Hard filters, then soft weights

The rankings engine is intentionally two-staged. First, hard filters remove schools that fail constraints you will not negotiate — wrong major family, wrong control type, tuition above your ceiling, graduation rate below your floor, wrong state or urban setting. That cut runs as SQL in DuckDB-Wasm in the browser.

Second, soft weights score the survivors. Affordability, earnings, graduation, selectivity, and Mobility (Pell Grant share) are min-max normalized inside the filtered subset, then combined with weights that always rebalance to 100%. Change a dial; the order updates immediately. Nothing is sent to a ranking server to be judged.

Why the order matters

If you soft-rank the entire universe first, impossible schools pollute the scale. A $90k sticker school can look “affordable” relative to even pricier peers. Filters first keep the math honest to your actual choice set.

What “on your device” buys you

• Reproducibility — same filters and weights yield the same order for the same Scorecard extract.
• Privacy — your preference profile need not leave the browser.
• Speed — re-scoring a filtered subset is cheap once Parquet is loaded.

Try the enginePick a major, set a price ceiling, move one weight.