Beta: Scores and links may change as the catalog improves.

About the Scores

What the scores are

Each sermon is scored on four dimensions: clarity, engagement, exposition, and theology. The overall quality score is the unweighted average of the four. These are signals, not endorsements.

A high score means the scoring model found strong evidence of that quality in the transcript. It does not mean the sermon is "good" or that you'll agree with its theology. Use the scores to filter and explore, then listen and judge for yourself.

The four dimensions

Each dimension is scored independently against its own rubric, so a sermon can be clear but shallow, or theologically rich but hard to follow.

Clarity
How clearly the message communicates. A clear sermon has one dominant idea stated early, logical flow between sections, accessible language, and examples that illuminate rather than distract.
Engagement
Whether the sermon offers fresh insight and concrete vividness. An engaging sermon makes familiar Scripture feel fresh and abstract truth feel tangible, rather than predictable and generic.
Exposition
How well the sermon explains the biblical text. A strong expository sermon is driven by the passage: its points come from the text, with attention to context and structure, rather than using the text as a springboard.
Theology
The doctrinal substance and faithfulness of what the sermon teaches, evaluated from a conservative Reformed confessional perspective. Deep theology articulates doctrine precisely and keeps the gospel central.
How scoring works

Scores are produced by an automated model that analyzes each transcript. The model was calibrated against detailed reference evaluations of hundreds of sermons, graded using the rubrics above. Scoring runs offline in batches, separately from this site.

In practice, most scores fall between 3 and 8, and a typical sermon lands around 6.5. Small differences of a few tenths of a point are not meaningful.

Known limitations
  • Automated scoring can be wrong. The model misses nuance and context that a human listener would catch. Treat scores as suggestions.
  • Transcript quality matters. Poor audio or transcription errors affect scores. Some sermons are penalized unfairly.
  • English only. Only English-language sermons are scored.
  • Unscored is not zero. Some episodes have no score yet; they simply haven't been scored, and they sort after scored episodes.
Recalculation policy

The scoring model evolves. When a significantly improved version ships, the whole library is rescored, so scores can change over time.

If you notice a score that seems off, it may predate the current model version. Scores are not permanent.