How the game list is scored

Score

The default order uses one score. The weights in the ranking code are the numbers in the table below. Featured, embed and cover art add 0 points.

score = 50 × popularity + 40 × experience + 10 × freshness

X engagement was captured on 2026-10-21. Play-test checkedAt values cover 2026-10-21–2026-10-21 (file date 2026-10-21).

Three games, worked out

Each line substitutes into the formula. The score is the sum of the three weighted parts, rounded to two decimals. The ×50, ×40 and ×10 columns show those same unrounded products to two decimals.

1. 3rd Mainland Biker, rank 1. popularity 0.82555567 × 50 = 41.27778333; experience 0.825 × 40 = 33; freshness 0.93303299 × 10 = 9.33032992. Unrounded sum 83.60811324, score 83.61.
2. BeetleJump, rank 22. popularity 0.25 × 50 = 12.5; experience 0.95 × 40 = 38; freshness 0.93303299 × 10 = 9.33032992. Unrounded sum 59.83032992, score 59.83.
3. Blueline, rank 2. popularity 0.84640291 × 50 = 42.32014544; experience 0.69375 × 40 = 27.75; freshness 0.93303299 × 10 = 9.33032992. Unrounded sum 79.40047536, score 79.40.

Where the points come from

Each bar is one weighted term summed across all 68 games. It shows how the weights land on this list.

The next chart is the top 12. Bar length is out of 100: popularity, experience and freshness sit end to end, and the empty track is the gap to a perfect score.

• popularity × 50
• experience × 40
• freshness × 10

• 13rd Mainland Biker83.61
• 2Blueline79.40
• 3fly.pieter.com77.64
• 4INKWAVE76.02
• 5Last Train to the Sea74.82
• 6Tiny Skies74.57
• 7Riposte74.46
• 8Bruno Simon73.02
• 9Playable Worlds72.71
• 10R.M.S. Titanic 191270.84
• 11Outermate69.66
• 12HELL//BREACH67.94

Score distribution

Full ranking

68 games, highest score first. Equal scores sort by slug, so English and Chinese stay in the same order and a reload does too. /games uses this order when the search box is empty.

Rank 1 is 3rd Mainland Biker at 83.61. Popularity 0.8256, experience 0.8250, freshness 0.9330.

Popularity

popularity = max(0.25, min(1, log10(1 + likes + 2 × reposts + views / 100) / 4))

A game with no recorded post, or a post whose formula result is under 0.25, stays at 0.25. The cap is 1, so one very large post contributes at most 50 points. The log spreads orders of magnitude and stops a single viral post at that cap.

Most-engaged public X post found linking to each game (X API search, top 25 relevant posts per game). Numbers are copied from X at capture time.

• Numbers were copied from X on 2026-10-21. Likes, reposts and views in the table match the JSON.
• One post per game: the most-engaged public X post found linking to the game (X API search, top 25 relevant results). The post URL is in the table.
• Refresh the numbers when the list is refreshed. A signal whose post is gone is dropped, and that game returns to 0.25.

53 of 68 listed games have a post. The signal file has 53 rows, and 0 of those slugs have no current listing.

32 games on the popularity floor

15 have no recorded post. 17 have a post, and reach = likes + 2×reposts + views/100 still gives log10(1+reach)/4 ≤ 0.25.

Experience

experience = 0.2 × load + 0.15 × mobile + 0.15 × noWall + 0.15 × errors + 0.2 × depth + 0.15 × polish

Each game URL opened in headless Chromium at 1280x800 (desktop) and as iPhone 13 (touch), waited for the load event plus 7 s, screenshot and console errors captured; depth and polish judged from the screens.

68 listed games, 68 rubric rows. Rubric slugs with no current listing: 0. A listed game with no row fails the build, so every game below has one.

loadMs

Recorded 67, missing 1. Min 274 ms, median 1183 ms, max 20008 ms.

Phone load event (mobileLoadMs, not in the score): Recorded 68, missing 0. Min 120 ms, median 981 ms, max 26389 ms.

Games whose desktop loadMs is null (that input scores 0): Contexto

mobile, noWall, depth, polish

The mobile chart uses the value stored in the rubric. 2 games have webgpu true, so the mobile term that enters the formula is half the stored value.

Console errors

42 games loaded with 0 console errors (errors = 1). 26 loaded with more than 0 (errors = 0.5). The highest count on one game is 242.

The table shows the 8 highest console-error counts.

Play-test inputs

Freshness

freshness = max(0.25, 0.5 ^ (ageDays / 30))

Age runs from the curation date to 2026-10-24 (Asia/Shanghai). The curation date is added, or, when added is absent, the play date written as 试玩于 YYYY年M月D日. An undated listing stays at 0.25. The value halves every 30 days and then stays at 0.25. Age is constant within a day, so the order can move once a day.

15 games on the freshness floor

Orders this score leaves alone

• Ties break by slug, ascending.
• Search results at /games?q= keep text-match order.
• The Oldest curation and Name sorts on category pages and the home list stay as they are.
• Categories that also hold products keep Newest curation order. The default sort uses this score only when every entry on the page is a game.

Notes that add 0 points

otherSignals has 3 notes. They stay in the file so the source can be checked, and they enter neither popularity, experience nor freshness.

How many games sit in each category

Counts come from the category of each listed game, the same records as the ranking above.

Limits

• X numbers are one post captured on 2026-10-21. Likes and views on that post can change later; the page keeps this capture until the JSON is copied again.
• Search keeps the most-engaged public post among the top 25 relevant results at capture time. A later search can find a different post.
• depth and polish are judgments from the screens. Re-running the check and updating checkedAt can change both.
• The popularity floor places games with no post and games with a tiny post on the same 0.25.
• Any console-error count above 0 sets errors to 0.5. A game with 27 errors and a game with 1 receive the same errors term.
• WebGPU games have mobile halved before weighting. The distribution chart shows the stored value from before that halving.
• Freshness uses the Asia/Shanghai calendar day 2026-10-24. While that city is still on this date, the order stays put even if the UTC date differs.
• Numbers on this page are computed from the JSON and the formula at render time. The page does not keep a second copy of the scores.

Questions

How is the game score calculated?

score = 50 × popularity + 40 × experience + 10 × freshness. This page scores 68 listed games with that formula. On 2026-10-24 (Asia/Shanghai) the highest score is 83.61 (3rd Mainland Biker).

Which games sit on the popularity floor?

32 of 68 games have popularity 0.25. 15 have no recorded X post. 17 have a post whose log-scaled reach still lands on the floor.

Do jam placements add points?

Jam placements and other non-numeric notes live in otherSignals and add 0 points. The file currently has 3 notes. Featured, embed and cover art also add 0 points.

When does the order change?

Ties break by slug, so the English and Chinese lists match. Freshness uses the Asia/Shanghai date 2026-10-24, so the order can move once a day. X counts stay at the 2026-10-21 capture until the JSON is refreshed. Play-test checkedAt values cover 2026-10-21–2026-10-21.

What if a game has no play-test row?

A listed game with no row in game-experience.json fails the build. This page scores 68 listings against 68 rubric rows.