What should I measure after an update?
Measure stability first, then the first session, retention, returning players and sales. Pick the metrics before you ship, so you do not search for a good number afterwards.
Pick one main metric per update, the one that matches its goal, and treat the rest as guard rails. A content update should bring players back without hurting crash-free sessions. A tutorial rework should lift the first session funnel without lowering D7.
| Goal of the update | Metric | Where to read it |
|---|---|---|
| Fix crashes | Crash-free sessions and errors per session by build | Game health board |
| Better onboarding | Tutorial and level 1 to 4 funnel | Funnel chart filtered by build |
| Keep players longer | D1 and D7 of players who installed on the new build | Game KPIs board with an App filter |
| Bring players back | Resurrected players and returning DAU | Users page lifecycle filter |
| Sell more | Units sold and returned per day | Steam sales sync (Early access) |
Sources: Manalens metrics, error tracking and Steam documentation.
How do I tag events with a build?
Set the app version and build when you start the SDK. Every event then carries them, and the App filter group can split any report by version, build, platform or storefront. Through the API, the filter is app_build.
Use a new build value for every release, including hotfixes. If the Steam and itch.io builds differ, set the storefront too. A browser build made with the web SDK can carry the same app fields through its app option, so the build filters work there as well.
GDScript
func _ready() -> void:
Manalens.init({
"publishable_key": "YOUR_PUBLISHABLE_KEY",
"endpoint": "https://YOUR_COLLECTOR_ENDPOINT",
"app_version": "1.5.0",
"build": "602",
"storefront": "steam",
"enabled": not OS.has_feature("editor"),
})Sources: Manalens Godot SDK and app events documentation.
How do I compare before and after fairly?
Compare cohorts, not calendar weeks. Take players who installed on the old build in the two weeks before the update, and players who installed on the new build in the two weeks after. Give each group the same number of days to return. Then compare D1 and D7.
Watch for things that change at the same time. A sale, a festival or a creator video brings different players, and they may retain differently. Note every marketing beat next to the release date. Wait for a few hundred new players per cohort, and test the gap with a significance calculator before you call it a win.
Sources: Manalens retention documentation; Manalens A/B test significance calculator.
How do I separate an update from a sale?
Split new players from returning players. Public numbers cannot do this. Dome Keeper went from about 200 to 5,274 concurrent players after its Double Dome update, which shipped with a 60% discount, so the public count cannot say which part came from each. In your own data, new installs show the sale, and owners who return on the new build show the update.
Valve shows sale events mainly to prospective buyers and update events to owners in the library, which supports the same split.
Sources: Game World Observer (10 October 2023); Steamworks Events and Announcements Visibility documentation.
Which live ops events should I track?
Track when a timed event starts for a player, whether they take part, and what they earn or spend. Use one event name per action and put the event ID in a property, so one chart covers every event you run.
For economy changes, send resource events with flow set to source or sink. The Game KPIs board sums them per currency, so you can see if an event floods the game with gold.
GDScript
Manalens.track("live_event_join", {"live_event": "winter_2026", "day": 1})
Manalens.track("live_event_reward", {"live_event": "winter_2026", "reward": "sled_skin"})
Manalens.track("resource", {"flow": "source", "currency": "gold", "amount": 200, "item": "winter_chest"})Sources: Manalens metrics and Godot SDK documentation.
How do I catch a bad update fast?
Set alerts before you ship. An alert counts one named event in each five minute window and can filter by app version. Use "fewer than" on session_start to learn when the new build stops sending events, and "at least" on app_unclean_exit to catch a crash spike.
Then check the Game health board, which shows errors per session by build and the top errors with the latest build.
Alerts go to Slack or to a signed HTTPS webhook. A fewer than alert starts with the first full window after you create it, so create it before the release, and test it with a short window on a test build.
Sources: Manalens alerts and error tracking documentation.
How do I measure a 1.0 launch after Early Access?
Treat 1.0 as a second launch, with its own cohort and its own baseline. The Planet Crafter sold 32% of its units in the two months after 1.0, and its 1.0 numbers tracked more than 50% ahead of the 0.7x median that GameDiscoverCo usually sees. It also added online co-op at 1.0.
Abiotic Factor, also covered by GameDiscoverCo, chose large, chapter style updates during Early Access instead of weekly or monthly ones. Its team said that timing store bundles with a major content update was key for traffic.
For your game, compare 1.0 players with Early Access players on D1, D7 and the share who reach the new content. Track the new content with its own events, so you know whether people play it.
Sources: GameDiscoverCo (24 June 2024; 2 September 2025).
How do I write an update review?
Write one short note two weeks after each update, with the same headings each time. Over a year, these notes show which kinds of update work for your game.
- State the goal of the update and the metric you picked before release.
- Record crash-free sessions for the new build against the old one.
- Record D1 and D7 for new players on each build, with the cohort sizes.
- Record Resurrected players and units sold in the two weeks before and after.
- List anything else that happened in the same weeks, such as a sale or a festival.
- Write one lesson for the next update.
Sources: Manalens retention, error tracking and Steam documentation.
How do I see what my last update changed?
Manalens is the analytics and growth engine built for games. Set the build in the SDK, then compare crash-free sessions, retention and returning players for the old and new builds on one dashboard.
Sources: Manalens product documentation.
Frequently asked questions
How long should I wait before I judge an update?
At least seven days after enough new players arrive, so D7 has data. Stability shows in hours, the first session in a day or two, and retention in a week or more. Judge each metric on its own clock, and do not judge D7 on day 3.
Can a free update increase sales?
It can, especially next to a discount. The Planet Crafter doubled its peak concurrent players after 1.0, and Dome Keeper jumped from about 200 to 5,274 players after a free update with a 60% discount. These are single cases, so measure your own update against a baseline.
Should I ship many small updates or a few large ones?
Both can work, but Steam rewards large ones with more visibility. Major Updates can reach the library and the store homepage, while patch notes stay out of high visibility places. Ship fixes as patches when you need them, and group new content into fewer, larger updates that deserve a Major Update post.
What if most players never update the game?
Then compare players by the build they actually run, not by the date. Every event carries the build, so filter each report to one build. If an old build stays common, check the platform and storefront filters to find where players are stuck.
Sources and implementation
- GameDiscoverCo: How The Planet Crafter hit 1 million sold, post-1.0 (Simon Carless, 24 June 2024) ↗
- GameDiscoverCo: How Abiotic Factor rode a hot Steam 1.0 release to 1.4m sold (Simon Carless, 2 September 2025) ↗
- Game World Observer: Dome Keeper sees 2,500% surge in concurrent players after latest update (Evgeny Obedkov, 10 October 2023) ↗
- Steamworks documentation: Events and Announcements Visibility (Valve, read 18 November 2026) ↗
- Steamworks documentation: Event Type: Major Update (Valve, read 18 November 2026) ↗