Appendix
Methodology
Every artifact in this report is reproducible from the same five-step pipeline. The questions and the source list change per study; the method does not.
- 01
YouTube + podcast search
For each study we identify the universe of public commentators most likely to be discussing the question. The 2024 autopsy keyword cluster was: "Why Democrats lost / 2024 election post-mortem / Harris campaign / coalition collapse / Trump 2024." YouTube search and podcast directory queries returned the candidate set.
- 02
Source discovery + political compass scoring
From the candidate set we sample across the political spectrum, across audience sizes, and across formats. Each source is scored on the political compass by sampling its actual content and auditing the score before inclusion.
- 03
Content analysis
Transcripts are extracted, then analyzed for claim-level findings — hypotheses with confidence and stance, anchored to the specific quote that justifies them and the timestamp where the quote occurs. Every finding has a verifiable provenance back to the original video.
- 04
Theme analysis
Findings are clustered into themes the commentary itself organized around — not categories imposed by us. Top-level themes and sub-themes emerge from the data; we author taglines and key insights and edit the prevalence ranking.
- 05
Comment + engagement analysis
Per-video view, like, and comment counts are joined to each finding's source for engagement-by-quadrant breakdowns. Daily timelines are reconstructed from publication dates. Comment sentiment is scored for the next iteration of the report; not yet surfaced here.
This dataset
| Study | 2024 Election Analysis |
|---|---|
| Date range | 2024-11-05 → 2025-05-05 |
| Sources sampled | 98 |
| Videos analyzed | 203 |
| Hours of commentary | 258.5 |
| Findings extracted | 8,844 |
| Themes (top-level) | 25 of 126 hierarchical |
| Total reach | 112M views |
| Engagement | 2.24M likes · 370K comments |
| Dataset SHA | d537383bfd0e |
| Exported | 5/7/2026, 2:17:19 PM |
FAQ
- What dates does the study cover?
- The study samples content created between 2024-11-05 and 2025-05-05.
- How is engagement data reported?
- We snapshot YouTube view, like, and comment counts at the time of analysis. The numbers shift after that.
- How are political leanings determined?
- Each source is scored on the Political Compass by sampling transcripts of its actual content. We audit before inclusion.
- How is finding extraction done?
- Transcripts are processed by a combination of Gemini, Claude, and OpenAI models, plus human review. Each finding's quote and timestamp are stored alongside the hypothesis.
- What was the primary research question?
- "How do political commentators across the spectrum explain the 2024 election outcome — what they blame, who they think should change, and what they think comes next?"
- Is the full source and content list available?
- Yes — see the Sources page for all 98 channels. Per-video citations live inside each PullQuote's "Why we trust this" expansion.
- Can I reproduce the dataset?
- The export pipeline is in
scripts/export-study.tsagainst the after-action analysis backend. The frozen JSON dataset has SHAd537383bfd0efor cache busting.