Ong-Lots is a podcast analytics platform scoring how oral versus literate speech is, episode by episode and speaker by speaker. It ingests 1,100+ Odd Lots episodes, transcribes with diarization, and assigns each named speaker an orality score.
GSV-A3
Ong-Lots
Ong-Lots (GSV-A3, pace-layer A) is a podcast analytics platform, in the media-ecology domain, that treats spoken conversation as a measurable artifact. Its central question, borrowed directly from Walter Ong and Eric Havelock, is how oral versus literate a given stretch of speech is: formulaic, repetitive, audience-facing, and agonistic sit on the oral end, while abstract, subordinated, hedged, and impersonal sit on the literate end. The system runs this measurement across the entire Odd Lots podcast corpus -- more than 1,100 episodes -- scored at the level of the episode, the individual speaker, and the named guest. Scoring runs two parallel methods, both producing a 0-100 scale where 100 is most oral. A rule-based analyzer counts seven oral markers (formulaic phrases, sound patterns, repetition, first/second-person engagement, parallelism, discourse-marker memory aids, emphatic language) against nine literate markers (abstract nouns, syllabic complexity, subordination, hedging, passive voice, sentence length, impersonal/descriptive style, academic markers), each normalized per thousand words. Alongside this, a learned model called 'Havelock 3.0' is served behind a hosted API and scores the same text. Scores are computed and stored at every granularity: episode-level, per-guest-per-episode via a join table, per-speaker host scores in JSON, and a denormalized per-guest average across all of a guest's appearances. The substantive engine behind the scoring is its ingestion pipeline: a daily cron job pulls new episodes and transcribes any that are missing using Deepgram Nova (per the strategy grounding), keeping the corpus current automatically rather than as a one-time batch job. Ong-Lots is part of the four-repo 'Ong' cluster alongside Ong Back and Ongtology, sharing the Havelock/Ong oral-literate theoretical frame across different media (podcast speech here, video-to-text in Ong Back, cultural artifacts in Ongtology). Status: shipped and live at ong-lots.vercel.app as of 2026-04-08, built on Next.js 16, React 19, Prisma 7, and Postgres -- the most mature of the Ong-cluster projects.