> For the complete documentation index, see [llms.txt](https://docs.tachyo.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.tachyo.ai/under-the-hood/scoring.md). # How wallets are scored ## The short version Every night, the platform recomputes a score for every wallet in its universe from that wallet's own track record, then bins it into a tier. Weekly, the model behind the top of the ladder is retrained. Nothing about a tier is hand-assigned. ## The cascade Not every wallet has the same evidence behind it, so scoring runs as a cascade and the **confidence attached to a tier reflects which path it took**: | Evidence | What happens | Ceiling | | ----------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------ | --------------------- | | **Full history**: deep, reconciled trade record | Scored by the model. Highest confidence. | Any tier, including S | | **Closed trades only**: a real but shallower record | Scored on trade features. Confidence climbs with depth: it starts at the minimum around five closed trades and reaches its ceiling around twenty-five. | Below S | | **Balances and PnL only**: position data without a usable trade history | A deliberately conservative percentile score. | Capped below S | | **Nothing usable** | **unrated**. No letter is invented. | n/a | Two consequences worth internalising: * **Thin data can never produce an S.** The top band is reserved for wallets the model can actually stand behind, which is why S-tier counts on a token are worth reading closely. * **unrated is a statement about our evidence, not about the wallet.** See [Wallet tiers](/concepts/wallet-tiers.md). ## Per chain, not global Scores are computed **per chain**. The same wallet carries different tiers on different networks because skill is local. See [Scoring is per chain](/concepts/per-chain-scoring.md). ## What the backtest says The number the platform is willing to be judged on: > **S-tier (the top \~0.1% of scored wallets) was profitable over the following 30 days in 73% of cases, against 41% for wallets generally.** Measured **as-of**, with **no look-ahead**, averaged across weekly cutoffs. As-of means each wallet is judged using only what was knowable at that cutoff, then evaluated on what happened after: the only way to test a score that is worth anything. A backtest that scores wallets using the full history and then congratulates itself on picking the winners is measuring hindsight. That \~1.8x gap over an unscored wallet is what the product is selling. It is not a guarantee about any single wallet, and 73% means 27% of S-tier wallets were not profitable in a given 30-day window. ## What is deliberately not in the score * **Follower counts and social reach.** Tested against forward returns and found to carry effectively no signal, so the platform refuses to use them, even though they would make the numbers look better populated. * **Single spectacular trades.** One outsized win next to four nothing trades is small- sample noise. The minimum-evidence rule exists precisely to keep it from producing a letter. * **Anything that has not closed.** A wallet holding a paper 40x has not proven anything yet, which is why realized results and total PnL are shown separately. ## How to sanity-check it yourself Take a wallet you already have an opinion about and look at its tier, its trade count, and its record on the chain you care about. Then use [Compare](/platform/compare.md) against a wallet you rate differently. If the model disagrees with you consistently in a way you can explain, that is worth sending through [Feedback](/platform/feedback.md). A meaningful part of what the scoring does today started as exactly that kind of note. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.tachyo.ai/under-the-hood/scoring.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.