1We ask whether a compact, transferable, multi-resolution model can learn minute-level structure that improves2calibrated probabilistic forecasts of the next 5, 15, 30 and 60 minutes of U.S. equity and ETF returns. We build a3Multi-Resolution Patch Transformer that reads the last 390 one-minute bars together with five days of 5-minute and4twenty days of 30-minute past-only aggregates, emits Student-$t$ marginals for each of the next sixty minutes and,5in a second stage, a low-rank Gaussian copula over them. The model is trained and evaluated on \num{234} million6one-minute bars of 300 instruments (2018--2025) from a private FirstRate Data lakehouse, selected point-in-time and7audited for timestamp, session and corporate-action conventions, against ten baselines sharing the same8information set, with CRPS, interval and calibration scores, trading-day cluster inference and overlap-aware9Diebold--Mariano tests on about one million evaluation origins per year, sixty held-out instruments and an untouched10final year. Every model beats a trailing-scale random walk by 3.5--4.9\,\% of CRPS; the gain is a volatility gain11with no directional or economic content. Sequence models lead at five minutes; a HAR-type variance regression with12a kurtosis-matched Student-$t$ leads and is better calibrated from fifteen minutes on. The multi-resolution context13adds at most 0.1\,\% in sample and nothing out of sample, model size (3.5 vs 20.7 million parameters) does not14matter, regime conditioning and coherent path sampling add nothing because future minute returns are conditionally15independent given the recent history, and pooled training transfers to unseen instruments without penalty. A model16that reads only the trailing volatility and the clock recovers nine tenths of the full model's skill. We release17the code, the protocol, synthetic fixtures and these null results.18