How scores work: no percentage until 10 weighted reports
Right now, no build qualifies for a percentage — every build you see on this site is below the threshold, and that's exactly what the site will tell you. We'd rather show nothing than something wrong. This page explains every rule the numbers follow, because a score you can't audit is just a vibe with a percent sign.
The confidence ladder
Evidence unlocks detail. Nothing skips a rung.
Why so strict? Showing a confident number off 3 reports is the single mistake that would end this project. The threshold is weighted evidence, not raw count — see the weights below.
The formula and the weights
score = (flawless + 0.5 × glitchy) / total — glitchy counts half, bricked counts zero. Reports are weighted before they're summed:
- Verified (screenshot submission) ×2 · community-sourced (collected from a public post) ×1
- Time on the firmware: a week or more ×1.5 · a few days ×1.0 · just updated ×0.5
A report that only says a version was mentioned — without the owner clearly running it — is stored as weak evidence and never moves a number.
Why scores are per model line
The same version string behaves differently on different hardware, and we can prove it: firmware 2.3.27.20 destroyed task areas on LUBA 3 — one owner lost 34 zones, unrecovered by reboot, reinstall or cloud restore — while YUKA mini 2 owners reported the same build as an improvement. Blend those and you get roughly 60%: a number that is true of nothing and useful to nobody. So a score is always one model line on one build, never a build alone. The rollout dates differed too, so they weren't even the same binary.
Where the reports come from
The site launched with 218 reports collected from public Facebook groups, Reddit, GitHub issues and owner forums — in English, German and French. Each is labelled community-sourced, classified by verdict, and carries a link to the original post so you can check our reading against the source. Screenshot submissions from owners are labelled verified and weigh double, because the firmware version comes off the Device Information screen rather than a prose mention.
Quotes are verbatim — with one editorial mark
Excerpts are byte-for-byte what the owner wrote, umlauts and all. The single exception: where a post is signed with a real name, the name is replaced in place by […] and the edit is recorded alongside the report — nothing else in a quote is ever altered. The judgement call, stated plainly: real given and family names are redacted; platform handles are retained. A handle like wise3 was chosen by its author for public posting, and posts that credit each other are incoherent without them.
The bias we correct for
Complaint bias is real, and a community member said it better than we could: “Reddit can be a bit of an echo chamber of negativity — people don't report anything when they don't have problems.” That's why we actively ask for flawless reports too. A “mowed normally, nothing to see” is data — often the most valuable kind.
Unlisted builds
7 builds owners are running don't appear in Mammotion's release notes at all. We record them and flag them “unlisted” rather than reject them — one of them appears to fix a turning regression owners complained about for 13 months, and another stopped a mower passing its self-test. An unknown version is information, not an error.
Known limits
- 16 reports on 1.15.21.1 name the version but not the model. They can't score anything — we won't guess which mower a stranger owns — so they're shown as context instead.
- Mammotion sometimes ships app and firmware updates together, so a few reported changes may be the app, not the firmware. Where an owner said so, we recorded it that way.
- Rollouts are staggered — two identical mowers can be offered different builds on the same morning — so “the latest version” is not one thing, and we track what owners were offered separately from what they run.
Questions this page doesn't answer? Email hello@mammostats.comand we'll add them here.