How tools are scored and ranked
435 apps, tools, plugins, skills and MCP servers, one score out of 10, rules written in plain language. This page describes exactly what the code does, and nothing else. The weights below are read from the database at render time: they cannot diverge from the ones used in ranking.
1. The nAIvigate score
Each card has four sub-scores out of 10. Only one is a hand-set grade (utility); the other three are computed by the same code at import, at the weekly recheck and for the monthly edition:
- Utility : hand-set grade: what the tool really does for the group’s task, judged on use, not on marketing promise. Without a grade, the neutral value 5 is used.
- Reliability : computed. Base 5, then: +3 if the last activity or release is under one month old, +2 under three months, +1 under six, −1 beyond six, −3 beyond a year; −3 if the project is declared unmaintained; +1 for an official publisher; −2 if the website no longer responds; up to +2 for repository popularity (log scale of stars); 0 for a dead card.
- Security : numeric score when one exists; otherwise derived from the badge: green 10, orange 6, red 2, not assessed 4. The card then shows the badge state, never an invented 10/10. The badge reflects where data goes, permissions requested and documented caveats.
- Accessibility : computed: 40% install (nothing or one click 10, five minutes 7, technical 3), 30% required level (beginner 10, comfortable 7, technical 3), 30% price (free 10, freemium 7, paid 4).
The overall score is the weighted average of these four sub-scores, bounded between 0 and 10 and rounded to one decimal. Weights depend on the tool’s primary group; without a dedicated row, the default weights apply.
| Group | Utility | Reliability | Security | Accessibility |
|---|---|---|---|---|
| Default | 45 % | 20 % | 20 % | 15 % |
| Photo & imagedefault | 45 % | 20 % | 20 % | 15 % |
| Videodefault | 45 % | 20 % | 20 % | 15 % |
| Audio, voice & musicdefault | 45 % | 20 % | 20 % | 15 % |
| Design & slidesdefault | 45 % | 20 % | 20 % | 15 % |
| 3D & printingdefault | 45 % | 20 % | 20 % | 15 % |
| Write & translatedefault | 45 % | 20 % | 20 % | 15 % |
| Codedefault | 45 % | 20 % | 20 % | 15 % |
| Automation & agentsdefault | 45 % | 20 % | 20 % | 15 % |
| Data & spreadsheetsdefault | 45 % | 20 % | 20 % | 15 % |
| Office & emaildefault | 45 % | 20 % | 20 % | 15 % |
| Meetings & notesdefault | 45 % | 20 % | 20 % | 15 % |
| Research & monitoringdefault | 45 % | 20 % | 20 % | 15 % |
| Learndefault | 45 % | 20 % | 20 % | 15 % |
| Autonomous assistantsdefault | 45 % | 20 % | 20 % | 15 % |
| Run locallydefault | 45 % | 20 % | 20 % | 15 % |
| Orchestratedefault | 45 % | 20 % | 20 % | 15 % |
| Route & savedefault | 45 % | 20 % | 20 % | 15 % |
| Feed (memory, context)default | 45 % | 20 % | 20 % | 15 % |
| Frame (methods, guardrails)default | 45 % | 20 % | 20 % | 15 % |
| Monitordefault | 45 % | 20 % | 20 % | 15 % |
2. Ranking in the Finder
The Finder filters first (group, level, budget), then ranks by overall score with four adjustments, all fixed and visible in the code:
- +0.8 point when the requested group is the tool’s primary group (a tool linked as secondary stays in the list but only takes a card if it truly dominates).
- +0.5 point for one-click install when you declare yourself a beginner.
- +0.3 point for a free tool when you ask for “free and open source only”.
- −1 point for a serious security caveat (red badge).
The three cards are roles, not a podium: “Most powerful” (best utility), “Simplest” (lightest install and level), then “Best free” if n°1 is not already free, otherwise “The alternative” (another tool type with a close score). The rest of the list follows the adjusted score order.
3. Kit mode
My AI toolbox uses the same engine, task by task, with three extra rules: a tool already picked is reused when it covers another task; outside “Run your AIs” and outside the technical level, only apps, built-in modes and plugins are offered (tools, MCP and skills are building blocks for an AI, not for a person); with a budget (or “no limit”), a paid tool, or a freemium’s paid plan, replaces the free pick only if it has at least 1 more utility point and the total fits the budget; the free option stays shown on the row. €0 budget = strictly free. If the total still exceeds, the most expensive paid slot is replaced by the best free or freemium option. A card only enters a kit if it is enriched and its price is readable, or it has a free tier.
4. What is checked, and when
Every Sunday a script rechecks website, repository, last release and activity of every card; an unreachable card becomes “to recheck”, then “stale”, then leaves the Finder. Detailed texts (description, features, tips, pitfalls, install) are written by a language model from official sources then reviewed in batches; it never writes a score or a security badge. Prices are those displayed by the vendor at the verification date shown on the card.
5. What is not done
- No placement is sold, not in the ranking, not in the cards, not in the monthly edition.
- No language model is involved in ranking or in the kit: same question, same answer, today and in a month if the database has not changed.
- No security guarantee is given: a green badge means “no caveat on record”, not “risk-free”.
An error on a card? The contact link is in the footer: a verified correction is published at the next recheck.