Rank Predictor

Overview

After every mock submission, the predictor recomputes:

1. Your percentage score on each attempt, in order.
2. A linear regression line fitted through those points.
3. The projected percentile at attempt n+1.
4. That percentile mapped back to a rank using the average number of participants across the matching tests.

How the math works

The regression uses the standard ordinary-least-squares formula on points {x: attemptIndex, y: percentScore}:

slope     = (n·ΣXY − ΣX·ΣY) / (n·ΣX² − (ΣX)²)
intercept = (ΣY − slope·ΣX) / n
projected = slope · nextX + intercept
rank      = ((100 − projected) / 100) · avgParticipants

avgParticipants is the average totalParticipants recorded across the most-recent rank snapshots for tests in the matching exam type. If no snapshot exists yet, the predictor defaults to 10,000.

Requirements

• At least 2 submitted mock test attempts in the same exam type.
• Attempts must be in status AUTO_GRADED, AI_GRADED, or GRADED — partial drafts don't count.
• At least one mock test series must exist for the exam type with rank snapshots, so the participant count is real.

If you don't yet have 2 attempts, the predictor tells you how many more you need before it activates.

Viewing your prediction

1. Open Dashboard → Rank predictor, or scroll to the bottom of any mock series page.
2. Pick the exam type. The prediction card shows your projected percentile, the rank band (min–max), and how many attempts the projection is based on.

Confidence band

The displayed rank is a band: rankMin = round(rank · 0.85) and rankMax = round(rank · 1.15) — a flat ±15% buffer. The band gets tighter in your perception as you collect more attempts, because the regression slope is more reliable, but the band width itself stays the same percentage.

Trend direction

The prediction includes a trend field:

• improving — slope > 0; your percentile is going up attempt over attempt. Keep doing what you're doing.
• declining — slope ≤ 0; the regression line is flat or going down. Check the explanations on your last few mocks and shore up the weak topics surfaced by score prediction.

Limitations

• Linear assumption. Real improvement is rarely a straight line. The predictor over-projects if you've had a recent breakthrough, under-projects after a slump.
• Participants ≠ exam takers. The participant pool is Tuition.in users on that test, not the full national applicant pool. Treat the rank as relative to a representative sample.
• Recent attempts weighted equally. The first attempt and the latest one influence the slope equally. Take a fresh mock to push the projection toward your current ability.