Every number on this site is derived from observed Discord trade messages. No editor sets prices. No hand-tuned multipliers. Our anchor is Mythical Fruit Chest = 10,000; every other item's value is worked out from its trade ratios.
A bot reads every message from seven opt-in Discord trading channels (trade_1, trade_2, booster, premium, certified, winloss, rumbling) and saves (timestamp, author_id, content). ~1M messages parsed.
Emojis resolve to a 237-item dictionary; text matches against ~970 aliases (PCC, ICC, ASE, CQASE, PCHAK). Two dedup layers kill spam: 60-second bump filter (same author reposting within a minute counts once) and per-user-per-day cap (each author/message pair counts once per calendar day).
Obvious lowball posts (tier letters ≥2 apart with no compensating mods, pure 1:1 cross-tier trades) are dropped before math runs.
Dijkstra ratio chains propagate values from the MFC anchor outward along the strongest trade paths. Multi-item bundle trades then refine: for each item, we take the weighted percentile of the implied values from bundles where it's alone on one side.
Each trade carries two weights: 12-hour half-life (short window, picks up fresh market shifts) and 30-day half-life (stable baseline). Per item we pick whichever is more trustworthy right now. Iterative updates use 35% EMA blending with a 30% per-round step cap to prevent oscillation.
Demand boost/discount of ±3–20% based on wanted-vs-offered ratio. Scarcity-aware 85th percentile for Collectables with thin bundle evidence. Learned +Adds value per trade-scale tier lets us include mod-bearing trades that were previously discarded. 1:1 alignment + chain-outlier cap catch items stuck at bad Dijkstra estimates.
Items with <200 trades get an orange LOW DATA badge. Items 200–999 trades are medium confidence. 1000+ trades are high confidence. The bands are purely trade-count driven with no subjective override.
+Adds and +Strong_Adds aren't priced.If you think a value is wrong, tell us in Discord and we'll show you which trades produced it. Eventually we'll publish a community-submission path with evidence review, per the backpack.tf model.