes the 10-18 … 10-20 gap. Launches: Nun 09-23 (v32+), Soulstone 10-05, Hero's Growth / Hero + Toys 10-11 (v36).

Did buyers spend more overall, or just on the hero pack?

23 buyers whose first new-hero purchase was on or before 10-12: other IAP in the 14 days before = $38.74 per buyer, 14 days after = $18.11, plus $13.70 on the hero packs themselves, so $31.81 in total after vs $38.74 before. Their total spend did not go up. The hero pack took part of what they were already spending, which matches the flat game-wide numbers.

Why. Hero packs sit behind deep gates (Nun: mine depth 40; Soulstone: stage 115; Hero + Toys: stage 251), so only ~5–10% of DAU can see them, and those players already pay. A hero pack can only grow the payer base if a new or mid-game player can see it. The Divine Cat pack ($19.99, Leo) is ungated but has sold 2. Its price is the likely problem (Selective Hero at $24.99 sold 14, mostly on day 0–3).

4. How many heroes do players get?

One, for most players: 68.7% of players who reach stage 31 own only the starter (Pom); payers typically own 2 (Pom + Knight). The client does not log hero acquisition. These counts are rebuilt from the signals that do exist (below), so read them as estimates. The four craftable heroes are counted when ≥ 50 fragments of a faction were earned. The 7-day-login hero (Sai) is counted when the day-6 claim was followed by a later visit.

Population	Players	Mean heroes	Median	1 (Pom only)	2	3	4	5	6+
All installs (08-25 … 10-12)	10463	1.10	1	92.2%	6.6%	0.6%	0.1%	0.5%	0.0%
Reached stage 31	1635	1.45	1	68.7%	24.4%	3.1%	0.3%	3.3%	0.1%
Reached stage 101	573	1.80	1	58.1%	24.3%	7.3%	0.9%	9.1%	0.3%
Payers	241	2.31	2	0.0%	75.9%	19.9%	1.7%	2.1%	0.4%

Where those heroes come from

Source	All installs (08-25 … 10-12)	Reached stage 31	Reached stage 101	Payers
Volt (5 rewarded ads, stage 10)	4.9%	18.7%	14.7%	16.2%
Sai (7-day login, day 7)	0.6%	2.6%	3.8%	1.7%
Knight (first purchase, any product)	2.3%	9.4%	20.2%	100.0%
Leo (Divine Cat pack, or 10,000★ in Lucky Machine)	0.1%	0.5%	1.4%	0.8%
The Nun (Lord of Hell pack)	0.1%	0.7%	1.9%	4.6%
Candle (SS Hero + Toys pack)	0.0%	0.2%	0.5%	1.2%
Selective Hero pack (pick one)	0.1%	0.4%	1.0%	3.7%
Crafted from 50 faction fragments	0.6%	3.5%	9.8%	1.7%

Population: installs 08-25 … 10-12 (Volt's ad grant is logged from v22 / 08-23; ≥ 2 weeks of runway). Everyone has Pom (starter). Free routes add Volt (5 rewarded ads, from stage 10) and Sai (7-day login). Any purchase adds Knight.
The 7 rank-7 (SS) heroes (Leo, Alahu, Poshedon, Skeleton King, The Nun, Candle, Satan) come from packs, the Selective Hero pool, or pass fragments. Each is owned by < 2% of players.
Instrumentation gap: add earn with currency = HERO_<id> (and the source as position) when a hero is obtained. Hero ownership is the most direct measure of whether hero packs and events deliver.

5. What we suggest

Re-run Lucky Machine as an experiment: one ≥ 90% build, defects fixed, 20% remote-config holdout, prizes and grants logged. Decide on more early events from its result.
Keep the event calendar on the paying group: Dice Adventure, then Fishing (gate 101), each with the same holdout. That is where the measurable money is.
Fix the early wall before adding early content. Stage 34 (v36/v37) and the stage-24 D1 cliff decide early retention more than any event can.
Hero monetisation: bring a hero offer forward. The Nun at $9.99 proves the appetite; it is just gated deep. Test a mid-game SS hero offer (stage 31–65, $4.99–9.99, 48 h) and watch first-purchase rate. That is the one lever here that could grow payers, not just ARPPU.
Price-test Divine Cat (Leo) at $9.99 (it sold 2 at $19.99), and stop handing Leo out free at 10,000★ in the same period.
Give every player a second hero early. 68.7% of stage-31 players still have only Pom; Volt needs 5 opt-in ads, and only ~9% of DAU watch any ad. Granting Volt (or a pick) from a main quest around stage 10–20 opens the roster loop that Hero's Growth, fragments and hero packs sell into.
Log hero acquisition (earn HERO_<id>), so the next report counts heroes instead of estimating them.

Method and data

Source: GA4 export, non-China, 2026-07-20 … 10-26. SQL dashboard/lm_hero_sql/user_days.sql (3.7 GiB) and store.sql. Builder dashboard/build_lm_hero_report.py.
Stage cleared = level_win(N) − 1. Lucky Machine gate = campaign stage 31, builds 28/29/32/33/34, 09-16 … 09-25. Rewards and gates from the v37 client data (2-EventSlotMachineData, HeroData, SevenDaysLoginData, pack assets).
Revenue = Play-billed in_app_purchase gross (incl. VAT) + MAX ad value. Export gap days (09-28, 10-18 … 10-20) excluded from per-day rates.
Build-based control groups were tried and rejected. Players who stayed on old builds during the event were already leaving (2 active days of 10, 25% returned), so comparing them with updaters measures who updates, not the event.