Trust methodology
How recommendations work
StackBuilder ranks tools with a daily-recomputed scoring formula built from real signals — never invented metrics. During beta, tool information should be treated as researched guidance, not a guarantee.
The scoring formula
Every tool's rank comes from one composite, recomputed daily from real evidence — nothing is invented, and nothing is guessed. The components and their weights: verification status 20% (verified, fresh, unverified, or stale, plus recency of the last verification check and whether the maker claimed the listing), freshness decay 15% (how recently the tool was actually verified, decaying with a 30-day half-life), stack usage 15% (how often the tool appears in public stacks and workflow guides), save momentum 10% (saves in the last 7 days), click quality 10% (outbound clicks in the last 7 days), trend velocity 10% (the daily trending score), pricing transparency 5% (whether the pricing model, starting price, pricing notes, and free-plan status are all disclosed), claim confidence 5% (a maker-claimed listing scores higher), category relevance 5% (listed under at least one category and one tag), and editorial fit 5% (set by a human reviewer — never by the scoring engine). A spam-risk penalty for thin or missing listing data is subtracted. Activity components are normalized against the current maximum, so scores stay relative as the catalog grows.
Rankings are not for sale
Money cannot buy rank — technically, not just as policy. The scoring function that computes every tool's score never reads sponsorship data: it does not reference who paid for a showcase because that field is never read by the scoring code. Rankings come only from the signals above.
Sponsored listings
Paid showcases exist to keep the lights on. They are always visibly labeled “Sponsored” in dedicated slots — never mixed into the ranked list, and never affecting a tool's score.
Fit labels, not raw scores
You will never see a raw score. Tools are presented with plain fit labels — Strong fit, Worth a look, Long shot — alongside the reason chips behind them (for example, “verified this month” or “saved by dozens of people this week”). The number is internal; the reasoning is visible.
Freshness signals
A tool is verified only after a real check — a successful link check or an editorial review — and we record when it happened. Freshness decays over time, so a tool verified a year ago does not outrank one verified last week. The freshness agent rechecks listings daily and flags stale ones for human review.
Stack usage
When a tool appears in public stacks and workflow guides, that is evidence it works alongside other tools in real workflows. The stack graph also powers alternatives and recommendations.
Corrections
Users and makers can submit corrections. Risky changes require human review before going live.