Performance

What is actually measured

Reporting play time honestly

• Report at session end or on a timer, not every second; each report is a row.
• Report wall-clock play, not a guess, because the average is what your listing's engagement figure becomes.
• Do not report for a tab you cannot see; a hidden page inflates the number and means nothing.
• Accept that a player can post the same call themselves; the metric is not tamper-proof.

POST /api/listing-play
{ "collection": "games", "listingId": "my-game", "playSeconds": 340 }

// playSeconds is rounded and clamped at 0
// a value of 0 writes nothing but still answers 200 { ok: true }

A fast web build

1. Ship the fewest bytes that run. The whole tree sits behind one URL with no bundler between you and the player.
2. Compress large payloads. The presign accepts a contentEncoding so a .data or .wasm blob can be served marked as encoded.
3. Get content types right. .wasm must be application/wasm or streaming compilation fails; that is inferred from the extension for you.
4. Split assets from code. A 300 MB initial download is a review, whatever the game is worth.
5. Keep one file under 5 GiB. The per-object ceiling, enforced twice, see Upload limits and file types.
6. Version the folder, not the file names. A version segment gives you a clean rollback target.

Downloads and caching

The desktop client

The Windows client is a browsing shell around the store with a local library, so its performance characteristics are mostly disk and process management rather than rendering.

• Games install under %LOCALAPPDATA%\Appsinity\library\<slug>, one directory per title; a large install is disk, not memory.
• The library manifest and play log are bounded lists, so the client does not grow without limit as you install.
• The performance overlay (CPU, GPU, memory) is a local HUD, laid out from the store's In-game overlay row and toggled with Ctrl+Shift+F12; it reports nothing to Appsinity.
• Close sends the client to the tray rather than quitting, so a background instance holds memory until you quit from the tray.
• GPU utilisation comes from one shared Windows performance stream, which is why the HUD updates together rather than per metric.