Tiny Knightfall — what each build changed, and what it did (last 90 days)

The short version

v25 is the one clear regression — and it was the build, not the traffic. D1 −2.6pt, D7 −2.3pt and revenue per install about −50% against v24, and all three survive mix adjustment and dropping the top payer. The 17/07 Tier 1 campaign (47% of all payers) did stop two days before v25 shipped, but correcting for traffic mix barely changes the drop — the payment bug is the better explanation. v26.2 fixed it: D7 8.2%, the best of the window.
v34's harder stages cost about 2.5pt of D1 (24.6% vs 27.1% on v29, mix-adjusted). Its higher revenue per install is only partly real: +60% vs v29 with the top payer dropped, about +21% once traffic mix is equalised — the 29/09 Tier 1 ROAS campaign started four days into v34.
v36 (Daily Quest + Battle Pass) did not deliver — and it made the early game harder. Its “campaign rebalance” moved the walls earlier: a new wall at stage 34 (losses 10% → 40%) plus 50–53 and 57–60, while removing the old 54–56 wall. D1 −3.2pt (21.4% mix-adjusted, the lowest of the window, on 1,243 installs); revenue per player-day only +3%. v37 made stage 34 harder still (45% losses, clear time doubled). v36's D7 and first-week revenue cannot be judged yet: the export gap leaves only 32 v36 installs with seven observed days.
v37 is v36 pushed to everyone — same features, but stage 34 got harder again. Its D1 is +2.9pt raw but only +0.9pt mix-adjusted (the rest is the install wave from 10-21), and fewer of its players get past stage 34 (12% → 8% within 3 days, early). Treat ARPU swings under ±40% and D1 swings under ±1pt as noise: v19 and v33 changed nothing visible and moved that much.

Headline numbers: all countries and campaigns; “mix-adjusted” and “drop top payer” are the controls above. Check any of them per country or campaign with segment A / B.

What each feature did the dev's changelog (23/10), checked against the data

Effects are mix-adjusted with the top payer dropped, against the previous build, for all countries and campaigns. ARPU swings under ±40% are within the noise floor. The dev numbers builds 30–36 differently from what devices report (dev 30–31 ≈ v32, dev 35 ≈ v36 by date), so those rows are matched by date — worth confirming with the dev.

Build	Dev says it shipped	Seen in the data	Effect	Verdict
v17	Campaign game mode; toy tutorial fix	Tiny Guide pass and 10-day ad pack start selling	ARPU +70%, D1 ±0	Can't tell v16 is a thin baseline and the ROAS campaigns were ramping the same week.
v18–19	Research + Astrology finished; new package shop	Gem research-skip spending (v18)	ARPU −31% / −29%, D1 +1.9pt then flat	Nothing clear Swings sit inside the noise floor.
v20	Sugar World mode + packs; main quest update	4 new packs sell	ARPU +54%, D1 −0.5pt	Probably helped revenue Modest, close to the noise floor.
v21	Astrology completed	No new events (an ads → hero switch on 08-21 is not in the dev log)	ARPU +3%, D1 −0.3pt	Nothing
v22	Hero system + Alchemy (big batch), Treasure UI	Alchemy / Astrology / Toy live; Treasure unused until v25	D1 +2.2pt, ARPU −73%	Engagement up, revenue down 774 installs and a 3-day life, but the drop survives every control. Worth asking whether Alchemy hands out heroes that packs used to sell.
v24	New pet; limited Astrology pack	Pets are not tracked	ARPU back to v21's level, D1 −1.6pt	Reversed v22's revenue drop The pet itself can't be measured.
v25	Pets, Loot (Village), new IAP packs — and the payment bug	Village live	D1 −2.5pt, D7 −2.3pt, ARPU −51%; stage-31 reach 9% vs 16%	Hurt The build, not the traffic — it survives mix adjustment.
v26.2	Pack/event update; 3–10-day ad packs hidden; fixes	Short ad packs gone	D1 +4.5pt, D7 +4.9pt (8.2%, best of the window), ARPU +41%	Helped Mostly undoing v25.
v28	FlyBox cycle (120s with remove-ads, 60s without); Lucky Machine	Fly-box reward reach 3.1% → 1.1%; Lucky live	D7 −3.7pt, ARPU +44% on 182 installs	Can't tell — too small The dev's two files disagree on whether FlyBox was v26 or v28; the data points to v28.
v29	Diary-quest progress fix; tutorial fix	Diary-quest reach +3.6pt; no Hope-pack buyers	ARPU −28%, D1 −0.6pt	Nothing measurable
v32 (dev v30–31)	Battle Pass; hero soul + The Nun packs; crash reporting — and No-Ads sale packs removed	Crash logging ✓, sale packs gone ✓; no Battle Pass or Nun/soul activity until v36	ARPU −43% on 241 installs, D1 −0.3pt	Battle Pass shipped dark The sale-pack removal is covered by the No-Ads report.
v33	— no dev entry —	Nothing new	noise	Nothing Use it as the noise floor.
v34	Account deletion; key-3 pack; 2 talisman packs	No talisman buyers. Difficulty raised — stage 54–56 losses 40% (was ~25%), 77–100 doubled — not in the dev log	D1 −2.5pt vs v29; ARPU +21% vs v29 (mix-adjusted)	Difficulty hurt D1 Ask the dev what changed in game config.
v36 (dev v35)	Campaign rebalance; Battle Pass 3 + Daily Quest; hero growth pack; 8 new IAPs; x2 speed only after a rewarded ad; recommend popup on first boss loss	Battle Pass + Daily Quest live; new difficulty wall at stage 34 (losses 10% → 40%) and at 50–53, 57–60 — the old 54–56 wall removed; x2-speed ad use unchanged (2.2% → 2.4%)	D1 −3.2pt (21.4%), ARPDAU +3%; D7 / ARPU on only 32 mature installs	Hurt retention The early lift in spend per player faded with more days. The rebalance moved the difficulty earlier, to stage 34, where far more players arrive than at 54–56 — the likeliest reason D1 fell, alongside the new early-game gates (x2 speed behind an ad, the popup).
v37	Version bump; forced update (Remote Config should-update → 36, force-update → 33)	No new or dropped events vs v36, but stage 34 harder still (45% losses, clear time 30s → 59s); 65% of daily players by 10-25	D1 +2.9pt raw, +0.9pt mix-adjusted; reach to stage 34 in 3 days 12% → 8%; D7 / ARPU not mature yet	Same features, harder stage 34 Not a clean control after all: the raw D1 rise is the 10-21 install wave, and fewer players get past stage 34 — on 203 installs, so early.

Things that never shipped to players or can't be seen: pets (not tracked), talisman / The Nun / hero-soul / Hope packs (no purchases), account deletion (no event), and the Battle Pass until v36. v37 has v36's features but an even harder stage 34. Something the dev log misses: v34's difficulty increase.

The timeline hover anything · click a build column to open it

Drag the timeline (or scroll sideways) to move along it.

Every update one card per build — its goal, what changed, and whether the numbers moved the way the goal needed (against the build before it, beyond the noise floor) · drag or use the arrows

Since v20: did the new features raise LTV? no — and the data says why

Between v20 (3 Aug) and v37 the game added Sugar World, finished Astrology, then Heroes, Alchemy, Toys, Treasure, pets, the Loot/Village loop, the Lucky Machine event, Battle Pass (three versions), Daily Quest, a campaign rebalance and 30+ new IAP packs. First-week revenue per install did not go up once: v20's $0.70 is still the best D7 LTV of the window, and later builds sit between $0.16 and $0.68 (top payer dropped). v34's $0.68 is mostly the 29/09 Tier 1 campaign — about +21% over v29 once traffic mix is equalised.

Why not

Fewer players pay; payers do not spend less. The D7 payer rate fell from 3.0–3.3% (v20–21) to 1.3–2.5% afterwards, while the median first-week spend per payer stayed at $10–14. The lost LTV is lost payers.
Part of that is traffic. The 17/07 Tier 1 ROAS campaign (47% of all payers) fed v17–v24 and stopped on 09-03. Equalising the traffic mix closes some of the gap, but builds after v24 are still ~25–30% below v20–v24.
The new content sits where few players arrive. Within 5 days only 15–19% of installs reach stage 31 and 3–7% reach stage 100 — and that share has not grown. Alchemy, Loot Storage and the campaign packs unlock deeper still (stage 115 / 200 per the roadmap), so 9 in 10 installs never see most of what shipped. Meanwhile D1 sits at 21–28%: the game loses three quarters of its players before the new systems can matter.
Many packs barely sell, or never reached players. Talisman, The Nun, hero-soul and Hope packs show no purchases; the hero growth pack is clicked by 0.4% and the premium pass bought by 0.4%. The Battle Pass sat in the build for three weeks before anyone could use it (dev v30 → device v36).
Ads grew, IAP shrank. Ad revenue per install rose 2.7× (8% → 24% of first-week revenue), helped by removing the cheap No-Ads offers — but it does not make up the IAP lost to falling conversion.
Difficulty kept moving the wrong way. v25's payment bug (D7 LTV $0.24) aside, v34 raised the walls (D1 −2.5pt), and v36's “rebalance” moved them earlier — a new wall at stage 34, where far more players arrive — which v37 made harder again. Each step trims the share of installs who ever reach the content built for them.

What this points to: work on the D0–D1 funnel and first-purchase conversion in the first ~30 stages, where every install actually is — not more late-game systems and packs.

LTV per install, first 7 days all traffic · raw · top payer dropped

D7 LTV per build (label under it: D7 payer rate)v20 level

Every number as a table

How to read this — definitions and caveats

Builds are the 17 releases that reached ≥30% of daily players (v37 is still rolling out). Rollout = first day seen → 20% → 50% of daily players. Stragglers and test builds (27.2, 30, 31, 35, 350) are left out of the columns. Device v35 is one of those test builds (12 players, 10-04 → 10-09): the content the dev calls “v35” reached players as device v36.
Install cohort = players whose first session was on that build. Retention and ARPU are cohort metrics, so a build is judged on the new players it received — not on veterans who updated into it (they are more engaged and would flatter newer builds).
ARPU D7 = revenue in the first 8 days (install day + 7) ÷ installs old enough to have had them. D1 / D7 = share of installs active exactly 1 / 7 days after install (the cards and table also show every day D3–D7, for retention and for LTV = cumulative revenue per install by that day). ARPDAU = revenue ÷ player-days, by the build actually running that day — the one metric that includes veterans.
Mix-adjusted re-weights each build's cohort to the same traffic mix — geo tier (US / other Tier 1 / rest of world) × traffic type (ROAS Tier 1 campaigns / other paid / organic / unattributed), using the window average of the selected segment and only groups with ≥20 installs. When raw and adjusted disagree, the movement came from who was installing, not from the build. ARPDAU is never mix-adjusted or whale-trimmed: it covers every player on the build, veterans included.
Drop top payer removes the single biggest IAP spender from each build's cohort. Revenue at this scale is whale-dominated; a change that disappears with one player removed is not a build effect.
Faded values have too few installs to trust (<200 for ARPU, or a retention change that is not statistically distinguishable from zero at 95%). Blank = not enough mature installs yet.
Campaign comes from AppsFlyer: about 45% of installs match a paid campaign; the rest are Organic (matched to AppsFlyer's organic list) or Unattributed. Filtering to one campaign samples it — it never covers it fully.
Overlap warnings fire when a campaign starts or stops, or a store / live-event change happens, between 3 days before and 10 days after a build's first day.
Android only (the iOS app is not in the GA4 export). China excluded. Revenue is gross (no refund events), and IAP tracking began 2026-07-20. The window ends 10-25. The export is missing or nearly empty on 09-28 and 10-18 → 10-20 (grey bands): those days are unobserved — installs on them are dropped, no retention day lands on them, and players who installed before 10-18 are followed only to 10-17. That is why v36 has fewer mature installs, and no mix-adjusted D7/ARPU (too few per traffic group).
Changelog: the dev's changelog_android.txt (23/10) merged with what the event data shows, checked against the roadmap, sync sheet and the 09-12 dev meeting. Tags: dev dev changelog · data seen in events · roadmap roadmap, confirmed by data · prior earlier report · unknown nothing visible — ask the dev.