The Start here pages ran 460 to 2,100 words and the first command of
Get your files sat 254 words down. Each page now opens on its action,
with short sentences and one fact per block. The facts a newcomer does
not act on moved one level down: the legal analysis to wiki/legal.md,
the project terms to wiki/reference-terms.md, the annotated run and the
installer failures to the installer page, collection figures to a new
Explore index.
tests/test_site_copy.py holds the budgets on the generated pages, and
the testing guide states them with their reasons.
Four statements the repository contradicts are rewritten: prboom.wad
is compiled into PrBoom (emulators/prboom.yml), Batocera reports a
mismatch as UNTESTED (batocera-systems checkBios), 13 hle_fallback
entries are required, and the gaps headline splits missing firmware
from game data and unread entries.
EntryCategories asked every core in turn: 3.8 million proofs for 7,430
entries. ProofIndex looks each key up once, and verify --all --json is
byte-identical (24 s to 17 s). generate_site and generate_readme now share
one coverage, cached in tmp/coverage-cache keyed by its inputs, so the
README step drops from 25 s to 3 s.
Five tabs from Start here to Contribute, a level on every page, a device picker on Get your files that lists every registered platform, and platform pages that open on where the files go, the command, the pack and how to check. FAQ and troubleshooting keep the questions asked before installing, the technical ones move to their own page, and getting started points to the pages that took its content.
A release recorded without a platform's pack linked its button to that release, which serves no pack for AmberELEC, EmuELEC, KNULLI, MinUI, NextUI or MyMinUI. The cell now names the release and points to the installer, which reads the file lists on main. Core folder links resolve each core name to the profile page that describes it (mednafen_vb is beetle_vb).
minarch hands every core SDCARD_PATH/Bios/<TAG> as its system directory
and opens no file in it. The scrapers read each pak's launch.sh (core,
tag, devices), each device's core makefile (the project a library is
built from: MyMinUI's dosbox_libretro.so is dosbox-pure) and the minarch
statement, at a pinned commit. The installer recognises the card from
.system/version.txt and refuses a folder inside it.
A browser upload into bios/ lands on a fork and opens the pull request, and CI already hashes the file and names its folder. CONTRIBUTING asked for a fork, a clone, five packages and the whole suite, and promised a PR comment that a fork's read-only token cannot post. The issue chooser links the upload page.
The quick install block gains the PowerShell form that passes options and a link to the installer page. The collection size uses the GB format of the download table and the site, counts get separators, and the sentences about RetroArch's existence check and the method describe the code instead of judging documentation.
The contributor install line left out mkdocs-material and
pymdown-extensions, so pipeline step 8b (render site) was skipped.
The ranges are the ones deploy-site.yml installs, and a test reads
them from the workflow.
profile_sync gains --changed (added or edited profiles, uncommitted
ones included) and --added, both against the merge base of --base
(origin/main by default). CONTRIBUTING runs --changed --strict instead
of one --emulator call per new profile, and the CI job runs --added
in place of its own diff loop.
The 3.12 guard rested on generate_readme f-strings that have since been
rewritten. Every script compiles under CPython 3.9.25, 3.10 and 3.11, and
only generate_pack's parenthesized context managers fail under 3.8. Under
3.9 the offline pipeline (--skip-packs) completed with no change to the
committed artifacts, the 15 manifests matched 3.12 byte for byte, and the
RomM pack was identical. CI now compiles every script under a real 3.9.
Quick Install was the one Title Case heading left in the published
text. The Files sentence was glued to the paragraph above it, two
paragraphs before the table whose column it describes. It now names
the column and sits just above the table.
--emulator accepts a buildbot core name from a profile's cores:, and a
mistyped target gets the closest names instead of every target. diff_truth
shows destinations and reports unpaired entries as evidence, not removals.
The verify summary counts discrepancies, a shared destination names both
systems, a truth hash conflict names both cores, and pipeline steps carry
their place in the real run instead of N/8. Every GitHub API caller takes
the token from github_token(), which falls back to gh auth token.
Troubleshooting, the FAQ and the download page sent people who only ran
the installer to verify.py, which reads the repository, not their BIOS
folder. Download links now name the exact asset of the recorded release,
the --dest example names the BIOS folder on the card, the slice layout
left with the last release that used it, and the shown verify output is
what the tool prints.
Dropping the double-dash asides left 72 comments and docstrings with
half a dash glued to the next word ("shape -a size or a size range -is
all"). They now read as a colon, a comma or parentheses, and a prose
style test names any that comes back.
The README opened on a sales line. It now says what a pack holds: the files its emulators read, as their source code names them, minus what the emulators install themselves.
The guides said files over 50 MB become release assets and that RPG Maker and ScummVM were the trees kept out of dedup. .gitignore decides the first, four committed files are over 50 MB, and NODEDUP_DIRS holds eleven trees.
CONTRIBUTING and the testing guide installed pyyaml and jsonschema only, while the CI test job also installs markdown. On a contributor's machine the wiki heading test skipped and the run read OK. Both name it now, and a test compares them to the packages the tests import.
generate_readme, generate_site and slots resolve against data/ but took no hold when run on their own, as the documented commands do. A refresh swapping a tree mid-read then gave counts from half of it. Each now holds the trees shared, like verify and the builder.
On a hash platform the builder leaves out a file whose declared hash the local copy contradicts, but the README coverage added every UNTESTED file to the present count. RetroDECK read 2,008/2,008 for a pack carrying 2,007. verify now lists the destinations the builder refuses, and the coverage subtracts them.
verify kept its own copy of the placement rule and missed the flat RomM folder, so the report still counted 543 unreachable files as packed. It now asks the builder's ExtraPlacement. The README fraction counted every held core file, including those a platform cannot place or that a same-named packed file hides. It now counts what the pack holds.
Profiles, the wiki, comments, docstrings, command output and schema descriptions carried em-dashes, clause-joining semicolons, unicode arrows, box-drawing banners and check-mark glyphs. Each sentence was rewritten rather than having its character swapped. Hashes, paths, source_ref values and every cited file:line are unchanged, which a comparison of each parsed profile against the previous revision confirms.
validate_pr comments now say PASS and FAIL in words, and source_ref values in a profile's structured blocks render as code. test_prose_style fails if any of these comes back, in tracked files or in the generated site.
argparse completed a prefix into the only flag starting with it:
slots.py --emulator read profiles from a directory named after the
emulator and reported nothing, and pipeline.py --all built every pack
variant. Every parser now refuses an abbreviation.
The documented command, discover tests, makes tests/ the top level,
so tests/__init__.py never ran and the variable it sets to keep tests
offline was missing from every local run. Every documented and CI
run now passes -s tests -t ., and a test requires it.
SHA256SUMS.txt sat beside the artifacts it vouches for, so whoever could
rewrite a release rewrote the list with it. The packs were already
reproducible, which answers corruption and lets a third party rebuild an
archive byte for byte; nothing answered a rewritten release.
The list is now signed with an ed25519 key kept for this alone, and the
public half is allowed_signers at the repository root, so verification
does not go through the release page: ssh-keygen -Y verify against the
committed file, then sha256sum --check. Rehearsed on all three outcomes:
a good signature, a tampered pack caught by the sums, a rewritten list
caught by the signature.
The release steps sign and upload the signature, the README points a
downloader at the procedure, and the reproducibility section says what
each half proves. Rotation keeps retired lines so past releases stay
verifiable. Three tests hold the trust root, the signing step and the
documented principal in agreement.
The home page and the stats export counted every file carrying a
provenance record, the provenance page and the README only the system
files. The site published 553 and 566 for the same quantity, one click
apart, and the export paired the wider count with composition.systems as
its denominator. common.count_catalog_matched is now the single source,
scoped to the systems bucket.
The FAQ had drifted from the profiles it describes: MAME pinned at 0.287
against 0.289 in mame.yml, Adler-32 attributed to Dolphin's IPL rather
than the DSP ROMs that carry known_hash_adler32, and the per-emulator
verbose report named as the only content check on an existence platform,
which skips the DISCREPANCY line the platform report raises itself.
Tests read both sides: no generator may count matches inline, and each
FAQ claim is checked against the profile or the script that owns it.
The contributors block is the only part of the README that comes from
the network, and a refused request returned an empty list, which deleted
the section. That happened during a pipeline run and the result was
committed; the freshness check then regenerated the section and failed
on the difference.
An unavailable list now republishes the one already there and says so.
Losing it is a worse answer than a stale one, and it makes an offline
regeneration additive rather than destructive. Reverting the change and
simulating the same outage empties the section again.
Abdessamad Derraz