Search Scoring — Interactive Playground

Play with each scoring clause (Gauss, near, exists, range, equals, boost, compound) and see live results, then quiz yourself.

interactive1 min readinteractiveatlas-searchscoringlearn-by-doing

📘 Informative architecture document. No credentials, IDs, vault names, or internal URLs.

🎮 Search Scoring — Learn by Doing

Odyssey ranks vehicles with Atlas Search by combining small scoring clauses. Reading about them is one thing — playing with them is how it clicks. Below is a live playground: move the sliders, flip the toggles, and watch the score change. Then test yourself with the quiz at the bottom.

#How scoring works (the 30-second version)

A search result's final rank is a sum of clause scores. Each sortType (RECOMMENDED, SHIPPING_FEE_LOWEST, HIGH_PRICE_NEW_CAR, …) is just a different recipe of clauses with different weights. The clauses fall into two families:

Match clauses decide whether a vehicle qualifies for a score: equals, range, exists, text.
Score shapers decide how much it scores: Gauss decay, near, boost, constant, and function expressions.

A compound clause glues them together — its should list adds the pieces up.

#The clauses, hands-on

Use the tabs below. Each one mirrors a real block from the production score-weights config (the same JSON you can open under Search Score Weights).

🔔 Gauss decay — smooth "closer is better" curve (used for shipping fee).
🎯 near — proximity reward with a half-score pivot (newest year, lowest mileage).
✅ exists — reward vehicles that have a field (free-shipping flag, 360° spin).
📏 range — keep only values inside a [min, max] window.
🟰 equals — exact match; splits NEW vs USED scoring lanes.
⬆️ boost / constant — multiply relevance vs assign a fixed score.
🧩 compound — combine clauses; should scores add up.
🧮 function score — compute a score from a field via an expression.

Used for shipping fee scoring: a vehicle whose shipping cost is near the origin scores ~1.0; the score decays smoothly as the cost moves away. Real config: { origin: 295, path: "dealer.stateShippingMap.#USER_STATE#", scale: 500 }.

origin (ideal value) 295scale (decay distance) 500this vehicle's value 295

score

1.000

↑ the decay curve across all shipping-cost values (highlighted = this vehicle)

🧠 Test yourself

0/6 correct · 0/6 answered

1. A vehicle's shipping cost equals the Gauss `origin` exactly. What's its decay score?

2. In `near`, what does the `pivot` represent?

3. Which clause would you use to keep only vehicles built 2020–2025?

4. How do `should` clauses inside a `compound` combine their scores?

5. What's the difference between BoostScoreBlock and ConstantScoreBlock?

6. Why split scoring into HIGH_PRICE_NEW_CAR and HIGH_PRICE_USED_CAR using `equals`?

#How scoring works (the 30-second version)

Match clauses decide whether a vehicle qualifies for a score: equals, range, exists, text.

Score shapers decide how much it scores: Gauss decay, near, boost, constant, and function expressions.

A compound clause glues them together — its should list adds the pieces up.

#The clauses, hands-on

Use the tabs below. Each one mirrors a real block from the production score-weights config (the same JSON you can open under Search Score Weights).

🔔 Gauss decay — smooth "closer is better" curve (used for shipping fee).

🎯 near — proximity reward with a half-score pivot (newest year, lowest mileage).

✅ exists — reward vehicles that have a field (free-shipping flag, 360° spin).

📏 range — keep only values inside a [min, max] window.

🟰 equals — exact match; splits NEW vs USED scoring lanes.

⬆️ boost / constant — multiply relevance vs assign a fixed score.

🧩 compound — combine clauses; should scores add up.

🧮 function score — compute a score from a field via an expression.