The Edit Room
The produced feed was edited progressively harder from 2015 to 2018; raw YouTube tape, from 2022, shows what the same conversations sounded like unedited. Here's the gap, month by month, and why every Ledger score uses raw tape only. Our transcripts keep every "um" the edit removed.
Chart A: one show, two tapes
Median guest disfluencies (um and uh per 1,000 words), month by month. ▬ the produced podcast feed · ▬ the raw YouTube tape, same show, same guests. Hover any dot for the month's roughest tapes. Click it to open the episode.
the full story
Same host, same guests; the difference is the edit. The feed descends from around 30 to a surgical zero by 2018, then YouTube arrives in 2022 and shows what the conversations really sounded like all along. Gaps in the gold line are months with fewer than five feed measurements (after the show moved to YouTube, feed audio becomes scarce). We never interpolate. The gold spike in April 2025 is real: a same day "Daily Deal" news episode shipped essentially unedited, and its guests measure 29 um and uh per 1k on the feed, full raw tape levels. Daily turnaround leaves no time for the edit room.
Chart B: speakers on both kinds of tape
132 guests appear on both kinds of tape: produced rate across, raw rate up, one dot per person. 109 of 132 sit above the diagonal.
what this shows
The diagonal means "the edit changed nothing". Sitting above it means the person sounds rougher on raw tape (median +9.9 per 1k). Same humans, different edit.
Chart C: the host across eleven years
Same host, eleven years. His measured filler rate tracks the production calendar rather than his speech.
why that matters
A speaker's own rate does not rise and fall on this schedule. If the metric were measuring Harry, his numbers would be steady across eras. They track the edit instead.
| Year | in the produced feed | on raw tape |
|---|---|---|
| 2015 | 5.8 /1k | — |
| 2016 | 5.9 /1k | — |
| 2017 | 2.5 /1k | — |
| 2018 | 0.5 /1k | — |
| 2019 | 0.4 /1k | — |
| 2020 | 0.4 /1k | — |
| 2021 | 0.5 /1k | — |
| 2022 | 0 /1k | 9.9 /1k |
| 2023 | 0 /1k | 6.6 /1k |
| 2024 | 0 /1k | 5.4 /1k |
| 2025 | 0.8 /1k | 5.8 /1k |
| 2026 | 0.8 /1k | 7 /1k |
Chart D: argument clarity, produced vs raw
Each dot is one person's argument clarity shift between produced and raw tape (1–5 scale, 149 guests). 104 of 149 score higher in the produced feed.
the full story
We re-scored the argument clarity of the same people on both kinds of tape, counting only guests with enough assessed exchanges on each. Produced tape scores about 0.2 higher at the median; one guest gains 1.43. The edit removes more than "um"s: it removes the tangents and dead ends that make an answer meander. This is why argument clarity scores are only issued from raw tape.
what this means for every score on the Ledger
None of this is an accusation. Editing is what good producers do. But it is why this site refuses to score anyone on produced tape: the microphone chain would be doing the talking. Every speaking style measurement and every argument clarity score on the Ledger comes from raw recordings only, and the methodology holds every metric to one rule: the same person must score the same on either tape, or the metric doesn't ship. It is also why a 2015 quote keeps the repetitions and fillers from the tape. The early feed was barely edited; the heavier edit began in 2016, and by 2018 the silence was surgical.