A pack folded for Windows and macOS keeps one spelling of each case twin. Extracted on a Steam Deck or Linux, the core asking for the other spelling finds nothing, and only the installer added it. Where the platform also runs on a filesystem that keeps case, the README now lists each spelling left out beside the one it holds: 12 for EmuDeck and 10 for RetroArch, the manifest's case_variants.
fetch_large_file verified a cached copy and returned its name, which another run replaces when the release serves a re-upload: the caller then packed bytes it never verified. The returned path is now a hard link under .cache/large/.by-sha1/<sha1>/, kept only if it holds the inode that was hashed. Revisions no name holds are dropped a day later. restore_large_files hashes each copy before moving it into place.
Step 5 compared every narrowed pack with verify and never the --split parts with the full pack, so a part writing a path the full pack settles passed every check. check_split_packs builds each platform's parts in memory and compares their union with the full pack, path for path and byte for byte.
generate_emulator_pack resolves with offline=offline and fetches a release asset the cache lacks, while verify_emulator resolved offline and called it missing. verify --emulator now applies pending_release_asset as verify --platform does, accepts --offline, and the pipeline compares emulator packs offline on both sides.
The RetroDECK verify excerpt counted 1800 core files in the pack. Companion entries no longer sweep cheats, GameIndex.yaml and DVD firmware to the BIOS root, and the same run prints 1764.
files/?sha1= of a file the collection no longer holds answered that no
file the site knows has the hash. generate_db.py now records each sha1 a
rebuild drops in retired.json, with the day and the file in its place,
backfilled from the history since 0d586ef6b (4,139 files). The lookup
answers with it and leads to the successor's card.
install.py tells a MiSTer card by its MiSTer binary and menu.rbf since
73b82375d. installer.md said so, the FAQ and the README did not, and sent a
MiSTer user to --dest the card's root, where nothing reads the files.
tools.md listed what keeps a platform entry out of removal_candidates
without the platform's launchers: records, which candidates.py checks
right after a path (_PROOF_ORDER).
testing-guide listed the generated artifacts without derivable.json and
the run samples of wiki/installer.md, and tools.md gave the manifest
command alone. The nightly's commit rules also had no place for the two
files its offline pipeline rewrites, so autorun stopped on them.
PCSX2 stats <bios>.rom1 then <stem>.rom1, case-sensitive on Linux (BiosTools.cpp:214-229), and so does PCSX2 1.6 for erom. The pack wrote the stored SCPH-70004.ROM1, never found. A companion entry also seeded a scan of its own with a 4 MB bound and swept cheats, GameIndex.yaml and DVD firmware to the BIOS root. A companion now scans its image's folder for its own extension only, any size, since the code reads min(size, filesize), and is written as the code spells it.
A data tree's archive came with no progress line, and the run ended on the BIOS bytes alone (310.5 KB where 82 MB arrived). Its download now draws the progress line and adds its bytes and time to the end line, a cut attempt taken back. It writes no progress event: those count the download event's files.
Lakka keeps its BIOS on its STORAGE partition (/storage/system). A computer now tells that partition by the retroarch.cfg RetroArch writes there, whose system_directory is Lakka's /tmp/system overlay, and --platform lakka on a computer asks for the card instead of writing to /storage/system here.
Batocera and KNULLI write the image name, not the board, to the card's board file (post-image-script.sh). The target scraper now records each board's images from Config.in, the target lists carry them, and the installer maps a card's image to its board. KNULLI's boot/knulli.board is read too.
data/ppsspp-assets was refreshed after the manifests were written: 46 lang files changed. Regenerated so the next commits carry their own changes only.
profile_sync.py finds and fetches the project's main by itself. The
documented --base upstream/main failed in any clone without that remote,
where the same command without --base passes.
The README, the setup pages and adding-a-platform tied archived to a
scraping schedule and a CI matrix that do not exist. Archived keeps a
platform out of the runs that cover every platform unless
--include-archived is passed.
The ci area of check_freshness read the PyPI and action pins only. The
exact versions website/package.json pins (astro, starlight) now get a row
each, STALE once npm publishes a newer release.
architecture.md said the pack and verify decide through slots.arbitrate(),
while a test forbids them to call it: they read slot_choices() through
pack_overrides(). build_claim_index, the name-keyed index the module
docstring described, had no caller outside its own tests.
Step 5 builds the --required-only, --no-game-data, --region, --target
and --system packs of every platform in memory and compares their paths
with verify run under the same narrowing. verify_platform gains
required_only, generate_pack a record mode. A run samples three regions,
targets and systems per platform (155 packs, about 3 minutes);
--full-narrowings compares all 1,150 (25 minutes). A test fails when
_narrowings() gains a dimension that is neither compared nor exempted.
The published platform contract validates against platform.schema.json, which does not carry the resolver's per-entry system key. The two inheritance tests and the alias call check read the new entry shape, and the testing guide lists test_maker_folder_names.
Demul 0.7 build 251220 (demul.exe sha1 e326ae66, the build RetroBat and Batocera 43.1 ship) loads its BIOS from MAME-style sets named in seven tables in the executable: dc, naomi, naomi2, segasp, awbios, hikaru, mie and segadimm, each member checked by size, crc32 and sha1. The profile lists every set with its members and cites the tables and the loader by address. dc_boot.bin, dc_flash.bin, naomigd.zip, hod2bios.zip, f355bios.zip, f355dlx.zip, airlbios.zip and jvs13551.zip are not read by this build and leave the profile.
Fourteen lists gain the data trees their cores read. AmberELEC and EmuELEC fetch the 192 blueMSX machine configs from the tree instead of holding them back, and RetroPie drops the blueMSX tree its lr-bluemsx module installs.
A platform pack carried only the trees its own systems declared, so for_platforms reached RetroArch, Lakka and RetroPie alone and every other platform running blueMSX, FBNeo or Dolphin got no tree. The pack, the install list and the pipeline's tree members now add the trees of the libretro cores the platform runs, below the system directory and build folder the core reads, unless the platform installs or never opens them. A platform entry the manifest resolved into a carried tree becomes that tree's member, which the installer can fetch: AmberELEC and EmuELEC no longer hold 192 blueMSX machine configs back. stonesoup reads its tree beside the content and pcee2 compiles its resources in, so neither tree reaches a BIOS pack.
AmberELEC's nec-pc-88 declares DISK.ROM with an empty md5. QUASI88 reads
that name as n88sub.rom, the 2 KB PC-88 disk ROM (libretro.c:1085), but
the name step served the 16 KB MSX disk ROM, the only primary of that
name. The quasi88 alias never reached by_name, which refuses an alias a
primary already carries.
load_platform_config now writes each entry's system as platform_system.
A platform entry with no hash and no owning emulator, whose system id
names a maker, takes no name candidate filed under another maker, in
the name step and in the hash_mismatch step after it. When nothing is
left, by_shadowed_alias, the profile aliases by_name refuses, answers
within the system's maker folder. names_manufacturer and BUCKET_FOLDERS
move to common.py, and generate_site imports them.
Across every registered platform the only resolution that moves is
AmberELEC nec-pc-88 DISK.ROM, from bios/Microsoft/MSX/DISK.ROM to
bios/NEC/PC-88/N88SUB.ROM. Manifests, truth and exports are otherwise
byte-identical.
The RetroDECK report quoted on reading-reports.md counts the two np2
FONT.ROM files, the testing guide lists test_np2_fonts, and the source
paths record follows fontv98.c and the new pcem upstream pin.
release-process.md listed tbblue.mmc, removed from bios/ on 2026-10-20, and left out uqm-remix-disc1.uqm. A test now reads the list against git ls-files.
reading-reports.md said the cross-reference checks whether a profile's files appear in the platform config, a rule by name. Since 2026-10-10 declaration_settles counts a declaration only at the core's slot or at a size the core accepts, and an alias only where the core reads it.
architecture.md said a file found by its name claims nothing. Since 2b535525e slots.py counts such a copy (size_exact) when it has the exact size the profile declares and the platform's file another size, as RetroDECK's sgb2.boot.rom shows.
pin_frozen was prose, and no tool read the build file it cites, so a platform building another revision left the profile holding one nobody ships. pin_build records the platform, its repository, the revision read, the file, the variable and its value. check_freshness reads the file again where the platform's YAML now stands and reports STALE when the value moved. 36 of the 46 held profiles carry it, each value read from the build file at the platform's scraped revision. The other ten name no build file and are counted on one row: six RetroBat binary packages and four holds not tied to a platform build.
Abdessamad Derraz