Each need is scored 0–100 from four sub-scores. Each sub-score is the percentile rank of an underlying signal within the full source dataset (ties get the average rank).

• Clicks = pct(total clicks past 360d) — proven traffic.
• Units Sold = pct(units sold midpoint) — proven money on the table.
• Avg Units Sold = pct(avg units per top product) — high means each top product sells well, suggesting room for new entrants.
• Top Products = pct(# top-clicked products) — more top products is treated as positive: a well-developed niche with multiple winning entrants.

Final = w_c·Clicks + w_u·Units + w_a·AvgUnits + w_t·TopProducts with weights renormalized to sum to 1.0. The deployed page bakes in the top 20,000 needs by an unweighted baseline (sum of all four sub-scores). POD/apparel categories are toggleable but excluded by default. Range strings like 250,000 – 300,000 are parsed to midpoints.