Transparency
How we calculate the Stack Score
Every tool in the catalog gets a Stack Score from 0–100. It’s the number you see in the top-right corner of every card and detail page. Below is the exact weighted formula that produces it, the six components that feed it, and the biases v2 was built to fix.
The formula
The Stack Score is a weighted average of six components, each scored on its own 0–100 sub-scale. No single component can dominate. Editor judgment is still in the formula — but it’s a named slot with a capped weight, not the whole answer.
The result is clamped to 0–100 and rounded to an integer. Weights sum to 1.00 — adjusting one means adjusting another. The current weights were set when we shipped v2 on 2026-05-25.
The six components
Role-Fit · weight 0.30
The highest of the five role scores we maintain per tool: bookkeeper, solo agency, e-com operator, freelancer, ops manager. A tool that’s excellent for one role is more useful in a role-based directory than a tool that’s mediocre for all five — so the max wins, not the average.
Editorial Quality · weight 0.20
The editor’s hands-on opinion of the tool. This is where “I’ve actually used it and it’s great” lives. Capped at 20% so it can’t drown out the structured criteria.
Pricing Transparency · weight 0.15
Computed from the data on file: starts at 50, adds 20 for a visible starting price, 15 for a free trial or free tier, and 15 for at least two named pricing tiers. Tools that hide pricing behind “Contact sales” legitimately score lower here — even when we love them.
Trust & Freshness · weight 0.15
Rewards keeping tool data current: starts at 40, adds 30 if we re-verified pricing in the last 90 days, 20 if we ran a full editorial pass in the last 180 days, and 10 if the tool is verified. A tool whose data goes stale loses points automatically until we refresh it — this is how the “no stale listings” pledge gets enforced in code, not just copy.
Differentiation · weight 0.10
A 0–100 editorial rating of what this tool brings that the alternatives don’t. Stored with a one-line written justification so the editor can’t hand-wave.
Compliance & Trust Signals · weight 0.10
Computed from the data on file: 15 points per compliance tag (SOC 2, GDPR, HIPAA, ISO 27001, PCI-DSS, FedRAMP), 25 if the tool has a public API, and 25 if it has five or more named integrations. Capped at 100.
What the bands mean
Best-in-class. We’d default to this for our own work. Rare by design — most tools we love land in the 80s.
Strong choice. Specific advantages over alternatives. The fat middle of our catalog lives here.
Solid. Good fit for some users, not all.
Worth knowing about. Niche or has meaningful trade-offs.
Wouldn’t typically be in our catalog. If you see one, we’re tracking it for a reason (often as a comparison point or because it used to score higher).
What changed in v2
v1 was a single editorial number with no published weights. v2 is the weighted formula above. We rebuilt scoring because the v1 number had no statistical relationship to the per-role data we were already collecting — a tool could score 92 while genuinely fitting only one of the five roles at the 70+ level.
When we ran v2 in shadow mode against the catalog, the corrections were exactly what you’d expect: single-role specialists like Cursor moved from 92 to 82, frontier LLMs like Claude held near the top (91), and mid-tier productivity tools that v1 systematically under-rated moved up into the 70s. The full audit is on file.
Known biases we’re still working on
v2 fixed the “no formula at all” problem. These biases remain, and we’re open about them:
English-language testing
We test tools primarily in English. Multilingual-strong tools (Qwen, Mistral, AIxploria) may score lower for non-English users than they deserve. The Role-Fit and Editorial Quality components are both downstream of this.
North American workflows
Our five roles are tuned to North American practices — QuickBooks-shaped bookkeeping, US/Canadian e-com platforms, etc. Tools strong in EU/APAC markets may score lower than they deserve for those audiences.
Editor tilt
The Editorial Quality and Differentiation components are still set by the editor and reflect what the editor reaches for daily. The cap on Editorial Quality (20% of the score) limits the damage, but doesn’t eliminate it.
How to challenge a score
Every component is auditable. If a score looks wrong, tell us which component you think we got wrong and why:
- Email [email protected] with the tool name, the component you’re disputing, and your reasoning.
- Material changes get published in the newsletter with the before/after numbers and our reasoning.
Methodology v2 — launched 2026-05-25. Replaces v1’s single editorial number with the weighted formula above.