Cold vs hot write analytics

# Cold vs hot write analytics

A per-student analytics view that compares cold write scores (In-Class Essays, no tutor, timed) against warm write scores (tutor-assisted regular essays) over the course of a year, tracking whether the gap is closing — the signal that the student is internalizing what the tutor teaches.

## What the teacher needs to see

The core visualization is two sets of plotted points over time for a given student:

- **Warm write points** — grades on tutor-assisted essays (the regular assignments)
- **Cold write points** — grades on In-Class Essays (unassisted)

No trend line needed right away — just plot the points. With 3–4 cold writes per semester, the data is sparse enough that individual points tell the story more honestly than a line pretending at continuity. A trend line could come later as data accumulates, but v1 is dots on a chart.

The insight is in the gap between them and how it moves:

- **Gap closing** = the student is getting better on their own. The tutor is building real skill.
- **Gap staying wide** = the student relies on the tutor but isn't internalizing. May need a different intervention.
- **Both rising** = best case. The student is growing in both contexts.
- **Warm rising, cold flat** = the tutor is doing the work, not the student. Red flag worth surfacing.

First In-Class Essay is just a baseline — no comparison yet, just the first dot on the chart.

## Where does this live?

The student-level popout feels like the natural home — it's already per-student and teachers are already going there to understand individual students. But the popout is also slated to get AI insights, so how to fit both isn't obvious yet. Probably a section or tab within the popout, but the layout needs thought.

Could also surface in multiple places eventually:
- **Assignment-level summary** — after each In-Class Essay is graded, show a quick comparison against that student's warm write average right on the submission view.
- **Class-level overview** — a view showing the whole class's cold/warm gap at once, for identifying which students are internalizing vs. relying on the tutor.

But v1 is per-student in the popout.

## Data volume and retakes

3–4 In-Class Essays per semester is realistic. That's 6–8 data points per year — sparse but meaningful, especially plotted against the more frequent warm writes.

Students can also retake cold writes — not revising the old one, but after a conference with the teacher, trying another cold write with a new prompt. This gives students a path when they're unhappy with a score: meet with the teacher, talk about what went wrong, try again under fresh conditions. The retake is a new data point on the chart, not an overwrite of the old one. Both dots stay.

## Granularity

Start with overall grade comparison. Over time, the analytics could become more granular — breaking out sub-scores (thesis quality, evidence, analysis, etc.) by cold vs. warm to pinpoint exactly where the tutor is carrying the student. "Marcus's overall cold write scores are improving, but his evidence scores stay flat — the tutor might be doing his evidence work for him." That's a v2 insight, but the data model should be ready for it.

## Audience

Teacher-only for now. Students don't see their cold vs. warm comparison. Could open to students later, but the framing would need to be careful — "here's your growth" not "here's how much worse you are without help."

## Raw data first, AI insights later

V1 is raw data — plotted points, the gap, the direction. Teachers interpret it themselves.

V2 could layer in AI-generated insights: "Marcus struggles with conclusions in cold writes but nails them with the tutor — the tutor may be writing his conclusions for him." That's powerful, but it needs the granular sub-score data and enough volume to say something meaningful. Build the foundation now, earn the insights later.

## Resolved decisions

- [x] Per-student first, starting in the student-level popout (exact layout TBD).
- [x] 3–4 cold writes per semester is enough — plot the points, no trend line needed for v1.
- [x] Start with overall grade, build toward granular sub-scores over time.
- [x] Teacher-only for now.
- [x] Plot individual points, not a trend line — data is too sparse for lines in v1.
- [x] First In-Class Essay is just a baseline dot.
- [x] Raw data first, AI insights in v2.
- [x] Retakes are new data points (new prompt, new cold write), not overwrites.

## Open questions

- [ ] How does this fit in the student popout alongside the AI insights that are also planned for that space?
- [ ] Should the retake flow be formalized (teacher initiates from the popout? student requests?) or just "teacher assigns another In-Class Essay to that student"?
- [ ] What does the class-level view look like — a table of students with their gap direction, or a multi-student chart?
- [ ] When sub-scores become available, does the teacher drill into them from the popout, or is that a separate deeper view?