A short field guide to the labels you see on every deck — and what to do when you disagree with one.
Each deck on SorceryRec gets one of four archetype labels:
If the heuristic can't make a confident pick, the label is simply — (Unknown) instead of laundering uncertain decks into Midrange.
On top of the primary archetype, a deck can carry up to three overlays — Sorcery-native flavours that explain how the deck actually contests the realm without replacing the macro label:
That's why a deck reads as Aggro · Burn or Control · Site-centric: the primary tells you the plan, the overlay tells you the flavour.
The label isn't guessed — it's computed from your deck's shape:
Each archetype gets a score between 0 and 1; the highest-scoring archetype wins, but only if it clears a confidence floor. Otherwise the label is Unknown. Confidence is shown in the score breakdown on each deck's detail page.
The label you see comes from the first source available, in this order:
Manual overrides display 100% confidence and a small badge indicating the source.
Every classified deck shows an I disagree link next to its label. Click it and a small modal lets you propose:
We don't flip the label automatically. Disputes go into a queue; an admin reviews each one and either accepts (which sets the deck's admin-archetype slot, overriding the heuristic going forward) or rejects (the heuristic stays). One person can't spam-vote a deck into a wrong label, and we don't aggregate proposals into a majority — the admin decides each time.
Anti-griefing: each user is rate-limited to 5 disputes in any 24-hour window, and a previously-rejected dispute on a deck doesn't auto-reopen if the same user files again.
The classifier reads from per-card tags grouped into three families: effect (what the card does), role (what it's for in a deck plan), and site effect (only for atlas Sites). Every card detail page shows its current tags — and an I disagree link to propose a different set, exactly the same flow as archetype disputes. Accepted card-tag changes update the tags everywhere they're used, which means the heuristic re-classifies every deck containing that card on its next render.
Same anti-griefing rules apply: 5 disputes per user per 24h, one pending dispute per user×card pair, no public dispute log.
We're honest about the limits:
Everything above is a summary. The full specification — every tag, every score component, every threshold, with worked examples — lives in our open repo:
docs/archetype-classification.md — versioned alongside the code, updated whenever weights move.
Found a tag-level or deck-level mistake? Get in touch, or just file a dispute on the deck in question.