Desk Post-Trade Review

⚠️ **Trading warning / Avertissement trading** : informational only, not investment advice. A review explains what a finished trade cost and how it was run; it cannot make the next trade profitable, and a good outcome on a broken process is luck, not evidence. No review here is a promise of return. Curated by Skill Harbor: the journaling and review procedure of HyperGrok, Galleon Labs' seven-agent trading desk for Hyperliquid, written for the Trade Reviewer role. It has four parts. First, the journal: one append-only file per active day, every line with a UTC time and an id, entries for openings, risk decisions, approvals, sends, fills, closes, limit changes, incidents and notes, with corrections added as new lines and no opinions allowed. Second, the trade review, triggered when a proposal closes: it works from the exchange record first (the proposal file, fills for the window, historical orders and order status by cloid, funding paid), and computes entry and exit slippage in basis points, fees in dollars and as a share of notional with maker and taker noted, funding over the holding window, the net result in dollars and in R, whether a reduce-only stop rested on the exchange for the entire life of the position, lifecycle completeness, and holding time. The grade keeps two columns strictly apart: process (clean, minor break or major break, with the stage named; any send without a PASS or an approval by id, any missing protection, any resend on an unknown result or any breached limit is a major break) and outcome, stated in R and dollars without adjectives. Each review ends with exactly one repeatable finding. Third, a weekly desk review rolls up counts, hit rate, expectancy in R, largest loss and drawdown, costs as a share of gross PnL, process breaks and limit changes, with no trade recommendations. Fourth, incident reviews are blameless in tone and exact in fact. From the galleonlabs/hypergrok-trading-desk repository (MIT). Honest caveats: the review is only as good as the fills and journal lines it reads, chat memory is explicitly not a source, and counterfactuals about holding longer are excluded as non-evidence. Skill Harbor never reviews the code, review it yourself before use.