Over 2 Years of Sales Data From a $2.99 Premium Mobile Game

01 · TL;DR

The numbers first

1. Full paid-game totals
2. Launch press and Taiwan outreach
3. Country mix
4. iOS discount windows
5. Android discount windows
6. What retention can tell us
7. Detailed source-data sections
8. Every monthly record
9. All 15 promotions
10. Store settlement records
11. Limits of the premium model
12. Current iOS experiment
13. Methods and limitations

02 · Context matters

Price changes did not happen in a vacuum

What happened

Data

The selected before-and-after sales average was 4.97 times higher. This is a descriptive comparison, not an estimate of the revenue caused by TouchArcade. The release itself and several articles occurred in the same period, and no referral-to-purchase data exists.

What I learned

03 · Country mix

The same nominal discount reached different markets

What happened

The fall event was dominated by South Korea. The first 30 days after the June 2026 reference date were dominated by Taiwan, where direct outreach and local coverage occurred at the same time.

Data

What I learned

Country mix changes the proceeds per sale through local prices, taxes, exchange rates, and platform exposure. It also reveals that these were not repeatable, market-neutral price tests. I cannot use either window to calculate a universal price elasticity or the number of sales required to reach a chart position.

04 · Store records

Results since launch, without combining incompatible currencies

Data

*The iOS total is the sum of observed purchase rows and includes dates with no available row; missing dates were not filled with zero. Android counts transaction rows, not unique buyers. A purchase-minus-refund count is neither a unique-buyer count nor a final cohort refund rate. The two platforms use different currencies, cut-offs, time zones, and accounting definitions, so their amounts are intentionally not added together.

Apple financial reports, converted at the fixed reference rate of KRW 1,380 per USD, total approximately $15,137.68 in net customer sales and $12,528.99 in developer proceeds for accounting periods from June 30, 2024 through August 29, 2026. Those conversions are context, not actual bank receipts. A manually entered Steam gross figure was excluded because the settlement export and period allocation were not verified.

What I learned

Units, gross customer payments, estimated analytics proceeds, financial statements, and bank receipts answer different questions. Treating them as one “revenue” number hid important differences, so the audited version keeps each source and currency separate.

05 · iOS case studies

Big unit gains, smaller and short-lived proceeds gains

What happened

Fall 2025 event

The event concentrated purchases, but the comparison windows each have two missing purchase-report days. The event's 261 net units came from 268 purchases minus seven refund units. Refunds are assigned to their report event date and do not necessarily belong to purchases made in the same window.

June 2026 reference-date comparison

What I learned

As a simple illustration, cutting a US list price from $2.99 to $0.99 requires about 3.02× as many sales to hold revenue constant if taxes, commission rate, refunds, and country mix are all unchanged. In reality, those assumptions did not hold.

06 · Android case study

The answer changed with the comparison window

Data

Over 60 days, charge rows increased 20.5% and developer proceeds increased 9.2%. Over 30 days, charge rows fell 1.6% while proceeds increased 3.3%. The pre-period was not a clean full-price control: US transactions included $2.99, $0.49, and $0.99 prices.

Across all 15 recorded Google Play promotions

What I learned

The choice of 30 or 60 days changed the answer, and the earlier periods already contained promotions. This is useful operational evidence, but it is not a clean full-price control versus a discounted treatment.

07 · Retention

Acquisition grew. Exact-day return rates were mixed.

Data

The comparison uses two 28-day first-visit cohorts: May 19–June 15 and June 16–July 13, 2026, in UTC+9. Both had completed their D30 observation window.

On iOS, D1 and D7 rose while D30 fell; the larger cohort still produced 29 exact-day D30 returners versus 10. On Android, D1 fell, D7 rose, and D30 fell slightly. D30 counts were only 10–29 users per group, and the 95% Wilson intervals overlap.

What I learned

08 · Limits of the premium model

Lower prices coincided with more discovery, but did not solve the long-term problem

What happened

Dungeon Tracer launched at $2.99 and later moved to $0.99. Short promotions and media exposure produced strong bursts of paid acquisition, but the bursts faded. The first 30 days after the June 2026 reference date produced 1,031 iOS net units; the next two 30-day windows produced 120 and 114 observed net units.

Data

• At unchanged taxes, fees, refunds, and country mix, a move from $2.99 to $0.99 needs about 3.02 times as many sales to preserve revenue. The real country mix changed, so this is only an illustration.
• All 15 recorded Google Play promotions increased charge rows per day, but two reduced developer proceeds per day.
• The strongest acquisition periods overlapped with press, regional outreach, localization, launch interest, or sale-tracking exposure. Those conditions were not always repeatable.
• The available data cannot identify whether the lower price itself improved retention.

What I learned

The historical pattern is consistent with the upfront payment being one source of friction, but it does not isolate price from press and store exposure. Lower prices coincided with bursts of demand, while proceeds per sale fell and discovery did not stay at the initial peak. The paid model was still generating revenue; the question was whether a free entry point plus one optional purchase could fund continued updates more consistently. That question required a separate experiment rather than another before-and-after comparison.

09 · What I am testing now

One month of free iOS access with one optional $9.99 purchase

What changed

After operating for more than two years without in-app purchases, I changed the iOS app from paid to free for one month and introduced the game's first IAP. Android remains a paid app.

What the experiment is designed to measure

1. How much new-player acquisition changes when the upfront price is removed.
2. Whether enough free players choose the optional IAP to support the cost of ongoing updates.

The experiment tracks App Store first downloads, GA4 new users and mature activity cohorts, newly created server identities, verified first class unlocks, refunds, and developer proceeds as separate measures. Downloads, players, server identities, and buyers are not treated as the same population.

Why iOS first

Google Play allows a paid app to become free, but does not allow that same app to return from free to paid. I kept Android paid so that the test would remain reversible. iOS is therefore the first platform used to test the free-entry model.

What I can conclude today

Nothing yet about whether the new model is better. The test needs the full observation window, verified IAP proceeds and refunds, and mature retention cohorts. Early downloads or a few purchases would not be enough to establish sustainable revenue.

10 · Operating lessons

What I would measure and do next time

Launch coverage, direct Taiwanese outreach, short promotions, and a longer-term base-price change created different conditions. My practical takeaway is to create a reason to talk about the game, communicate it to the right local audience, improve the post-purchase experience, and measure proceeds and return behavior at each step. A category #1 or a large sales spike is meaningful, but it is not recurring monthly revenue.

11 · Methods and limitations

How the figures were audited

• For App Store Purchases DAILY files, the latest processed revision for each event date was selected. Weekly, monthly, annual, and older revised files were not double-counted.
• iOS days without a purchase row are marked unknown, not zero. Android days within collected settlement months with no matching transaction are counted as zero.
• Apple Analytics dates (UTC) and Sales Summary dates (US Pacific Time) remain separate. Transaction prices do not prove the exact configured start or end time of a promotion.
• Google transaction timestamps were converted from PST/PDT to UTC+9 before window comparisons. Legacy and current product-identifier columns were both included back to the July 2024 launch.
• Sales, refunds, and proceeds retain their source definitions. Store-estimated analytics, financial statements, and bank receipts are not treated as interchangeable.
• No order IDs, customer identifiers, or individual activity records are published here.

These are observational before-and-after comparisons, not randomized experiments. Trends, updates, press, country mix, other promotions, measurement consent, and delayed corrections can confound the results. I do not claim a causal price effect where the data cannot identify one.

Primary metric references

• Apple: App Store Purchase report definitions
• Apple: data completeness and corrections
• Apple: financial report fields
• Google Play: monthly reports
• Google Analytics: cohort metrics

This public English edition condenses the full Korean audit into a shareable case study while preserving the principal figures, corrections, and caveats. Source snapshots have different cut-off dates; the historical report is a fixed record and may be corrected if store reports are revised. The current experiment is documented separately because it began after the historical report's main data cut-off.