Spirit of the Island (PPSA13934)

[JJenkx]

Existing thread

• I searched this category for the title ID and no thread exists for this title on any platform.

Title

Spirit of the Island (PPSA13934)

Content version tested

01.000.000 (the title shows build 1.3.7.1 on its menus)

AnyPS5 commit

Libraries from main 7feb37e (2026-11-03), Release, MinGW-w64 GCC 15.2.0 UCRT, with no local changes. The executable was converted on this machine on 2026-10-20 by the relinker of that day's main. The run went through the shared queue with scripted controller input (APS5_PAD_INPUT_FILE with tools/pad_input.py: cross every 10 s from 60 s) and no trace or profiling switches.

System tested

Windows 11 Enterprise 10.0.22621, Ryzen 9 5950X, Radeon RX 6700 XT (AMD proprietary driver).

Static notes

• Engine: Unity with IL2CPP (Il2CppUserAssemblies.prx and its companion modules are bundled).
• Saves: the title creates and reads save data through libSceSaveData.

Furthest point

New Game character creation. In a 240 s run on the libraries above:

• 60 s: title screen (Start, Options, Credits, "Code Unlock") over the 3D town scene.
• By 120 s: Start has opened the New Game form (name, property, town and region fields, with the character tabs on top) in front of the 3D character room, and the on-screen keyboard opens on the name field.
• Rest of the run: with only cross pressed, the scripted input types "g" into the name field until the run ends.
• Crashes: none. The run ended at its 240 s limit.

Earlier result on 2026-10-21 (Windows, main of that day): name entry was checked below SDL. Published IME input reaches the title's name buffer, and the name was shown in 2 of 3 runs. A physical key press through SDL was not tested.

What is shown

• Menus and 3D scenes: the title menu, the town behind it, the New Game form and the character room all render with correct geometry, colour and lighting in the screenshots taken at 60, 120, 180 and 230 s. I saw no corruption.
• Input: the on-screen keyboard follows controller input; cross types the highlighted key.
• Frame rate (window title counter): 9.7 fps at 60 s on the title screen and 26.7 fps from 120 s on in character creation. The 60 s figure is early in the run, so shader compilation may still have been running. This is not a measured performance claim; no trace was taken.

Current blocker

No emulator blocker seen up to character creation. The next step is scripted input that fills the name fields (or confirms with Done) to reach the island. That needs a longer input script, not an emulator change.

[JJenkx]

Update, same libraries (main 7feb37e): reaches gameplay on the island.

A second run (330 s, no crash, ended at its limit) used a longer input script:

• Start: cross on the title.
• Names: the name and property fields were filled by posting text to the window through the IME path (Ayla, Isle) and confirming with Enter.
• Confirm: R1 through the character tabs, then Options to confirm.

What the screenshots show:

• 193 s: the New Game form holds the typed names. The name reads ggAyla: two cross presses typed g on the on-screen keyboard before the text was posted. This confirms the 2026-10-21 name-entry result end to end on current main.
• 230 to 320 s: the new save starts and the character stands on the island beach with the full HUD:
  • the quest log ("Learning the Basics", "Farming Basics");
  • the Year 1 / Spring clock;
  • the money counter, minimap, health and stamina bars and the hotbar.
• Rendering: water, sand, rocks, shadows and the HUD look correct.
• Frame rate: the window title counter reads about 31 fps in gameplay (again not a traced measurement).

So the furthest point is now in-game, on the island. Movement and other in-game actions are not scripted yet.

[JJenkx]

Gameplay frame-time trace and a crash at the new-save load (same libraries, main 7feb37e).

Trace. In a third run, with only APS5_TRACE_SECONDS=290:10 set, the window covered gameplay on the island beach: 348 presents in 10 s.

• Frame time: p50 30.9 ms (about 32 fps), p90 34.6 ms, p99 65.9 ms.
• GPU: busy 16.5 ms per present, so it is not the limit.
• CPU: the one queue worker is 93.7 % busy. Driver.Packet takes 26.5 ms per frame, of which Driver.Packet.draw is 21.8 ms. On the draw committer (67 % busy), Graphics.Draw takes 13.3 ms and Recorder.Sync 7.6 ms. Gameplay is CPU-bound on per-draw command processing in one thread.
• Shader compiles: Driver.Draw.shader_compile_and_link still takes 4.8 ms per frame at that point (a fresh pipeline cache).
• Trace buffer: the queue worker's event buffer filled after 7 s (trace_dropped). The summary totals below are complete; the detailed events after that point are not.

{"commit":"7feb37ee3e30","dirty":false,"env":{"APS5_PAD_INPUT_FILE":"<path>","APS5_TRACE_DIR":"<path>","APS5_TRACE_SECONDS":"290:10"},"gpu":"AMD Radeon RX 6700 XT","driver":"AMD proprietary driver 26.10.1 (AMD proprietary shader compiler)","os":"Windows 10.0.22621","cpu":"AMD Ryzen 9 5950X 16-Core Processor","cpu_threads":32,"memory_mib":65457,"title_id":"PPSA13934","title":"Spirit of the Island","content_version":"01.000.000","resolution":"1920x1080","window":{"spec":"seconds 290:10","first_present":8276,"ms":10018.887},"presents":348,"frame_ms":{"p50":30.938,"p90":34.617,"p99":65.896,"max":88.894},"gpu_busy_ms_per_present":16.518,"worker_busy_pct":{"queue worker 0x0":93.664},"top_phases_ms_per_frame":{"Driver.Packet":26.545,"Driver.Packet.draw":21.760,"Driver.Draw":21.614,"Graphics.Draw":13.313,"Driver.Draw.draw_enqueued":11.933,"Graphics.Draw.draw_and_resource_release":11.142,"Recorder.Sync":7.622,"Driver.Draw.shader_compile_and_link":4.822},"trace_dropped":true,"trace_dropped_threads":{"queue worker 0x0":{"events_full_after_s":7.079,"events_lost":113104,"totals_full":false}},"trace":"aps5-trace-7feb37ee3e30-200272.json.gz"}

Crash. One of the three runs on these libraries crashed 6.7 s after confirming the new game (216.8 s), while the island loaded. The game's dlmalloc-style heap walk at app.exe+0x166cfbf read a chunk whose size field held its own address with the low bit set (0x225a7ffa1 at chunk 0x225a7ffa0), so heap metadata had been overwritten. The other two runs passed the same point. Goat Simulator 3 (#10040, UE4) shows the same class of heap corruption during its loading burst on Windows, so this may be one emulator problem rather than a title bug. I have not found the writer yet.

[JJenkx]

Windows status on current main 1c8c0efc8c: Windows 11 22621, RX 6700 XT, AMD 26.10.1. Scripted new game to the island, 330 s per run.

• The heap crash from Windows: Spirit of the Island (PPSA13934) crashes in its heap allocator during the new-save load; Goat Simulator 3 shows the same class #15350 no longer reproduces: 0 crashes in 4 runs on 1c8c0efc8c and 0 in 4 on 086c0686c8. Every run logs one fix(agc): drop depth and stencil plane results over memory unmapped since they were drawn #15466 depth-plane drop (0x225190000+0x870000). On 4f41f4c375 it was 2 in 8.
• The remaining stop is the 8-sample D32S8 depth target at about 290–300 s, where the game reaches the island town:
  • plain main: 2 of 4 runs stop with native multisample depth publication cannot distinguish an unwritten byte from a same-value guest store; 2 run to 330 s;
  • APS5_NO_HOST_IMPORT=1: 4 of 4 stop at about 287 s with guest writes raced native multisample depth import;
  • with open fix(agc): import a native multisample D32S8 target whose stencil plane the guest stored its clear value into #13023, that raced-import refusal is gone, and 3 of 3 then stop at about 287 s with guest writes over native multisample depth are not implemented.

So the next piece this title needs is merging guest writes into a native multisample D32S8 surface.

AI-assisted: yes.