Pass 3—Corpus Metrics
Pass 3 is one of two passes in the pipeline (along with Pass 5) that involves no AI. It is pure database computation—SQL procedures that aggregate what Pass 1 and Pass 2 have already recorded, producing corpus-wide metrics that characterize each motif’s behavior across the full Text.
Where the earlier passes ask “what is this motif doing here?” (Pass 1) and “how does this motif develop over time?” (Pass 2), Pass 3 asks a simpler, more statistical question: what kind of motif is this?
What it does
The Pass 1 and Pass 2 data is rich but local. A motif occurrence tells you about one motif in one section. A motif trail tells you about one motif across its appearances. Neither gives you a way to compare motifs to one another—to say that one motif is a dominant structural pillar while another is a rare spotlight that appears intensely when it does.
Pass 3 creates that comparative layer. It computes how often each motif appears (frequency), how strongly it tends to perform when it does appear (intensity), and then classifies motifs into descriptive role categories based on the combination of the two. These metrics and labels are what allow the site to present motifs not just as a flat list but as a landscape, with core pillars, atmospheric constants, spotlights, and other recognizable roles.
How it works
Pass 3 runs in two stages, executed as stored procedures in the database. The first stage computes frequency and intensity metrics. The second stage derives classification bands and role categories from those metrics. Both operate on the same underlying data: the motif occurrences recorded by Pass 1, enriched by the rankings implicit in salience scores.
Stage 1: Frequency and intensity
For each motif, the following metrics are computed across the fixed corpus of analyzed units.
Frequency measures how often a motif shows up.
| Metric | Definition | What it tells you |
|---|---|---|
| units_any | Count of units where salience ≥ 0.10 | How broadly the motif appears, including echoes |
| units_sub | Count of units where salience ≥ 0.25 | How broadly the motif appears in a substantive way |
| freq_any | units_any / total units | Echo-or-higher presence rate |
| freq_sub | units_sub / total units | Substantive presence rate |
Intensity measures how strongly a motif performs when it does show up.
| Metric | Definition | What it tells you |
|---|---|---|
| i_strength | Mean salience across occurrences where salience ≥ 0.25 | How central the motif tends to be when substantively present |
| i_foreground | (Occurrences ranked in unit’s top 3 by salience) / (total occurrences) | How often the motif is one of the dominant motifs in its unit |
| i_index | 0.5 × i_strength + 0.5 × i_foreground | A blended intensity score combining strength and foregrounding |
A note on foreground ranking. Within each unit, motifs are ranked by salience in descending order using a dense rank, which means ties are allowed at the top-3 boundary. If three motifs share the same salience in a section, all three count as top-3 rather than arbitrarily excluding one. This is a small decision, but it matters for motifs that frequently co-occur at similar salience levels.
Stage 2: Bands and role categories
Once frequency and intensity are computed, Pass 3 classifies each motif along both axes using a tercile-by-rank scheme.
How banding works. All motifs are rank-ordered on freq_sub (for the frequency axis) and on i_index (for the intensity axis), independently. Each ranking is then split into three groups of roughly equal size—Low, Mid, and High. Ties are broken deterministically by motif ID so that every motif lands in exactly one band on each axis. The bands are relative—they describe a motif’s position within this corpus and this motif set, not an absolute threshold that would hold across a different analysis.
How role categories are derived. Each motif’s frequency band and intensity band are crossed to produce a role category:
| Frequency | Intensity | Role category | What it means |
|---|---|---|---|
| High | High | Core pillar | Shows up often and is strong or foregrounded when present. These motifs are the structural backbone of the Text. |
| High | Low | Atmospheric constant | Shows up broadly but usually not in a dominant role. These motifs color many sections without driving them—more background fabric than foreground. |
| Low | High | Spotlight | Rare, but when it appears it’s a big deal. These motifs are highly foregrounded in the few sections where they show up. |
| Low | Low | Fringe | Rare and generally not foregrounded. These motifs are at the periphery of the Text’s thematic landscape. |
| Mid | Mid | Midfield | Moderate presence and moderate intensity. These motifs function as connective tissue—useful for understanding transitions and context without dominating the landscape. |
| Other combinations | Mixed | Any cross not covered above (e.g., high frequency / mid intensity). These motifs don’t fit neatly into one of the primary categories. |
These categories are navigation labels, not interpretive conclusions about the Course. A “fringe” motif is not unimportant—it is simply rare and typically not foregrounded in this analysis. A reader may find a fringe motif deeply significant for their own study. The labels describe the motif’s statistical behavior across the corpus, not its spiritual weight.
Global cutoffs
When the bands are computed, the effective boundary values—the maximum freq_sub in the low band, the minimum freq_sub in the high band, and the equivalent values for i_index—are stored for reference. This gives the site stable reference points for displaying band boundaries and allows future re-computation to be compared against prior runs.
What the data looks like after Pass 3
Every motif in the dataset now carries three layers of characterization:
- Unit-level (from Pass 1): where the motif appears and how it functions in each section.
- Trail-level (from Pass 2): how the motif develops across the Text—its seed, milestones, zenith, and the novelty of each stop.
- Corpus-level (from Pass 3): how the motif compares to other motifs—its frequency, intensity, and role category.
Together, these layers let the site answer a wide range of questions. A reader can ask “what motifs are in this section?” (Pass 1 data), “how does this motif’s treatment here relate to its earlier appearances?” (Pass 2 data), and “is this motif a core pillar of the Text or a rare spotlight?” (Pass 3 data), all from the same underlying dataset.
What the data still cannot do is speak to the reader in an accessible voice. The scores, codes, and metrics are precise but clinical. Translating them into student-facing editorial content is the job of Pass 4.
Full documentation of the frequency and intensity metrics is on the Frequency and Intensity page. Full documentation of the role categories is on the Role Categories page.