The rules of the scoreboard

Methodology

Every list site tells you what exists. caniselfhostit tells you how to run it. We answer the question that costs you money and weekends: should you? For that answer to be worth anything, it has to be earned the hard way — so here is exactly how it's earned, and who is not allowed to influence it.

The rubric

A verdict is YES ("worth it"), KINDA ("worth it if…") or NOT REALLY ("keep paying"), computed from four inputs:

  1. Capability — the primary alternative covers ≥80% of the jobs people actually pay the SaaS for.
  2. Economics — break-even inside 18 months at $20/h for your time (every page has a slider — the derivation is the same code at every rate). We count VPS share, storage, and maintenance minutes. Hiding the time cost is how other people lie to you.
  3. Effort ceiling — YES requires the core workflow running in ≤2 measured hours; KINDA within one weekend. Past that, no software is good enough.
  4. Project health — a primary alternative with no commit in 12 months caps the verdict and gets a ⚠️ flag, automatically, nightly.

The rubric means a verdict can be argued against criteria — open a dispute with evidence. It also means we can say NOT REALLY about software that self-hosts beautifully. Capable but not worth your weekends is a verdict our format can express — and honest NOs are why you can trust the YESes.

Test protocol v1

  • Reference environment: Fresh Debian 12, Docker + Compose preinstalled, 2 GB RAM / 2 vCPU ($5-class VPS).
  • AI assistant allowed and stated — protocol v1 assumes Claude Code (or equivalent) as your sysadmin, because that's how people actually self-host in 2026.
  • Setup is timed, wall clock, empty directory → core workflow working. "47 minutes" on a page is a measurement, not vibes.
  • Every verdict ships its compose file and timed log. CI boots every compose file and checks its health endpoint — a broken install claim cannot merge.
  • Every verdict names who verified it and when. Verdicts older than 12 months dim and get a re-test issue.

Who can never touch a verdict

  • Votes. Display-only, forever. Brigading is inevitable; that's why the scoreboard and the verdict are separate systems.
  • Sponsors. Sponsorship buys a labeled slot, never a word of a verdict page's rubric sections.
  • Affiliate links. VPS links appear only below the math they help you act on, are labeled, and never affect rankings. (caniselfhostit is proudly "no ads, no affiliate links" — we're transparent instead of abstinent, and this page is the contract.)
  • Automation. Bots refresh metrics and propose PRs; humans merge every verdict change. The changelog is public.

Data & licenses

Capability seed (app → replacement pairings, pricing snapshots) from caniselfhostit (MIT, © Jashanpreet Singh), marked per-record with provenance. Our decision layer — verdicts, economics, timings, votes — is CC BY-SA 4.0: take it, credit it, keep it open. The full dataset is free at /api/apps.json and summarized in /llms.txt.