Paper-trading pilot

Every day, on the record.

Each row is one day's decision by Jev, recorded in the morning and scored over 24 hours, against the next morning’s entry. An evening checkpoint (UTC) is also shown for reference. Earlier days are never changed or removed, including the ones that didn't go well. A re-run on the same day corrects only that day's row.

Loading the record…

Daily loghistory.json ↗ · data CC BY 4.0
DateDecisionConfidence24-hour resultentry to next morning’s entryEvening checkpointentry to that evening’s price; not the experiment score

How to read this. Jev is given the price plus the recent trend, moving averages, momentum and volume, and answers buy, sell or hold. Each day is scored over 24 hours. The 24-hour result (the first column) compares that morning’s entry with the next morning’s entry, so the overnight move counts; the newest day stays pending until the next morning. The evening checkpoint compares the entry with a fixed snapshot of the price that evening. Solana trades around the clock, so this is not a market close, and it is not the experiment score: it covers a shorter window than the 24 hours Jev is asked about. For a HOLD, “avoided a loss” means the price fell and “missed a gain” means it rose. A buy or sell is scored win or loss on a one-SOL paper position, before fees. Confidence is the number Jev itself returns. The dashed line on the chart marks the move between one evening checkpoint and the next morning’s entry. Each morning’s decision is published here before the day’s close, as a row with no outcome yet. The evening only adds the outcome to that row, the publishing code refuses to change a decision once it is published, and any edit would show in the repository’s history, which is protected against rewrites. New decisions also record the Jev model version and a SHA-256 fingerprint of Jev’s raw answer (the small text under the confidence), so the private record can be checked against what was published. The question to Jev changed on 2026-10-03. Until 2026-10-02 it was “Should we buy, sell, or hold SOL right now?” From 2026-10-03 the question states that no position is held and that Jev may go long, go short or stay in cash over the next 24 hours. New rows carry a promptVersion field in history.json; earlier rows are unchanged.

This is a research pilot with paper money, not investment advice. Nothing here is a recommendation, and a short record says little about future results. The first days, marked "pilot start", were recorded before confidence and the price source were stored. Their prices fall within Kraken's real range for those days, but the source can't be confirmed.