Pass 2—Trail Synthesis
Pass 1 tells you what motifs are present in each section. Pass 2 asks a different question: what happens to a single motif over time?
Where Pass 1 operates one unit at a time against the full motif vocabulary, Pass 2 operates one motif at a time across the full Text. It takes the isolated section-level observations from Pass 1 and transforms them into a coherent, navigable trail—a structured account of how the motif emerges, develops, shifts, and reaches a culminating articulation across the Course.
What it does
For each motif, Pass 2 receives the complete ordered sequence of Pass 1 occurrences and compares each occurrence to the ones that came before it in the trail. It assigns scores that characterize development (novelty, milestone importance), flags that mark structural anchor points (seed, zenith, milestone), and writes editorial summaries that describe what changed at each stop.
Pass 2 is not a second detection pass. It does not look for new motif occurrences, re-read the source text, or reinterpret what Pass 1 recorded. It treats the Pass 1 dataset as its entire evidence base. The question is not “is this motif here?”—Pass 1 already answered that—but “what does this occurrence add to the motif’s story so far?”
How it works
Pass 2 is executed by a specialist custom GPT with a different instruction set, rubric, and output schema from Pass 1, designed for trail-level comparison rather than unit-level detection.
For each motif, the GPT receives a CSV file containing every Pass 1 occurrence of that motif, ordered by the Text’s chronological sequence. Each row includes the full Pass 1 record: salience score, all six controlled codes, evidence pointers, key terms, and first-pass context.
The GPT processes the trail sequentially, comparing each occurrence to earlier rows. It does not look ahead—an occurrence is evaluated only against what has come before it in the trail, not against what comes after. This forward-only constraint mirrors the reader’s experience: when you encounter a motif in Chapter 15, you have the context of Chapters 1–14 but not Chapters 16–31.
Gaps in the trail—chapters where the motif does not appear—are treated as neutral rather than as evidence of thematic regression or absence. The trail is built from documented appearances, not from silence.
Echo vs. substantive gating
Because Pass 1 stores echoes (salience 0.10–0.24) alongside substantive occurrences (salience 0.25 and above), Pass 2 begins by distinguishing these categories using the Pass 1 salience threshold.
| Type | Salience range | Role in the trail |
|---|---|---|
| Substantive | ≥ 0.25 | Eligible for seed, zenith, and milestones. Can carry development. |
| Echo | 0.10–0.24 | Valid trail marker. Typically low novelty. Not eligible for seed, zenith, or milestone flags. |
This gating prevents the trail narrative from being shaped by faint mentions while still preserving echoes for completeness. In the companion interface, echoes can be displayed as markers along the trail or hidden entirely depending on the reader’s preference.
What it produces
Each occurrence in the Pass 2 output receives the following fields.
Novelty score
A decimal value between 0 and 1 estimating how much this occurrence adds or shifts the motif relative to prior occurrences in the same trail. Novelty is driven by observable differences in the Pass 1 record—not by general knowledge of the Course, but by what can be seen in the data already collected:
- Changes in the first-pass context (a new function, consequence, or application attributed to the motif).
- Shifts in controlled codes (a change in tone, frame, agency, relational focus, abstraction, or teaching level).
- The appearance of new key terms that indicate a genuinely new angle rather than a synonym for something already established.
| Range | What it signals |
|---|---|
| 0.0–0.2 | Reinforcement or echo—the motif is doing what it has done before |
| 0.3–0.5 | A small new emphasis or application |
| 0.6–0.8 | Substantial development—a new dimension, register, or function |
| 0.9–1.0 | A turning point in the motif’s trail |
Echo occurrences typically receive low novelty scores unless a clear meaning shift can be demonstrated from the recorded evidence.
Milestone score
A decimal value between 0 and 1 estimating the importance of this occurrence as an anchor point in the motif’s overall arc. Milestone scoring is conceptually distinct from novelty. An occurrence may be novel without being a durable anchor—it might introduce a tangent that doesn’t recur. And an occurrence may consolidate earlier developments in a way that makes it milestone-worthy even if it is not maximally novel.
| Range | What it signals |
|---|---|
| 0.0–0.2 | Incidental—not an anchor |
| 0.3–0.5 | Useful but not anchoring |
| 0.6–0.8 | A major stop in the motif’s arc |
| 0.9–1.0 | A core anchor point |
Milestone flag
To keep milestone highlighting sparse and readable on the site, the total number of milestone-flagged occurrences per trail is capped. The seed and zenith are always flagged as milestones. The remaining milestone slots are assigned to the top-scoring substantive occurrences by milestone score, using a formula that scales with trail length:
K = clamp(round(N_sub × 0.15), min=2, max=7)
Where N_sub is the count of substantive occurrences (salience ≥ 0.25) and K is the total number of milestones including the seed and zenith. This enforces editorial sparsity—a reader following a long trail sees a handful of marked anchor points, not a wall of highlights—while allowing longer trails to have proportionally more stops.
Seed flag
The seed marks the first major establishing occurrence of the motif in the trail. This is not necessarily the first mention, but the earliest substantive occurrence that plausibly begins the motif’s story for a reader in a meaningful way. Each motif has exactly one seed.
Zenith flag
The zenith marks a peak occurrence where the motif reaches its most mature, integrated, or culminating articulation. Each motif has exactly one zenith, just as each has exactly one seed. The zenith must be a substantive occurrence, but it is not necessarily the last mention of the motif.
Variation summary
One to three sentences describing what changed at this stop relative to earlier occurrences. For high-novelty stops, the variation summary explains the shift—a new register, a new application, a new contrast. For low-novelty stops, it describes how the occurrence reinforces or echoes the established pattern. The variation summary is not displayed directly to site visitors. It serves two purposes: auditability (documenting what Pass 2 observed about each stop’s relationship to earlier stops) and input to the editorial layer (the editorial content in Pass 4 draws on variation summaries when writing student-facing discussions).
Second-pass context (decision log)
An internal record documenting the rationale for the novelty score, milestone score, and any seed/zenith/milestone assignments. Like the variation summary, this field is not displayed to site visitors. It exists for auditability—allowing the analyst or a future reviewer to understand why Pass 2 made the decisions it did—and as input to the editorial layer.
What the trail looks like after Pass 2
When Pass 2 is complete, each motif has a structured trail: an ordered sequence of occurrences, each one carrying both the Pass 1 annotation (what the motif does in this section) and the Pass 2 synthesis (how this occurrence relates to the trail so far).
The trail now has narrative shape. It has a designated beginning (the seed), a sparse set of marked anchor points (milestones), and a peak (the zenith). Each stop is accompanied by internal documentation—variation summaries and decision logs—that record what changed and why. A reader following the trail on the site won’t see these internal fields directly, but they inform the editorial content generated in later passes, and they ensure that every judgment in the trail can be traced and reviewed.
What the trail still lacks are corpus-wide metrics that characterize the motif’s overall role in the Text (how frequent, how intense, what kind of motif it is). Those come from Pass 3. The editorial layer that turns all of this into student-facing content comes from Pass 4.