JoshiOnChain

under the price, day 4 BTC dipped to the 73,400 area this evening and a $10.7M liquidation print marks the spot. price is back at 74,250 now. a $38.4M bid wall sits at 73,200, 1.4% under price. the book within 1% is 52% bid, $116.4M against $107.1M. big trades over the last hour netted $1.08M of buying.

under the price, day 3 BTC is down 2.12% on the day, but ETH is down 5.40% and XRP 6.04%. the rest of the market fell harder than bitcoin did. the BTC book is dead flat, $98.0M of bids against $98.3M of asks within 1%. the nearest bid wall is $28.0M at 73,500, 1.1% under price. big trades over the last hour netted out to $216k of selling. that is almost nothing next to the $7.24M long forced out on the low earlier.

under the price, day 2 the book is dead even around 8pm IST. $107.9M of bids against $103.7M of asks within 1% of BTC, 51% bid. the only wall left on the map is a $32.7M bid at 74,900, 1.4% under price, parked there more than two hours. in the last hour big trades sold 16 times for $6.73M and bought 17 times for $3.72M. more size sold, with price still sitting near 76,000.

thousands of stops at one obvious level is a pile of liquidity. price goes after that pile first, then moves on

under the price, day 1 one wall has owned BTC all evening. more than $30M of asks parked near 75,800, there for over four hours. in the last hour big trades sold 22 times against 8 buys, and every red drop on the chart is a long that got forced out on the way down. posting what this chart shows, every day.

question for anyone trading with AI in the loop. what is the single thing you still wont let it do? mine is anything that touches money without me clicking the button myself.

the agent race wont be won on benchmarks. it wins when people trust a model inside a browser logged into their real accounts. capability is table stakes now. restraint is the actual product.

building trading tools with agents taught me one main thing. models write great code but have no idea which number is real. so every number on my chart points to its source. that rule was for me before it was for users.

proof of reserves tells you the money was there at one moment. it says nothing about who holds a key to it the moment after. after this week i care a lot more about the second thing.

a lot of my best claude code sessions end with me typing "no, smaller". the model will happily build you a cathedral. the skill isn't prompting anymore, it's knowing when to say stop.

every AI agent token i look at has the same shape. the agent is a thin wrapper and the token is the actual product. flip that and you have a business. most of them won't, because the token is easier to ship.

we spent ten years teaching people not to paste a seed phrase into a website. now we hand agents our logged-in browser and our whole codebase and call it productivity. same lesson, bigger blast radius. the permission prompt is the new seed phrase, read it like one.

your indicator isn't late because it's bad. it's late because it reads price, and price is the last thing to move. the book moves first, the liquidations move second, and the candle just writes down what already happened.

US midterms today and the timeline is pricing it like a coin flip with leverage. the people who get hurt tonight won't be the ones who called it wrong. it'll be the ones who sized it like they were sure.

the most useful thing my agent does isn't writing code. it refuses to post a number it can't trace back to a source. everyone's building AI that sounds confident. the edge is in the one that knows what it doesn't know.