The ranking changed. Did the population change too?

The arithmetic behind the almost-flat share

The baseline contains 1,057 games, with disclosure found for 330. Today's slice contains 1,021, with disclosure found for 320 and no disclosure found for 701. Those fractions are 31.22% and 31.34%: a difference of about 0.12 percentage points. They are selected-slice descriptions, not estimates of AI use across the whole festival.

Among the 895 retained entries, all disclosure labels are unchanged: 270 are disclosure-found and 625 are no-disclosure-found in both snapshots. The outgoing group contains 60 disclosure-found entries among 162 games; the incoming group contains 50 among 126. The numerator therefore goes from 270 + 60 to 270 + 50, while the denominator goes from 895 + 162 to 895 + 126.

That accounts for the change in this observed share without any label changes among the retained games. It does not tell us when an entering game's disclosure was written, whether a developer changed production methods, or why a game moved. Most importantly, the 162 exits are outside today's cutoff, rather than known to have left the festival or lost all players.

The cutoff is a rule, not a permanent club

We keep the selection rule fixed: top 1,000 positions by rolling 24-hour demo peak concurrent players, adult entries included, plus every tie at the boundary. The membership and exact denominator can still change. Today the boundary is displayed rank 951 at a rolling peak of 10 simultaneous players. All 71 games tied there are included, producing 1,021 rows. The baseline had 1,057 rows and yesterday's selection had 1,015.

Competition ranks repeat when values tie. That also means a band such as ranks 101–500 can contain more than 400 games. We report the actual denominator in each band rather than silently trimming tied entries.

The baseline shares in these bands were 26/101 (25.74%), 114/404 (28.22%) and 190/552 (34.42%). Today's denominators and members differ, so these are repeated cross-sections, not the trajectories of three fixed groups. Genre comparisons would require verified tags, adequate denominators and a separate design; we do not infer an AI advantage or penalty from this table.

Keep watching games that leave the table

Our separate 12-game panel remains frozen to the baseline, with four games from each original rank band and equal disclosure-found and no-disclosure-found counts within each band. Today 11 appear in the selected table. We read the twelfth through its native PlayRecon card: Chinese Pancake Simulator is at rank 1,391 with a rolling peak of five simultaneous demo players, rather than zero.

All 12 panel peaks are observed at the same 16:20 UTC provider update. Six are above their baseline rolling peaks and six are below; none is unchanged. This small research panel is neither representative nor a set of causally matched pairs. Its role is to preserve observations of the same games when ranking membership changes. See yesterday's explanation of the fixed-panel approach.

What I would take from this as a developer

When comparing ranking reports, ask both whether the metric changed and whether the set of games changed. Keep an explicit membership reconciliation alongside the aggregate, and follow a stable reference panel outside the cutoff. Here the near-flat disclosure share masks 126 incoming and 162 outgoing entries. That is useful context for reading a chart; it is not a claim that either group succeeded because of AI.

Genre, prior popularity, demo age, marketing, visibility and quality remain potential confounders. Disclosure itself is a broad flag: PlayRecon's With AI filter includes disclosure on the game or its demo. No disclosure found does not mean AI-free. A found disclosure establishes neither quality nor the proportion of development done with AI, and does not by itself qualify a game for AIM's recommendations.

Source clocks and coverage

We read the public native PlayRecon UI in All and With AI modes. Both show the 25 October 2026, 16:20 UTC update. All 1,021 rows were collected at 16:44:20.540 UTC; the 320 With AI rows at 16:44:51.303 UTC. The badge-derived ID set and filtered ID set match exactly. Full cutoff ties are preserved; all 12 panel observations are present, including the separate card.

This update is 75 hours after the 22 October 13:20 UTC baseline and 25 hours after the 24 October 15:20 UTC update. Observations are not promised at identical UTC times or equal daily intervals. This is the third daily observation before the festival closes, not a final festival report.

PlayRecon remains version 0.1.8. We refreshed its Metrics and Data definitions. Rolling 24-hour peak CCU is the maximum simultaneous demo-player count over that window, not festival/all-time peak, unique players or sales. Six-hour and 24-hour rank changes are relative positions, not player counts. Followers and reviews refresh more slowly; game followers differ from wishlists and curator followers. Quantitative wishlist conclusions remain excluded while displayed ranks conflict with the stated top-3,000 coverage.

Steam's Content Survey concerns generative content consumed by players, rather than every development tool. We do not infer AI-production percentages, exact demo App IDs from provider card keys, or firsthand play. This article publishes aggregates, not the full catalogue. Start with observation 1 on disclosure scope or the baseline and method for the earlier context.