Hour-Aware Adaptive Risk Management for Autonomous Memecoin Trading on Solana DEXs: Evidence, Theory, and Design Lessons from a 15-Day Deployment

This paper reports a 15-day paper-traded deployment of an autonomous memecoin trading system on Solana decentralised exchanges, framed against three classical microstructure predictions (Kyle 1985 informed-flow; Precup-Sutton-Singh 2000 off-policy evaluation; Bailey-Lopez de Prado 2014 deflated-Sharpe). The 190-trade sample (29 March to 12 April 2026, UTC) shows a 40.5 percent win rate, mean per-trade return of +0.62 percent, cumulative +117.7 percent (net SOL +0.03873), population skewness -1.21, population excess kurtosis 6.61. Stratifying on the derived UTC entry hour (`date - heldMin`), a Mann-Whitney U test of three exploratorily identified poorest-performing hours (2, 13, 23; combined n=17, mean -11.60 percent) against all other hours (n=173, mean +1.82 percent) yields U = 1,345, p = 0.5634 (two-sided asymptotic with continuity correction), rank-biserial r = 0.085. The comparison is directional and non-confirmatory. A parallel counterfactual rejection-tracking system collected 4,874 forward-sample observations across 184 distinct rejection events; 33 of 184 events (17.9 percent) reached a 50 percent drawdown from reference within 24 hours, and 1,267 of 4,874 sample observations (26.0 percent) recorded the rejected token below half-reference. Removing the top three trades (1.6 percent of sample) flips cumulative return unprofitable; profitability rests on a small number of large winners and is structurally fragile. The manuscript derives three transferable design lessons for future autonomous-trading measurement studies on AMM venues: counterfactual logging as standard telemetry; fragility disclosure as a first-order manuscript deliverable; hour-of-day stratification as a required check on any pooled-trade analysis, with the choice of stratification timestamp (entry versus exit) declared explicitly and justified. Data and reproduction: the full trade log, rejection-sample corpus, four supporting SVG figures, per-filter provenance table (PROVENANCE.md), and an assertion-based reproduction script (audit.py) that exits zero if and only if every headline number in this paper reproduces from the deposited CSVs are archived in the companion dataset bundle (Zenodo concept DOI 10.5281/zenodo.20043301, current version DOI 10.5281/zenodo.21636182). Journal status: manuscript at Ledger (MIT Press / University of Pittsburgh), #632. Editorial pre-review decision 2026-07-27 with invitation to resubmit; this v3 preprint corresponds to the v4 manuscript that addresses the four editorial asks (theoretical framing, significance of contribution, generalisable design lessons, expanded literature positioning). Companion preprints: SSRN 6564803, arXiv 2606.08232. Competing interests: the trading system is the subject of pending U.S. Provisional Patent Application #64/099,108 (filed 2026-06-25, Micro Entity). The verbatim exit-trigger taxonomy, which embedded per-tier threshold values, was disclosed in the companion dataset v1.0 (Zenodo v1.0 version DOI 10.5281/zenodo.20043302) and is removed from the companion dataset v1.1 onward; per-tier thresholds, scoring weights, circuit-breaker triggering rules, scanner polling intervals, and per-filter category mapping remain undisclosed.