Methodology
How Decision Widget Recommends Products
Every recommendation on this site is generated by deterministic scoring rules applied to structured product data. No editorial opinions. No sponsored rankings. No mystery algorithms.
01 — Product Selection
Curated, not scraped
Each category starts with a hand-curated product universe. Only products that meet baseline quality, value, and availability thresholds are included in scoring — typically 8–15 products per category, not hundreds.
Deterministic
Given the same answers and the same product catalog, the same recommendation is produced every time. There is no randomness, no A/B variation in results, and no editorial override.
Transparent
Scoring rules are defined per category with explicit weights. Each product earns points based on how well its attributes match your stated needs.
No Sponsored Rankings
No brand or product pays for placement. Affiliate commissions are earned after a recommendation is made — they never influence which product is recommended.
Curated Catalog
Each category starts with a hand-curated product universe. Only products that meet baseline quality and availability thresholds are included in scoring.
02 — Question Design
Five questions, maximum signal
Each category quiz contains five primary questions designed to discriminate between products. Questions target the attributes that matter most for that category — budget, use case, environment, and key preferences.
Every question option maps directly to scoring rules. If a question cannot split the product set meaningfully, it is replaced. The goal is maximum signal in minimum time — under 60 seconds for most users.
03 — Scoring Rules
Deterministic, not probabilistic
When you answer a category quiz, your responses are matched against a set of scoring rules defined for that category. Each rule ties a product attribute to a question answer and a numeric weight.
The scoring pipeline
- 1Filter. Products that violate hard constraints (e.g., exceeding your budget) are removed from the candidate set via allow-list rules.
- 2Score. Each remaining product is evaluated against every scoring rule. If a rule's conditions match (question answer + feature value), the rule's weight is added to the product's total score.
- 3Rank. Products are sorted by total score, highest first. Ties are broken alphabetically for consistency.
- 4Present. The top-scoring product is your best match. The next two become alternatives for comparison.
Attribute matching
Scoring rules support several match types to handle different kinds of product data:
| Match Type | Description | Example |
|---|---|---|
| eq | Exact value match | noise_level = "quiet" |
| gte / lte | Numeric comparison | battery_life ≥ 120 minutes |
| includes | Array contains value | floor_types includes "hardwood" |
| exists | Feature is present | has_hepa_filter = true |
Rules can also apply per-unit weights (e.g., +2 points per decibel below threshold), enabling nuanced scoring for continuous attributes.
04 — Affiliate Disclosure
Same commission, every product
Decision Widget is a participant in the Amazon Associates Program. When you click a product link and make a purchase, we may earn a commission at no additional cost to you.
Key point
Affiliate commissions are earned after a recommendation is generated. They are never a factor in the scoring algorithm. The same product would be recommended whether or not an affiliate relationship exists.
For full details, see our Disclosure page.
05 — Ongoing Maintenance
Products change. We update.
Product data is sourced from Amazon's catalog and verified periodically. Each product listing includes a link health check that detects unavailable or delisted items. Products with broken links are automatically excluded from scoring to prevent recommending unavailable products.
Category catalogs are reviewed and updated as new products enter the market. Pricing shown reflects approximate price points at time of catalog update and may differ from current Amazon pricing.
For Writers and Researchers
You are welcome to reference or cite this methodology in your own content. If you write about product recommendation systems, affiliate transparency, or deterministic scoring models, feel free to link to this page as a source.
Questions about our approach? Reach out via our contact page.