How Shumi Traded the Spring 2026 Rally

The trades

Top winners: ENA +49%, TIA +33%, OP +24%, COMP +21%, JUP +18%.

Biggest losses: INJ −24%, SOL −23%, DOT −20%, PENGU −20%, ETC −18%.

Twelve trades exited on a clean reversal signal. Every one of those twelve was a winner. The other eleven hit the maximum hold without the bounce arriving. Two of those eleven still timed out green (ALGO +17%, SKY +6%); the other nine timed out red. Zero trades hit the 30% emergency stop. The 30-day cooldown blocks re-entry on coins that just exited.

Five winners carried most of the upside. A handful of trades pays for the rest of the ledger. That's normal for this kind of system. Full table sits in the panel.

Why no shorts

Same strategy with shorts allowed, same window: 46 short positions still open and underwater on May 7. One ZEC short, down 143%.

The trend gate blocked all of them. That's what it's there for. Mean-reversion shorts in a real rally don't snap back. The stretched move keeps stretching, with no clean exit catalyst. The gate reads the bigger trend first and refuses shorts when the trend is up.

What Shumi sends each morning

One of the surfaces Shumi will speak through is Telegram. A daily brief at 09:00 Bangkok with the trend status, every open position, and the watch list. Same format every day. (Public Telegram channel coming soon.)

Below is the brief from May 1. Eight open positions. Seven up, one down (WLD). Shumi also runs a separate trend-following strategy; the one trend-following position (LINK) is visible at the bottom of the message but isn't part of this article. Seven of the eight mean-reversion positions are still open today.

The nine tests

The strategy has to clear nine specific tests before any change ships. Each one looks for a different way the result could be fake.

• Beats random ✓. The strategy's results beat random entries by enough to call it real edge.
• Reproducibility ✓. Re-running on the same data produces identical trades. No hidden randomness.
• Cross-strategy sanity ✓. The mean-reversion strategy and the trend-following strategy agree on the basics where they should.
• Correlation in stress ✓. When the market falls, Shumi's holdings move together the way the design predicts.
• Data freshness ✓. Every market input is current, nothing stale.
• Runnability ✓. Every coin in the universe can actually be processed end-to-end.
• Parameter sensitivity ⚠. A known soft spot: small changes to one timing setting move results more than we'd like.
• Entry-timing tolerance ⚠. The strategy survives small jitter in when bars arrive. We haven't closed this out as a non-issue.
• Held-out test ⚠. The big one. A clean slice of data Shumi has never seen, run blind. Queued. Not yet shipped.

What's public, what stays private, and what you can do with each

Public. Enough to audit the method. The strategy type (mean reversion with a trend gate). The four exit paths and the 30% stop. The 79-coin universe. The nine tests, with which passed and which didn't. Every trade in the panel: real entry and exit dates, real Hyperliquid prices, real percentage returns. A careful reader can use this to check whether the structure is reasonable, whether the tests catch the right failure modes, and whether the trade ledger is consistent with the strategy description.

Private. The specific timing thresholds. What counts as "stretched-down enough" to buy. What counts as "trend is up." The exact ranking weights when multiple coins qualify. These numbers came out of 400+ experiments, and they're the thing that competes the edge to zero if published. Without them, a clean-room rebuild would land in approximately the right neighborhood and then have to do its own optimization from scratch, which is the work we did.