[jdsika]

Closes #61.

Problem

prefix.cc has no curation, and anyone can register a prefix there.

• Every tagged release since v0.1.1 has shipped 7 to 13 spam or broken prefix.cc rows.
• The spam removed by hand in Feb 2024 (d202004) came back with later refreshes.
• main now has 15 such rows in prefixcc.csv and in each merged* context. Almost all are canonical, so for example load_converter("merged") compresses https://onlyfans.com/sammuwuu/x to fugas:x.
• prefix.cc keeps changing. A refresh today would bring in more, such as leaks pointing at a different spam site.

The same problem has been reported to prefix.cc itself (cygri/prefix.cc#42). But prefix.cc's code hasn't changed since 2021, so we can't rely on it being fixed there.

Change

The PR has two commits, so the freeze can be dropped independently if you prefer.

1. fix: remove the spam and block it from coming back

   • Adds src/prefixmaps/data/prefixcc.excluded.yaml, which lists excluded hosts (subdomains included) and prefixes, each with a reason.

   • from_prefixcc() drops matching entries before building the context.

   • The loader rejects unknown sections, so a typo can't silently disable exclusions.

   • Removes the 15 rows from prefixcc.csv, merged.csv, merged.monarch.csv and merged.oak.csv:

     Rows	Why
     fugas, leaks, przecieki, veelana	onlyfans profiles
     onlyfans	cutt.ly short link
     faptica, check, dick	adult sites
     rankrage, seokoeln	SEO agency site
     fs, qualityze	software product pages; qualityze is the spammer reported on Blacklisting spam sites cygri/prefix.cc#42
     type, ljkl	teste.com junk
     bgcat	prefix.cc serves this namespace with its Cyrillic replaced by ?

   • No other row shares a prefix or namespace with these, so no other row's status changes.

   • A new test (tests/test_core/test_prefixcc_exclusions.py) fails if any bundled context contains an excluded host, or an excluded prefix from prefix.cc. Run against main's data, it fails for prefixcc and all three merged contexts.

2. feat: stop refreshing prefix.cc by default, as asked in freeze prefix.cc refreshes #61

   • The ETL uses the bundled prefixcc.csv snapshot, minus exclusions, so adding an exclusion takes effect on the next run.
   • Canonical rows are loaded first, because merging depends on the order in which rows arrive. In plain sorted-CSV order, 73 prefix.cc rows in merged would change. With canonical rows first, prefixcc.csv and all three merged CSVs rebuild byte for byte from the snapshot. So make etl causes no churn, and the CSV edits in this PR are exactly what the ETL now produces.
   • slurp-prefixmaps --refresh-prefixcc still downloads prefix.cc, and prints every added or removed prefix.cc row for review.
   • load_context(..., refresh=True) also uses the snapshot for prefix.cc. This is documented.
   • Semweb prefix fixes now belong in linked_data.curated.yaml, which already takes precedence over prefix.cc in the merged contexts. README and docs say so.

Not included

These look like junk rather than spam, so they're left for a maintainer to decide:

Prefix	Namespace
sirext	http://t.me/sirextt/247:
hello	a YouTube channel
beth	http://www.google.com/
twitter	http://stocktwits.com/
leak	http://maco.com/
da	a wowman.org query URL
vwdms, vwgw	intranet hosts
test, test2, meat	example.com / .invalid placeholders

Each would be one line in prefixcc.excluded.yaml.

The pgo link rot mentioned in #61 needs no change. A namespace IRI doesn't have to resolve, and PGO is a real vocabulary.

Verification

• An independent adversarial review re-checked:
  • the row removal (byte level, CRLF intact)
  • the rebuild identity, including order-independence on a fresh prefix.cc download and 20 random permutations
  • the host matching
  • the refresh=True behaviour
  • an end-to-end --refresh-prefixcc run into a temporary directory
  • the tox -e refresh path
• Tests: mutation checks confirm the new tests fail if the ingest filter or the snapshot filter is removed, or with main's data.
• Lint and suite: flake8 with the tox -e flake8 plugin set passes. The suite passes locally on Python 3.12 and 3.8, except test_w3id, which also fails on main and is fixed in fix: read w3id namespaces from ids/ after upstream move #84. CI will show that one failure until fix: read w3id namespaces from ids/ after upstream move #84 is merged.