Methodology

• Each track record is graded on its own rules, shown below, and kept separate from the others.
• Results are graded after the fact against prices we record. Past results are never edited or removed.
• Generated content (reports, predictions, calendar events, imported filings) must pass automated checks before it is shown: stale dates, completed events described as upcoming, claims that contradict the item's own figures, implausible prices or units, and duplicate or impossible schedules. Items that fail are withheld, not deleted, and re-checked daily.
• Nothing on Atradian is personal investment advice or an instruction to trade.

• Deterministic: reconciliation, valuation, weights, rule states, attribution and scenarios are computed by fixed code with regression tests, never by a language model. AI only reads screenshots (you confirm every field) and explains a finished brief; an explanation that contains a figure not in the brief is rejected.
• States: PASS, WATCH (close to a limit), REVIEW (a rule tripped), BLOCKED (a restriction on new additions is active) and UNKNOWN (a required input is missing or stale — never shown as green).
• Freshness: daily market series older than two sessions are stale; a stale input makes the rules that depend on it UNKNOWN. Every claim shows its source and observation date.
• History:the cash-flow-adjusted unit value is built forward from your accepted snapshots only; today's holdings are never backcast into a fictional past.
• Attribution: a move is company-specific only when a dated company event (SEC filing, results, guidance, material news) exists; otherwise it tracked its benchmark or is marked unexplained.
• Report-only: the brief cannot place, open, close or recommend an order.

Universe: about 1,400 stocks across US, ASX, LSE, TSX and XETRA, plus the top crypto assets, rebuilt nightly. Inputs:four 0–100 sub-scores. Technical (35%) covers trend vs the 50- and 200-day averages, 20- and 63-day momentum, and distance from the 52-week high. Sentiment (20%) is the tone of the last 7 days of news. Fundamental (25%) combines the latest research classification, earnings trend, analyst consensus and EPS surprises. Smart money (20%) is insider buys vs sells plus unusual options for stocks, and perpetual-futures open-interest change for crypto. A missing sub-score counts as neutral (50), so thin evidence can't produce an extreme score.

Grading:each score's return over the next 30 calendar days vs its home-market benchmark, in local currency. Benchmarks: S&P 500 (SPY) for US, ASX 200 for ASX, FTSE 100 for LSE, S&P/TSX Composite for TSX, DAX for XETRA, BTC for crypto. Scores began on 2026-09-25, so the first grades appear from about 2026-10-25. No fees apply because nothing is traded. Results are on the Scores page.

Every Sunday the AI panel publishes a direction, a percentage range and a confidence for a fixed list of symbols. Grading:the move runs from the latest close on or before the week's Monday to the latest close at week-end. A call is in range if that move lands inside the printed range, boundaries included. This is the headline rate. Direction is graded separately: bullish needs more than +0.5%, bearish less than −0.5%, neutral within ±1.5%. The chart compares these results with a naive trailing-momentum baseline. See Predictions.

Every BUY signal each strategy produces is logged whether or not it was traded. A signal counts as scored once its 72-hour forward return is known, and a win is a positive 72-hour return. Figures cover the trailing 60 days. The “all” row is the exact sum of the per-asset/per-regime rows. Averages are weighted by the number of scored signals. Rejected and underperforming strategies stay visible.

These are simulations only: no orders are placed. Qualifying wallets are chosen weekly on a minimum resolved sample and win rate. Each copy is sized at 5% of the paper book's equity, not at the source wallet's size, and Hyperliquid leverage is capped at 10×. Fees, funding and slippage are not modelled, and copies open at the source wallet's price. Open positions are marked hourly (Polymarket outcome price, Hyperliquid mid), and equity is cash plus position value. Hyperliquid losses are floored at the position's margin. The full assumptions are listed on each page.

Reports are generated by a multi-step AI pipeline (models and method version are shown on each report) using Finnhub, Financial Modeling Prep and exchange price data plus dated news headlines. The prompt includes the current date. Prices and targets are as at publication, in the listing's currency. Every report shows its sources and passes the checks under “Principles” before it is listed. Report Q&A rankings (for example “highest upside”) are computed from the stored numbers, not by the AI.

Market and macro data are fetched on a schedule, but each source publishes on its own calendar. Every macro card shows its series ID, observation date, retrieval date and normal publication lag, and flags a series that has stopped updating. Economic-calendar events link to their source. FOMC dates come only from the Federal Reserve's published schedule. 13F holdings come from the most recent SEC filing, which can be up to 135 days old, and older filings are flagged.

Atradian is research software, and scores and reports are produced by the automated systems described above, not hand-picked. Atradian's operator also runs an automated trading system that may hold positions in securities covered on the site. No fee is received from any issuer for coverage. Questions: support@atradian.com.