The rarity table is open
The first evening of the Discord alert channels produced exactly the question the channels were built to provoke. An A330-300 landed in #epic-sky, and the first reply was: is an A330-300 really epic?
Probably not. Roughly a sixth of the tiers in the Skydex reference were set by hand. The rest were inherited from an older dataset nobody had audited, and some are inverted, with a routine widebody at epic while its neo sibling sits at common. Until today the only place you could see a tier was one type page at a time, and the only place to argue about it was a chat message that scrolls away by morning.
Now the whole table is public: github.com/SkyDexOnline/skydex-taxonomy.
What is in the repository
data/aircraft-types.jsonis the curated table: ICAO designator, model
name, manufacturer, typical seats, range, engines, and rarity where a person has set it. This is the file pull requests change.
data/tiers.jsonis the effective tier of every type production knows
about, regenerated from the live reference on every publish, with a source on each row: curated or dump. That word tells you whether anyone has looked at the tier yet.
rarity.mdexplains what the six tiers mean and what they don't. A tier
answers one question: how often does a spotter inside the network's coverage actually see this type overhead? Not how many were built, not how interesting the aircraft is. The 747-400 is epic today because almost none fly passengers any more, not because Boeing built few of them.
data/legends.jsonholds the ten sky legends, which are a separate axis. A
legend is about the event of seeing one; a tier is about frequency.
data/aliases.jsonholds the markers ADS-B feeds emit when they don't know
a type, plus the invented family keys the old dataset used. Rows under those keys are rejected automatically.
- A validator runs on every pull request. There are issue forms for "this tier is
wrong" and "this type is missing", and the licence is CC BY 4.0: take the data, credit Skydex, no share-alike strings.
How a change travels
The repository is a mirror and an inbox, not the source of truth. Every merge to the Skydex application republishes these files, so what you read there is what production runs. Proposals come in as issues and pull requests, a maintainer replies, and an agreed change is ported upstream by hand. Once merged, the new tier reaches the reference within half an hour. Cards you already own keep the tier you were shown; a lowered tier only affects future catches.
Two things link into the repository from the product. Every type page on skydex.online prints its tier with a "why this tier?" link that opens a prefilled issue for that designator, and every alert card in #sky-alerts and #epic-sky carries the same link. The card that made you frown is two clicks from the thread where you say so.
What is not open
The list of light-aviation types whose airframes Skydex shows anonymously is a privacy rule and stays out of the repository. So does the airline registry, and any per-country policy. Military types carry a tier like everything else, but they never trigger an alert and can't be caught.
What comes next
The receiver network already records, per day and per 0.5° cell, which airframes were heard. Turning that into sightings per week per type per region is the plan for putting a measured number next to every tier, so a tier debate can be about the number instead of about two people's memories. Until then a tier is a judgement, and the repository is where it gets questioned.