Spurious season/episode numbers: title digits, absolute-number-in-parens, and 'Season N - EE' read as a range

[Toilal]

Summary

A number that is not the season/episode — a digit that is part of the title, an absolute episode number repeated in parentheses, a stray year, or a release-group id — gets captured as a season/episode, usually producing a spurious multi-value list (or a wrong scalar). Surfaced by the cross-parser test corpus (#873); ~15+ release names are affected, concentrated in anime-style names.

Patterns & examples

1. A number inside the title is captured as an episode

Release name	Expected	guessit today
Mob Psycho 100 - 09 [1080p].mkv	episode=9	episode=[100, 9]
Space Battleship Yamato 2199 (2012) episode 18	episode=18	season=2012, episode=[2199, 18]
[Golumpa] Star Blazers 2202 - 22 (… Yamato 2022) …	episode=22	episode=[2202, 22]
[SubsPlease] Fairy Tail - 100 Years Quest - 05 (1080p)	episode=5	episode=[100, 5]

2. Absolute episode number in parentheses double-counted

Release name	Expected	guessit today
[Hatsuyuki-Kaitou]_Fairy_Tail_2_-_52_(227)_…	episode=52	episode=[52, 227]
[Hatsuyuki] Dragon Ball Kai (2014) - 002 (100) …	episode=2	episode=[2, 100]
[Deep] Tegami Bachi (REVERSE) - Letter Bee - 29 (04) …	episode=4	episode=[29, 4]

3. Season N - EE read as a season range N..EE

Release name	Expected	guessit today
[SCY] Attack on Titan Season 3 - 11 (BD 1080p …)	season=3, episode=11	season=[3, 4, 5, 6, 7, 8, 9, 10, 11]
[Erai-raws] Granblue Fantasy … Season 2 - 08 …	season=2, episode=8	season=[2, 3, 4, 5, 6, 7, 8]
[HR] Boku no Hero Academia 87 (S4-24) …	season=4	season=[4, 5, …, 24]

4. Release-group id captured as episode

Release name	Expected	guessit today
[0x539] Somali and the Forest Spirit - S01E01 (WEB 1080p …)	season=1, episode=1	episode=539

Notes / scope

• Common thread: when a single season/episode is already anchored (e.g. - 09, S01E01, Season 3 - 11), a second numeric token nearby — title digits, an absolute number in (), a year, or a bracketed id — should not be merged into the season/episode list.
• Pattern 1080p screensize #3 looks like Season <n> - <ee> being interpreted as a season range <n>..<ee>; the <ee> after the dash is the episode, not a season bound.
• Pattern Part III & VI issue #4: a leading [0x539]-style group tag should be treated as a release group / id, not an episode.
• Logic lives in guessit/rules/properties/episodes.py (number extraction, ranges) and the group/title markers.

Reproduce / verify

These are part of the cross-parser corpus; run the opt-in suite and inspect the manifest:

uv run pytest -m cross_parser

Affected names are listed under season/episode in guessit/test/cross_parser/baseline.json. As the false numbers stop being captured, those entries should drop out on the next python scripts/import_cross_parser_tests.py run.

Acceptance criteria

• A title-internal number (Mob Psycho 100, Yamato 2199, Star Blazers 2202) is not added to episode.
• Season N - EE yields season=N, episode=EE (not a season range).
• An absolute episode in () after the main number (52 (227)) does not create a 2-value episode list.
• A leading [0x539]-style group tag is not parsed as an episode.
• No regression on the English corpus (episodes.yml).

[opensubtitles]

FWIW: we've implemented fixes for all four acceptance criteria of this issue over in guessit-js (the TypeScript port that tracks this project's corpus), released in v4.3.0–v4.3.2. The approaches may be portable back:

• Title-internal numbers / absolute-in-parens — an "anchored weak episode" scoring pass: when several weak numbers compete in one filepart, a zero-padded or dash-delimited number (- 09, 002) outranks title digits (Mob Psycho 100) and numbers quoted in brackets (52 (227) → 52, 29 (04) → 4). Ranges and value-agreeing duplicates are protected so 101 (01) and 16-20 (191-195) keep working.
• Season N - EE — a space-padded dash after a season word reads as season + episode (Attack on Titan Season 3 - 11 → s3e11); a glued dash (Season 1-3), plural Seasons, or a Complete marker in the filepart keeps range semantics (so Coupling Season 1 - 4 Complete is unaffected).
• [0x539]-style ids — a (?!0x[0-9A-Fa-f]) guard on the NxNN chains.
• No regressions on the full episodes.yml corpus in our suite (all cases above are fixtures now).

Happy to elaborate on any of these if a port is of interest.