UrBroToddWiki

Sources & method

Meta Info3 min readLast updated: Oct 31, 2026

Everything in this wiki is derived from the channel's own transcripts. This page records what the corpus is, how it was verified, and where the numbers come from — so any claim can be checked or re-run.

The corpus

Item Detail
Long-form videos 453 (Mar 21, 2023 → Oct 24, 2026)
Shorts 29 (Feb 2023 → Oct 2026)
Words 1,904,339 long-form
Total views 70,724,140 (snapshot 2026-10-28)
Verification Compared against the live channel on 2026-10-28: 453 long-form + 29 Shorts, no gaps

Transcript source: YouTube auto-captions, one .txt per video in Data/transcripts/, named NNN_videoId.txt.

Verification

  • Long-form: all 453 transcripts were checked against the full caption stream. 451 are byte-identical to the concatenated captions; [451] and [452] were added after the original snapshot and match at 99.9–100% token level.
  • No truncation: an audio spot-check of three videos found no missing speech.
  • Shorts: all 29 were checked against audio with Whisper; six were re-transcribed from audio because the captions were incomplete ([005], [020], [021], [024], [025], [027]). [029] has no captions at all and is entirely Whisper-derived.
  • Timings: Data/transcripts/451_*.srt and 452_*.srt plus Data/shorts/NNN_*.srt; rolling VTTs in Data/raw/late_uploads_vtt/. Converted with Scripts/vtt_to_srt.py.

Counts

Scripts/count_terms.py produces Data/metadata/term_counts.md. Method:

  • Case-insensitive substring counts over the lower-cased transcripts.
  • Character rows are "videos with at least one name match / total mentions", with caption variants folded in (Kerry/Carrie, Jay Bablo/Pablo, Donny/Donnie, Gregors/Gregor, Mato/MTO).
  • Quoted counts in the wiki come from that file unless a page says otherwise.

Treat every count as a floor. Captions mishear words, drop names and merge speakers. A character can be on screen without being named. The counts are for scale and recurrence, not for precise screen time.

Known false positives. The character table in term_counts.md includes Reef (3 videos, 4 mentions) — that is a coral reef in [239], not a person. Similar noise is possible for common-word names ("Ted", "Reed", "Clyde"). The wiki's character pages were checked against the analysis cast list; Reef is excluded.

Name mentions are not appearances. Santa has 476 mentions but 14 videos; Jay Bablo has 150 mentions across 71 videos — both are devices that get named a lot without carrying scenes.

Episode ids

Every episode has a stable id from its filename: 045_zOpLCRmHS9c.txt → [045]. The ids were assigned when the transcripts were extracted, so they are not chronological ([000] is Dec 2025, [045] is Oct 2026, [443] is Mar 2023). Always use the date column when order matters.

Wiki/Episodes/Index.md is the lookup table: id → date, title, length, views, YouTube link.

The wiki build

Scripts/wiki_build.py generates:

  • Wiki/Characters/*.md — 22 character pages. The prose (role, blurb, notes, see-also) is a curated dict inside the script; the appearance tables are computed from index.json + the transcripts.
  • Wiki/Characters/README.md — the roster.
  • Wiki/Episodes/Index.md — the episode lookup.

Everything else under Wiki/ is hand-written. Do not hand-edit generated pages — edit the script and re-run:

python3 Scripts/wiki_build.py

Refreshing after new uploads

python3 Scripts/fetch_new_uploads.py --write --shorts
python3 Scripts/fetch_metadata.py
python3 Scripts/build_catalog.py
python3 Scripts/rebuild_stats.py
python3 Scripts/count_terms.py
python3 Scripts/wiki_build.py

Limitations

  • No speaker labels in most of the corpus. Speaker-switch markers (>>) appear in 103 videos, all from 2025–26, so quotes are attributed by episode rather than by character.
  • Caption spelling is unreliable. The wiki normalises known errors; unknown ones remain. See the table in the glossary.
  • Views are a snapshot (2026-10-28) and will drift.
  • Titles promise more than the episode delivers. "99 YEARS" does not mean the video covers 99 distinct years; "FULL MOVIE" is a compilation.
  • The analysis document is not a source of truth. Where it disagrees with the transcripts, the transcripts win — for example, it dates Baby Todd's introduction to Mar 2024, while the name first appears in [399] (Sep 2023) and the adoption episode is [285] (Jul 2024).

See also