[528z4b2kgc-ux]

What and why

Unreal Engine 4.25 writes version 9 paks (PakFile_Version_FrozenIndex). Their footer has a frozen-index flag byte between the SHA-1 and the compression method names, so the magic is 0xcd bytes from the end, and uepak.py never found it. unreal_paks therefore skipped the paks, so neither the archive decoder (uepak.job) nor the shader split saw them. Past the footer the pak is version 8's.

uepak.py reads version 9's footer at 0xcd and its index as version 8's (PER_FILE = LEGACY + (9,)), and refuses a frozen index with a reason. Unlike 4 and 8, a version 9 pak is not taken as inline outright: inline_shaders reads its file names. Monster Jam's pakchunk0 holds the two .ushaderbytecode archives and goes to uepak.job; pakchunk1 holds only cooked packages and goes to ue_legacy_jobs. Its indexes are encrypted; since 050580d they are opened with main's key through uepak.Keys like versions 4 and 8.

Rebased onto current main (Nov 12): this PR's earlier key handling (pak_key, main= parameters in inline_shaders, unreal_paks, ue_legacy_jobs and profile_guess.py) is gone, since main now does that for every legacy pak. What's left is the version 9 footer and index.

Reference entry: Unreal Engine 4.25 paks (version 9).

Checked

• Tests: new dm/tools/tests/test_uepak_v9.py (synthetic paks and a synthetic main, no game files or keys): 6 tests (footer, frozen index refused, Zlib and stored files read, a pak with archives not inline, encrypted index opened with the key from main, key carried in the jobs), all of which fail on main. The test_uepak_* tests (60), test_docs, test_profile_guess and test_nsp_* (426) pass. ruff 0.17.2 is clean on the changed files; pyright not run locally.
• Game files: Monster Jam Steel Titans 2 (010051B0131F0000, 1.0.2), its two paks (4.7 GB) with this branch: main finds neither; with the change pakchunk0 is an archive pak (44,512 + 1,722 records, the 44,148 + 1,717 unique shaders of the game's build) and pakchunk1 an inline one. The game played on the Mac and an iPhone with those shaders, with the earlier version of this change.
• Devices: M5 Pro, macOS 27.2; iPhone (iPhone19,4), iOS 27.2.
• Not checked: a full NSP build on this branch (the archives' Oodle decode); regress.py on other Unreal games (no builds on hand). Version 4, 8 and 11 paks take the same code paths as before.

Docs

• Reference entry added and docs_symptoms.py run
• No game files, keys or extracted game data in the change

🤖 Generated with Claude Code

[528z4b2kgc-ux]

The two red jobs (host tests and sanitizers) are the same single check in t_save_sync.c: "both changed: the report says conflict, with the cloud's device". That run started before dd0f2ef landed, so it still had the old test, which expects a "Mac" prefix where the runner reports VirtualMac2,1. This PR only touches uepak.py, its test and two docs. A re-run against current main should come back green.