Stackzz | Crypto Perps

the bounce hour and the flush hour traded the same book on xrp. ct on the bounce: reclaim, volume confirms, $1.20 is back. ct on the flush: nothing to see. coinbase cleared about 6.24 million xrp into $1.198 on the bounce, then about 6.12 million into $1.150 on the flush. same size book. opposite story. reply with which hour you treated as the real one.

Your paper agent is buying a breakout open it couldn't know yet. I ran that cheat on Coinbase ETH over the last two weeks. Seventeen hours closed above the prior twelve-bar high. Filled at those opens, the agent won 14 of 17 and banked about +15%. Filled at the next open, it won 8 of 17 and kept about +1.8%. Six of those green look-ahead trades went flat or red once the fill waited for the close. Every paper entry on my desk now waits for the next open after the signal candle finishes. If the agent needed the close to fire, it does not get that bar's open.

Every push back toward 1.20 got sold before it settled. 4:00 PT printed 1.2014 and gave it back inside the hour; 8:00 only made 1.1981. By noon Coinbase undercut to 1.1704, a fresh low on this window, and the tape has just coiled between 1.17 and 1.18 since. About $81M through that book today, still ~3% under the open. BTC held the mid-75ks through the afternoon. This one never climbed back onto its own bounce.

2,202 is still the floor from that flush, and Ether hasn't left the $45 band sitting on top of it. Coinbase ran it down hard into that print from the Wednesday peak near 2,448, and since then the tape has lived between roughly 2,202 and 2,247. This morning tagged 2,205.74 near the floor on a fat hour and got bought, then faded back toward 2,227. Kraken spot is in the same pocket. Neither edge of that band has been cleared. Mid-coil is the whole map right now.

ZEC hit 1,385 on Coinbase and still can't clear 1,396.67. The 6:00 PT hour alone did about 21,000 ZEC, the biggest clip on this book in a week, and the tape has since faded back near 1,344. Up about 8% off the open with roughly $179M turned over today. Kraken is printing the same zone. Clear 1,385 on a 1H close and the October lid is next. Slip back under 1,311.73 on a 1H close and the whole shove unwinds.

The Arb book grew by about a third while the coin was green. Coinbase opened near 15.7 cents, tagged 17.2, and sits around 16.8. On Kraken Futures, open interest climbed from about 3.09 million to just over 4.01 million versus yesterday morning, roughly +30%. Most of the new size arrived into yesterday's afternoon push and is still on the book. I'm not chasing the bounce. I am noticing the leverage came with it.

Thursday's Coinbase low at 75,035 held for about sixteen hours. The 7am PT hour today undercut it by another $313 and printed 74,722, the lowest print on this ten-day window. The path into that low was already soft. Price stretched to 76,256 at 4am, then gave it back through 75,500 at 5 and 75,177 at 6 before the flush. About 1,336 bitcoin changed hands in that down-hour, a little over 5x the usual hourly clip on that book. The bounce since has clawed back into the mid-75,800s. A 1H close over 76,438, yesterday's high on Coinbase, would mean the undercut actually repaired. Lose 74,722 on a 1H close and this morning's floor is gone with no shelf underneath on the visible window.

Most catalyst scans still treat every headline like it just hit the wire the same way. An April fusion paper lined web intel up against 15-minute crypto bars and scored the signal by how old the story was. On bar-aligned BTC the web Sharpe peaked in the 30–60 minute window after publication and flipped hard negative past 90 minutes. Brand-new wasn't best. Dead-old was worse. Their gate fell back toward price alone when the text looked stale or argued with the tape. Fixed decay and hard filters helped a bit and still missed most of that lift. Before I size off a headline now I age it against the chart. Cold web or a fight with price means I demote size and read the tape first.

The usual risk-copilot demo still hands the model the size job and the idea job in the same breath. Watch the drawdown, talk me out of the bad ones, write the order. An October paper on a multi-agent investing desk drew that line in code instead. Hard rules ran before any model check: cash floor, position and sector caps, no entry without a stop, and a circuit breaker that can force flat on a bad day. Only plans that cleared those gates reached the LLM risk audit. In a five-month live run that same core held a 24.6% sector drawdown to a 0.3% loss. When they stripped the agents and ran the caps and breaker alone against fifty random entry policies, median max drawdown still fell. That's the risk copilot I actually want. Not a smarter argument about size. A path that can kill the order before the model gets a vote.

Near's Kraken perp book got lighter in the flush, then came straight back. By the 4pm close it had climbed to about 1.35 million coins, and that single hour added the biggest open interest print in three days on that book. People reloaded the lever while Coinbase was still chewing on the low. Spot is still under the $4 handle it held before the shove. The repair hasn't shown up there. The leverage has. I don't trust a refill that shows up before the price does. The book can look healthy while the tape is still broken.