Changelog

This release's changes. Every release, with how each change was checked: CHANGELOG.md.

1.0.0 - 2026-10-02

The first release on Thunderstore.

New seeds only. Location ids moved: DLC1 107 -> 105, DLC2 124 -> 123. 0.4.6 seeds: finish them on 0.4.6, or regenerate.

• A Skip no longer leaves the post-level panel over the next puzzle.

  • On a level the game sends straight on with solutions left (Cupcakes, Water Glasses, Fountain Pens), the game's skip moves the level out by itself, and the panel the mod asks every run puzzle for came up on top: the next puzzle opened under it, in RetryUI_GameState. DevTools trace: SkipLevel, TransitionLevelOut, RetryUI, ShowMenu, the tween-out.
  • Found by the 1.0.0 DLC gate, which stopped on it. It was in every DLC gate since 2026-09-30 and in droha's morning log; the harness used to get out by a route no player has.
  • A Skip on such a level now keeps the game's own route (Core AfterPuzzleRoute, Skips.FinishingTheLevel): the next puzzle opens once, in play. A level built for the panel moves on through it as before.
  • The gate checks it: "no puzzle opened under the post-level panel" (full 30, quick 28).

• On Thunderstore. The mod installs with r2modman or Thunderstore Mod Manager, BepInEx with it. The README now opens with how to install, in the order of Archipelago's own setup guides (required software, a mod manager first, then by hand, then joining), and the setup guide, installation page and package README carry the mod-manager steps.

From a sweep of the backlog (2026-10-01).

• A finished generator no longer relaunches itself.

  • The cause: the game asks for the level already running (StartLevel with no index and no reload) just after a puzzle finishes, and again as it goes back to gameplay when a level select closes over the level. The game does nothing with that call.
  • The mod's launch prefix read the call as a card click, took the running level, and made it that slot's seeded reload. So:
    • a finished generator came back about a second after "navigation: next": 3 to 5 times in every gate since 2026-09-27, and after DLC1 Trophy Cabinet and DLC2 Water Glasses on the DLC route;
    • a seeded level started over when a player left it through the level select and closed back to it. Breadtags did this mid-solve in the quick gate.
  • The call is now left as the game asked (Core LaunchRequest; the log says track: the game asked for the running level ... left as asked).
  • The quick gate's seed, before the fix: 6 relaunches and 1 mid-solve refund. After it: none of them, 26/26.

• The Seeing Stars Boss files each phase as it is solved.

  • Its Lock and Compass endings reach the save as those phases are solved, with no completion of their own, so until now they were filed only at the finish.
  • The mod now reads the save again on the frame after the game records a solution (SaveData.SaveLevelData, SaveSystem.SaveGame). Once a completion has run on the slot, that completion keeps its own filing.
  • Checked with DevTools marksolved:1231:DLC2Boss_Lock and ..._Compass on the Boss's run slot: each was filed at once, with no completion.
  • droha's playthrough (2026-10-01): Compass and Lock were each filed about 30 s after their phase was solved, when the game saved it, and Knife at the finish.

• Seed Pods and Clover reload when Symmetry arrives mid-level (ObjectLock.ResetOnUnlockLevels), as Cat Food Cans and the stacked levels do.

  • The game tints their pieces to 0.9 once, after the intro, and only the ones not locked then. Unlocked in place, every piece came back at 1.0, brighter than it should be.
  • The reload replays the intro. Every piece in both levels is Symmetry's, so it loses nothing.

• Trophy Cabinet is just its Solution (DLC1 ids after it move).

  • Its Cupboard Doors were a check of their own, where Clock Cupboard's and Tea Cabinet's are not. The doors are the way in (Gadgets), not a puzzle: droha's play on 2026-10-02 sent Cupboard Doors, Items Placements and Solution within 6 s.
  • The doors are notALocation. The level is then one group, so Items Placements' part check goes too; Gadgets is still required. DLC1 locations 107 to 105.

• Figurines' second sort is a Solution of its own (DLC2).

  • droha did all three solutions on 2026-10-02 and the hover showed 2 of 3. The two sorts are one controller's two arrangements (SortingItemsDraggables_0 and _1); the ending table had only _0, plus Groupables-(Achievement)_0, an id no game has reported. So _1 fell back to the first ending, Draggables, already in, and filed nothing.
  • Endings are now Draggables, Sorting Items 1 and Sorting Items 2. The Groupables (Achievement) controller is notALocation: never solved in play, so as an ending only a Skip could earn it. DLC2 still 124.
  • fixtures/solution-ids-observed.tsv re-harvested: Figurines is the only row that moved, and Core's every-observed-id test fails on the old table.

• Canapes is just its three Solutions (DLC2). Its Cracker Positions check fired before every solution, for the same three crackers each time; droha: "it's not really a mini solution". solutionOnlyParts; DLC2 locations 124 to 123.

• Every different ending files its own check, even where the ending table is wrong (Core CheckRouter.ForEnding).

  • An id the table did not know filed the first ending in the list even when that one was already in; that is how Figurines lost its star.
  • Now each id found on a slot takes the ending it names if free, else the first free unseen one, else the first free one. A wrong guess in the table costs the check's name, never the check. The answer still depends only on the ids found before it, so it is the same at every launch.
  • Why it matters: a sweep of every level (2026-10-02) found no other wrong id, but 11 levels with several controllers list ids no game has reported yet (Bookshelf credits all three of its solutions to its top left shelf). Those are guesses until played; this makes them safe.
  • droha then played the ten (the Boss's Lock and Compass were filed from the save on 2026-10-01): all 23 unreported endings came in as the table had them, each its own check, so no name moved. fixtures/solution-ids-observed.tsv holds them now (223 of 244 ids name a controller).

• Robots keeps every robot's state where it is without Ordering.

  • Its Indexables group (Ordering) shares Robots 2, 3 and 7 with IndexedDraggables, which needs nothing. The lock frees a piece any free group holds, so those robots could be extended without Ordering.
  • Robot 9's antennas and Robot 1's arm buttons are locked, but they take clicks through colliders under their free robot (DevTools objects:, 2026-10-01; droha: "not just the antenna guy").
  • Now, while Ordering is withheld, the colliders of every button (IndexIncrementTrigger) on the level's ten states are off, so no press reaches them. They come back on with Ordering.
    • A press that gets past anyway is refused and logged as a warning (the new index hold patch). There were 0 in the hand test.
    • Not a refused press, which two earlier builds made. A press refused mid-handling, whether its state change or the whole handler, left Robot 3's hearts container switched off once Ordering came, so no heart would go in. With no press while held they went in (droha's A/B).
    • Not the game's own LockIndex, which the first build set. Robot 2's key clears that when it comes out, and the head then moved.
  • The big robot's chest-and-arms layer, which carries its heart, is grey while its state is held (Core ObjectLock.IndexHoldArt). Its body stays lit.
  • The robots still drag: their group needs no ability.
  • droha's hand test, 2026-10-01:
    • Ordering withheld: every robot dragged, and none changed state, including the key robot with its key in and out.
    • Containers withheld: nothing went into a robot.
    • The big robot "looks perfect".
    • Ordering given back mid-level: every button worked, the hearts went in, and the level finished.
  • Math Set too. Its half circle and compass are shared with its free drag group the same way, and droha extended both without Ordering. Now neither changes; the Inside solution, both closed as they start, can still be finished, and its check waits for Ordering as the logic says.
  • Robots and Math Set only (Core ObjectLock.IndexHoldLevels):
    • Water Glasses and Music Box share no state piece, so the lock already held theirs (droha's hand test).
    • Junk Drawer Transforming and Combs have the pairing, but their logic asks only for Drawer. Junk Drawer Transforming was finished with Ordering locked.

• No run puzzle goes to the Daily Tidy page any more, and every run level ends the same way. droha: "If we know what levels go there can't we just prevent those levels from doing it? ... a set ending sequence for all levels so we are in control".

  • Until now every finished generator went to the game's Daily Tidy state, and the daily guard pulled the player out to the next slot: 105 times in the second player's 0.4.2 playtest, 32 in droha's, every generator in the gates.
  • Where the game decides: LevelManager.OnLevelCompleteTweenOutComplete, which every way out of a finished level passes through. It runs IsDailyTidy's body inline (randomized or holiday-dated, and not an archive level), so the daily guard's postfix on the getter never saw it. Found from the interop's xref cache and the game's machine code, then seen with DevTools trace: on Stamps (Randomized): the tween-out, get_IsArchiveLevel, StopGame(True), the Daily state.
  • The new DailyDecision answers that check's one real call, get_IsArchiveLevel, for the level leaving, inside that one call only, and the game takes the branch every other level takes. It logs daily decision: <level> leaves as a run puzzle, not a daily.
  • Generators now leave by the panel's route like every other level (Core AfterPuzzleRoute: the panel with solutions left, otherwise on to the next slot, or the track when nothing is playable).
  • On the quick gate's seed, 18 fresh runs of Stamps, Batteries and Post-It Notes in a row: 54 generator finishes, 0 Daily Tidy states, 0 rescues, the next slot opened each time.
  • Every generator, the player's way:
    • a run of only the 16 base generators, each finish moving on by itself: 16 of 16 moved on with the decision, no Daily page. That includes Books, Pencils, Buttons and Spice Jars, which left by the panel's Continue with an ending still to find;
    • the 4 DLC generators, each from its card: 4 of 4.
  • droha's hand test, 2026-10-01: two generators in a row each opened the next puzzle directly, never a Daily page "not even for a flash"; the pause menu's Exit went to the title; Books showed the panel, and its arrow went on to the next puzzle. droha had never seen the page in recent releases either: it was happening unseen, and only the logs showed it.
  • The daily guard's rescue stays as a backstop and now warns. The gate has a new check, "the run never went to the Daily Tidy page" (quick 27, full 29).
  • The ReactToDailyCompleting postfix is gone. That getter decides nothing: the game reads the field directly. Its machine code is also shared with 14 other getters (Level.get_HintUsed among them), which the postfix risked answering false during a run.
  • The gate's arrow check may now start on a generator, since generators end the same way.

• Post-It Notes (Randomized) no longer logs 15 render-texture errors when it finishes.

  • Each note takes a texture from a shared pool about 5 frames after the level is built, and switches its line camera on to draw into it. In about 1 load in 3 the cameras stay on for good. Then, when the level is torn down, Unity logs Releasing render texture that is set as Camera.targetTexture! once per note.
  • That happened only when the level started straight after another level finished, never from a card. Measured with the new DevTools cameras, values: and trace: (7 of 7 runs that logged it had the cameras on at load, none of the others).
  • The mod now switches any note camera still drawing off as the level is torn down or rebuilt (PostItCameras). Six runs: it fired in the 3 with the cameras on, 0 errors in all 6.
  • Why the cameras stay on: not the notes. In every run with them on, each note had finished its snapshot (DevTools values:NoteVisualSetter, 2 of 8 runs on the generator seed; the backup fired in both, 0 errors in all 8). The game switches the note cameras off in one place, PostItNote_LevelRandomizer.OnLineGenerationComplete, after a frame count rather than when every note is done (read from its machine code), and that switch-off sometimes misses them. It is the game's own; a player sees nothing, as each note shows a copy of its picture.

• A generator launched outside the run's slots is built, not empty. The launch prefix gives such a level a stable seed of its own so it is not left empty. That seed was the level index times 2654435761 cast to an int, which wrapped negative for 6 of the 20 generators (Books, Batteries, Post-It Notes, Trim Plant, Microscope, DLC2 Figurines). A negative seed reads as "no seed", and Post-It Notes came up empty 9 of 9. Core LaunchRequest.FallbackSeed keeps it above 0 (the seeds that were positive are unchanged); booted outside the run, Books, Microscope and Calendar were each built (13, 5 and 24 objects). The log line now names the seed. Found by a DevTools boot; no player route to it is known.

• The run's track is taken down if it is left up over the title (DlcGuard TickTrackLeftOver). droha's screenshot from the DLC gate showed the cards, the strip and the Close X over the main menu. The route behind it was the harness's own (below), but a screen like that should not stay. The DLC gate on player routes (2026-10-01 23:10, 29/29): 161 menu listings, none with the track over the title, and the takedown never had to fire.

• The gate moves only the way a player can. Every visit leaves a level by the retry panel's Menu or the pause menu, then Levels, and opens the next puzzle by clicking its card on the run's track. Where the mod moves on by itself, the gate waits for it.

  • It used to leave by the post-level Level Select, Close and a forced title, then boot: the next level. It also pressed the post-level Level Select on a level the mod had already moved on from, which no player can. That built the track twice and left it over the title (droha: "you shouldn't be hitting it either. it's sloppy").
  • test_scheduler.py pins it: the gate's visit functions use no DevTools shortcut, and its next press is read from the open menus.
  • DevTools pause now enters Menu_GameState and pauses the game, as a player's Esc does: the same log lines as droha's Esc (2026-10-01), and a Reset from it closes the menu. It used to raise only the menu-open event. The state then stayed at Gameplay, so a Reset or a Skip from that menu went back to gameplay "already active", nothing took the menu down, and its Levels stopped answering - a pause no player can reach. The old form is pause:event.

• The DLC navigation probe goes to the track the player's way, and its two one-off failures are explained.

  • "The second session opened no card": the probe reached the track from the title with a forced menu:levels. The title stayed the menu system's active menu under the track (menu=Title Menu), and every card click selected its level and launched nothing (48 clicks, 0 launches, 2026-10-02). It now presses the title's or the pause menu's own Levels. Three runs since: 30 of 30 checks, every card launched.
  • "retry panel: moving on failed: NullReferenceException": the probe pressed the hidden panel's post-level Level Select in the frame between the mod hiding the panel and moving on, then the mod's move-on called the panel's NextLevel over the track it had opened (seen once in those three runs). No player can press a hidden panel; the probe now presses it only when the panel is on screen. The mod's move-on also stands down when the game is already on the track, the DLC level select or the title ("left the post-level screen another way first"). Forced 8 times on the generator seed, it never threw (7 times the move-on came first; once the press came before the panel was hidden, so there was nothing to stand down).

Written off

Tried again, in more than one way, and not seen again; gone from the backlog.

• Trim Plant's Grow NRE (twice, 2026-09-30; Sprite.get_bounds in the plant's Grow coroutine).
  • By the player's own routes, 0 of 50: Esc then Levels, and Esc then Reset, 25 each, 0.5 to 3 s after it loaded (2026-10-02).
  • It does reproduce, 4 of 25 (4 of the 5 exits at 0.5 s), but only with a pause that keeps time running: DevTools' old event-only pause, now fixed to enter Menu_GameState as Esc does. A player's pause stops time, so the grow cannot run on into a level being torn down.
  • Both gate sightings also relaunched Trim Plant mid-grow (the relaunch above, now gone).
• A boot over an unfinished Junk Drawer 2 breaking the DLC1 Boss's first solve (harness only).
  • In game: 0 of 3. One part of Junk Drawer 2 solved, then boot:1128 and a forced solve.
  • Gate logs: 0 in the 13 since.
• Books (Randomized) finishing on its Draggables rule by itself.
  • In game: 0 of 5 symmetric layouts left alone for 60 s. Three were booted and two booted twice, with no run connected so the layout was the seed's.
  • Also 2026-09-28's 100 layouts.
  • Every gate Books finish since ended on Shuffle_0.
• The arrow session's lost Post-It Notes completion (2026-09-24).
  • In game: three --only-arrow sessions passed both checks, each solve followed by PartSolved and LevelComplete.
  • Kept arrow logs: 32 of 33 completed. The other sent its solve while the game had paused itself on losing focus (the next item).
• A gate run that stops receiving game events (2026-09-24).
  • Cause: the game's own Pause(true) holds every event. boot undoes it, and the gate warns of a solve sent while paused.
  • No stop in the 67 gate logs since 2026-09-25.
• Mirror's candle and Ghost Cat's cats not grey while locked: correct, and droha left them (2026-09-30).
• Held states drawn into a piece's one picture stay lit (Robots' key robot and Robot 7, Math Set's half circle and compass). Their states do not change while Ordering is withheld; each is a single flipbook, so a tint greys all of it or none, and they still drag. droha: leave them (2026-10-01), no overlays.
• Steam Cloud conflicts while testing: the tip is in docs/dev/testing.md.
• The DLC gate leaving its paper plan at visit 14 (2026-10-01 14:05). The DLC gate of 23:10, on player routes, followed its plan through all 22 visits, 29/29.
• A full gate's harness dying after visit 7 (2026-10-01 13:58, its traceback lost). The full gate of 22:46, stderr kept, ran to the end: 29/29, nothing on stderr.