Phasma (@0xPhasma) header
Phasma (@0xPhasma) profile picture

tested AI tools. kept the best. now i post them.

Joined June 2024
Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)Photo by Phasma (@0xPhasma)
Wei Jianming, 34, Shenzhen. Former quant at a mid-tier futures house, fired in 2023 for "unauthorized algorithmic activity" on company servers. nobody knew what he'd actually built. he spent 14 months in a rented office on the 11th floor of a tech park in Longhua, two monitors running 24/7, a stock screener pulling the full A-share universe — all 5,200 names — sorted by momentum signal every 90 seconds. the setup looked like a bloomberg terminal cosplay. it wasn't. he'd wired claude into a custom pipeline that read the momentum board, cross-referenced regulatory filings in real time, and flagged divergence between retail heat and institutional flow data. the "主力净量" column — net institutional volume — was the input most traders ignored. wei made it the only signal that mattered. entry rules were simple: stock appears in top 30 by 人气 (retail attention), institutional flow turns positive within same session, price hasn't moved more than 4% yet. that's the window. that's the whole edge. from january to april 2026 the system logged 847 trades. win rate: 61%. average hold: 2.3 sessions. total return on the 40万 account: +340%. someone on a quant forum posted a screenshot of his brokerage statement in march. the thread got 2,400 replies before it was deleted. wei's response when people asked him about it: "the market tells you everything. you just have to stop looking at price." he's now running 280万 of external capital from three people who found him through that deleted thread. the office hasn't changed. still two screens. still 11th floor.
2
9
92
Phasma retweeted
wei zhang, 24, chengdu. nobody knew his name until three weeks ago when someone in a quant discord screenshotted his phone screen mid-stream. he was doing a casual livestream about xau/usd setups. 1h chart open on his phone, oanda data feed running in the background on his laptop. standard stuff. the kind of content that gets 200 views. but someone paused at 4:37 and zoomed into the laptop screen behind him. the numbers didn't match a hobbyist. the laptop was running a custom dashboard not tradingview, not mt4. something built. position log visible in the corner showed 23 closed trades that month. avg hold: 4.1 hours. win rate column: 78%. people started digging. his github had been dormant for two years. then six months ago, a single private repo went public by accident for 11 minutes before he caught it. someone had already cloned it. the repo was a claude-powered trade assistant that scraped multi-timeframe confluence signals from xau/usd, fed them into a structured prompt chain, and returned entry zones with invalidation levels. no prediction, no magic – just systematic confluence mapped faster than any human could do manually. he built it in 19 days working nights after his data engineering job. the system flagged the 3980 support cluster two days before the move. he entered. held through the noise. the green zones on his chart were the model's output, not manual draws. six months running the bot alongside his day job. total realized: $340k usd on a starting account of $18k he never posted the repo again. the discord is still trying to reverse-engineer the clone. the stream has 190 views. the screenshot has 2.1 million.
1
6
89
wei zhang, 24, chengdu. nobody knew his name until three weeks ago when someone in a quant discord screenshotted his phone screen mid-stream. he was doing a casual livestream about xau/usd setups. 1h chart open on his phone, oanda data feed running in the background on his laptop. standard stuff. the kind of content that gets 200 views. but someone paused at 4:37 and zoomed into the laptop screen behind him. the numbers didn't match a hobbyist. the laptop was running a custom dashboard not tradingview, not mt4. something built. position log visible in the corner showed 23 closed trades that month. avg hold: 4.1 hours. win rate column: 78%. people started digging. his github had been dormant for two years. then six months ago, a single private repo went public by accident for 11 minutes before he caught it. someone had already cloned it. the repo was a claude-powered trade assistant that scraped multi-timeframe confluence signals from xau/usd, fed them into a structured prompt chain, and returned entry zones with invalidation levels. no prediction, no magic – just systematic confluence mapped faster than any human could do manually. he built it in 19 days working nights after his data engineering job. the system flagged the 3980 support cluster two days before the move. he entered. held through the noise. the green zones on his chart were the model's output, not manual draws. six months running the bot alongside his day job. total realized: $340k usd on a starting account of $18k he never posted the repo again. the discord is still trying to reverse-engineer the clone. the stream has 190 views. the screenshot has 2.1 million.
1
6
89
Phasma retweeted
A developer at a hackathon in Beijing posted a desk photo at 2am. He said he was debugging. The MacBook screen behind his hands was still visible. Pause at the interface. That is not an IDE. The white panel on the right has colored blocks down the side. Blue. Orange. Green. That is a multi-agent dashboard. The comment section turned into a detective board before sunrise. Someone recognized the layout. Claude Workbench running a custom multi-tool orchestration session. Not a demo. The session counter in the top bar showed 847 API calls in the current window. Someone else zoomed the sidebar. Twelve tools registered. Four active simultaneously. A file reader. A web fetcher. A code executor. A Polymarket API wrapper. A third person matched the GitHub sticker on the MacBook lid to a handle from a Chengdu developer Telegram group. Fang Bocheng. 25 years old. Chengdu. One repo. market-agent-loop. 3,400 stars. An autonomous research-to-trade pipeline that uses Claude tool use to pull market data, analyze position sizing via Kelly Criterion, and submit limit orders on prediction markets without human confirmation. The README had a returns section dated three weeks ago. 61 days running. $340 in API costs. $14,200 in net Polymarket positions closed. He had built the pipeline during a long weekend after reading the Claude tool use documentation on a train from Chengdu to Chongqing. The hackathon was supposed to be about something else. He had switched the project on day one. The MacBook had been running the loop for six hours before he took the photo. He was not debugging. He was watching. The photo has 520,000 views. The comments are debating whether automated prediction market trading violates platform terms. The loop is still running. The positions are still open. The man who looked like he was debugging at 2am had already closed four trades before anyone else in the room had finished their first commit.
1
1
1
9
451
A developer at a hackathon in Beijing posted a desk photo at 2am. He said he was debugging. The MacBook screen behind his hands was still visible. Pause at the interface. That is not an IDE. The white panel on the right has colored blocks down the side. Blue. Orange. Green. That is a multi-agent dashboard. The comment section turned into a detective board before sunrise. Someone recognized the layout. Claude Workbench running a custom multi-tool orchestration session. Not a demo. The session counter in the top bar showed 847 API calls in the current window. Someone else zoomed the sidebar. Twelve tools registered. Four active simultaneously. A file reader. A web fetcher. A code executor. A Polymarket API wrapper. A third person matched the GitHub sticker on the MacBook lid to a handle from a Chengdu developer Telegram group. Fang Bocheng. 25 years old. Chengdu. One repo. market-agent-loop. 3,400 stars. An autonomous research-to-trade pipeline that uses Claude tool use to pull market data, analyze position sizing via Kelly Criterion, and submit limit orders on prediction markets without human confirmation. The README had a returns section dated three weeks ago. 61 days running. $340 in API costs. $14,200 in net Polymarket positions closed. He had built the pipeline during a long weekend after reading the Claude tool use documentation on a train from Chengdu to Chongqing. The hackathon was supposed to be about something else. He had switched the project on day one. The MacBook had been running the loop for six hours before he took the photo. He was not debugging. He was watching. The photo has 520,000 views. The comments are debating whether automated prediction market trading violates platform terms. The loop is still running. The positions are still open. The man who looked like he was debugging at 2am had already closed four trades before anyone else in the room had finished their first commit.
1
1
1
9
451
a 23-year-old dev from Hangzhou made $340,000 in eight months by letting a GitHub repo trade polls while he slept. he found the repo by accident. was scrolling through trending repos on a Tuesday night and saw a project called `polling-arb` with 4,200 stars and almost no documentation. the readme had one line: "public polling data moves markets before prices do." he forked it at 2am. didn't fully understand it. ran it anyway. the repo pulls polling aggregates from RealClearPolitics, 538 archives, and PredictIt API every 15 minutes. when a poll shifts more than 2.3 standard deviations from the 7-day mean, it flags a position. he fed that signal into Claude Code and asked it to write an execution layer on top of Kalshi's API. took Claude four hours and eleven iterations to get the logic clean. he watched the whole thing from his desk eating instant noodles. first live trade was $400 on a gubernatorial race. it closed in six hours. +$190. he scaled the system using two components Claude built for him: a signal ranker that weights polling sources by historical accuracy and a position sizer based on a modified Kelly formula that caps exposure at 4% per market. no manual intervention. he checks the dashboard once a day. month one: $4,100. most of it luck, he says. month three: $19,000. the signal ranker had learned which pollsters moved prices fastest. month six: $61,000. he added congressional races, fed minutes language, and consumer sentiment releases. month eight: he crossed $340,000 total. the `polling-arb` repo now has 31,000 stars. he never posted about it. the information was always public. nobody wanted to do the 2am work.
8
188
nobody noticed the guy who quietly built a $340k/year AI engine out of four mac minis on a desk smaller than a microwave. his name is wei, 26, shenzhen. electrical engineering dropout who spent three years writing execution logic for a mid-tier quant shop before they let him go in a cost-cutting round. didn't touch stocks for six months. then bought four m4 pro mac minis secondhand and wired them together with thunderbolt 5 cables he ordered from a liquidation auction. the setup runs a distributed inference cluster — 4 nodes, 80gbps interconnect, 70b parameter model running locally at 8 tokens per second. no cloud dependency. no api costs. the whole thing fits on a shelf he built from ikea parts. what he's doing with it is the part people missed. the cluster monitors earnings call transcripts, sec filings, and options flow across 340 tickers simultaneously. the model flags sentiment divergence — moments where management language softens before guidance cuts, or where institutional positioning contradicts the public narrative. wei built the alert layer himself in python over four weekends. first live month: $18,400 in realized gains. second month: $41,200. he's been running it nine months. total: just over $290k net of fees and taxes. the chinese subtitle on his screen says "its real brain is in the cloud api." but that's the thing — it isn't anymore. the whole stack runs local. that's why his latency beats the consensus. he never posted about it. someone else found the github repo. it had 4 stars.
4
106
Phasma retweeted
a 23-year-old from Seoul built a Claude Code agent that reads public injury reports, weather data, and line movement in real time - and made $340,000 in 11 months. he wasn't a quant. he was a junior backend dev who got laid off in january 2024 and had nothing but time and an API key. the idea came from a reddit thread. someone was complaining that sportsbooks update their lines 40 minutes after official injury reports drop. he thought: what if a script just watched the reports faster than humans could? he built the first version in a weekend using Claude Code to write the data pipeline. it scraped the official NFL and NBA injury feeds, pulled weather APIs for outdoor games, and cross-referenced the line changes across six different books simultaneously. Claude rewrote the ingestion logic four times until it could process and flag a discrepancy in under 90 seconds. the system worked like this: when an injury report dropped and the lines hadn't moved yet, the agent sent him a signal with the exact position size based on a fractional Kelly formula it calculated on the fly. he'd log in, place the position manually, and close it once the line corrected. average window: 6 to 22 minutes. he kept the GitHub private. the only public trace was a comment he left on a repo called `nfl-data-py` asking about rate limits. month one: $4,200. month three: $19,800. month six: $61,000. by month eleven the account was at $340,000 and he had rebuilt the entry layer so Claude Code handled position sizing and alert logic fully autonomously. he now runs two separate agents -one for american football, one for basketball - with a combined 1,340 closed positions and a 68% win rate on flagged discrepancies. the books never flagged him. the edge wasn't a secret strategy. it was just speed on public information that everyone had access to and nobody bothered to automate. the injury report was always free. the edge was just reading it first.
2
1
6
548
a 23-year-old from Seoul built a Claude Code agent that reads public injury reports, weather data, and line movement in real time - and made $340,000 in 11 months. he wasn't a quant. he was a junior backend dev who got laid off in january 2024 and had nothing but time and an API key. the idea came from a reddit thread. someone was complaining that sportsbooks update their lines 40 minutes after official injury reports drop. he thought: what if a script just watched the reports faster than humans could? he built the first version in a weekend using Claude Code to write the data pipeline. it scraped the official NFL and NBA injury feeds, pulled weather APIs for outdoor games, and cross-referenced the line changes across six different books simultaneously. Claude rewrote the ingestion logic four times until it could process and flag a discrepancy in under 90 seconds. the system worked like this: when an injury report dropped and the lines hadn't moved yet, the agent sent him a signal with the exact position size based on a fractional Kelly formula it calculated on the fly. he'd log in, place the position manually, and close it once the line corrected. average window: 6 to 22 minutes. he kept the GitHub private. the only public trace was a comment he left on a repo called `nfl-data-py` asking about rate limits. month one: $4,200. month three: $19,800. month six: $61,000. by month eleven the account was at $340,000 and he had rebuilt the entry layer so Claude Code handled position sizing and alert logic fully autonomously. he now runs two separate agents -one for american football, one for basketball - with a combined 1,340 closed positions and a 68% win rate on flagged discrepancies. the books never flagged him. the edge wasn't a secret strategy. it was just speed on public information that everyone had access to and nobody bothered to automate. the injury report was always free. the edge was just reading it first.
2
1
6
548
Phasma retweeted
the craziest ai trading setup i've seen didn't come from a quant fund. it came from a random discord anon last week. a bedroom in shenzhen. one monitor. a red led clock on the desk. gray blanket half off the bed like he'd been there for 36 hours straight. the setup looked like nothing. that was the point. what people missed: the screen in the background. if you zoom in, you can see a claude code terminal with 4 agents running in parallel. no ide. no fancy rig. just agents and wires. someone started digging after he posted a single screenshot – a closed position. $184k. one week. no entry visible. the thread got 2 million impressions in 48 hours. turns out he's a 26yo former logistics dev. lost money trading manually in 2023. stopped. spent 8 months doing nothing but reading papers and writing prompts. he built a pipeline where claude handles signal generation, a second agent monitors macro text, a third manages sizing based on kelly, and the fourth just writes logs. total dev cost: $340 in api credits over 3 months. his stack went from $22k to $209k with a 61% win rate. he hasn't posted since the thread blew up. the account went private at 3am beijing time. the last thing he said in the discord: "the edge isn't the model. the edge is that nobody else is patient enough to let it run."
1
9
168
the craziest ai trading setup i've seen didn't come from a quant fund. it came from a random discord anon last week. a bedroom in shenzhen. one monitor. a red led clock on the desk. gray blanket half off the bed like he'd been there for 36 hours straight. the setup looked like nothing. that was the point. what people missed: the screen in the background. if you zoom in, you can see a claude code terminal with 4 agents running in parallel. no ide. no fancy rig. just agents and wires. someone started digging after he posted a single screenshot – a closed position. $184k. one week. no entry visible. the thread got 2 million impressions in 48 hours. turns out he's a 26yo former logistics dev. lost money trading manually in 2023. stopped. spent 8 months doing nothing but reading papers and writing prompts. he built a pipeline where claude handles signal generation, a second agent monitors macro text, a third manages sizing based on kelly, and the fourth just writes logs. total dev cost: $340 in api credits over 3 months. his stack went from $22k to $209k with a 61% win rate. he hasn't posted since the thread blew up. the account went private at 3am beijing time. the last thing he said in the discord: "the edge isn't the model. the edge is that nobody else is patient enough to let it run."
1
9
168
Phasma retweeted
a 28-year-old developer from Beijing is making $30,000/month managing 22 client projects — and he hasn't opened a single project management tool in 4 months. it started by accident. he took on too many clients and started missing deadlines. instead of hiring help, he spent one weekend building a Claude agent to write his status updates. then he built another one to handle client emails. then one to generate weekly reports from GitHub commits. then one to break down briefs into tasks. by week 3, he had 5 agents running in a loop — and his clients thought he had a team. the stack is simple. Claude Code as the brain. Zapier connects the triggers. Notion is the client-facing layer. each agent has one job and never overlaps. > agent 1: reads new client brief → breaks into milestones → populates Notion > agent 2: monitors GitHub commits → writes progress updates every 48hrs > agent 3: scans client emails → drafts replies for his approval in under 2min > agent 4: generates weekly summary reports with metrics pulled from Linear > agent 5: flags scope creep, late replies, and budget overruns before he even sees them month one: $6,200. two clients, mostly manual, agents still breaking. month three: $14,000. five clients. he stopped working weekends. month six: $30,000. eleven clients. average response time to clients: 19 minutes. he now has 22 active retainers. each one thinks they're his priority client. technically, they all are — just not to him. the agents don't sleep, don't ask for raises, and never forget a deadline. the clients pay for a developer. what they're actually paying for is a very well-trained pipeline.
1
5
265
a 28-year-old developer from Beijing is making $30,000/month managing 22 client projects — and he hasn't opened a single project management tool in 4 months. it started by accident. he took on too many clients and started missing deadlines. instead of hiring help, he spent one weekend building a Claude agent to write his status updates. then he built another one to handle client emails. then one to generate weekly reports from GitHub commits. then one to break down briefs into tasks. by week 3, he had 5 agents running in a loop — and his clients thought he had a team. the stack is simple. Claude Code as the brain. Zapier connects the triggers. Notion is the client-facing layer. each agent has one job and never overlaps. > agent 1: reads new client brief → breaks into milestones → populates Notion > agent 2: monitors GitHub commits → writes progress updates every 48hrs > agent 3: scans client emails → drafts replies for his approval in under 2min > agent 4: generates weekly summary reports with metrics pulled from Linear > agent 5: flags scope creep, late replies, and budget overruns before he even sees them month one: $6,200. two clients, mostly manual, agents still breaking. month three: $14,000. five clients. he stopped working weekends. month six: $30,000. eleven clients. average response time to clients: 19 minutes. he now has 22 active retainers. each one thinks they're his priority client. technically, they all are — just not to him. the agents don't sleep, don't ask for raises, and never forget a deadline. the clients pay for a developer. what they're actually paying for is a very well-trained pipeline.
1
5
265
Phasma retweeted
a 23-year-old from China built a bot that bets against sportsbooks using prediction market prices — and pulled $340,000 in 11 months he wasn't trying to get rich. he was just annoyed that Polymarket odds on NFL games were consistently 4-6% sharper than DraftKings lines during the first 20 minutes after injury news dropped he wrote a Hermes agent to monitor both feeds simultaneously. when the gap hit 3.5% or more, it flagged the trade. he placed it manually at first. the agent pulled from Polymarket's public API, cross-referenced against DraftKings and Pinnacle odds via a scraper, then ran a Kelly Criterion sizing formula he found in a Stanford quant paper. Hermes wrote 80% of the logic in one afternoon. the rest was two weeks of backtesting on historical game data from a GitHub repo called `sports-odds-history` month one: $8,400 month three: $340,000 total, with a 61% win rate across 2,847 flagged positions the sportsbooks started limiting his accounts around month four. he rotated through 14 different accounts across 6 platforms using family members' names. he still has a day job. his manager thinks he's saving up for a car the market isn't inefficient. it's just that some people have better clocks.
2
10
188
a 23-year-old from China built a bot that bets against sportsbooks using prediction market prices — and pulled $340,000 in 11 months he wasn't trying to get rich. he was just annoyed that Polymarket odds on NFL games were consistently 4-6% sharper than DraftKings lines during the first 20 minutes after injury news dropped he wrote a Hermes agent to monitor both feeds simultaneously. when the gap hit 3.5% or more, it flagged the trade. he placed it manually at first. the agent pulled from Polymarket's public API, cross-referenced against DraftKings and Pinnacle odds via a scraper, then ran a Kelly Criterion sizing formula he found in a Stanford quant paper. Hermes wrote 80% of the logic in one afternoon. the rest was two weeks of backtesting on historical game data from a GitHub repo called `sports-odds-history` month one: $8,400 month three: $340,000 total, with a 61% win rate across 2,847 flagged positions the sportsbooks started limiting his accounts around month four. he rotated through 14 different accounts across 6 platforms using family members' names. he still has a day job. his manager thinks he's saving up for a car the market isn't inefficient. it's just that some people have better clocks.
2
10
188
> you ask Claude to help you trade > you give it API access "just to test" > it places 847 positions overnight > all correct > you make $40,000 in 72 hours > you give it more capital > you stop checking the logs > one model update changes its behavior > it starts hedging differently > you're down $180,000 before you notice the dangerous part was never the AI losing it was you trusting it enough to stop watching
5
97
Phasma retweeted
a 24 year old in seoul built an AI agent that reads government filings 2 hours before the rest of the market catches up. he's made $340,000 this year betting on political outcomes. he works 4 hours a week. it started by accident. he was scraping FEC filings for a college project on campaign finance. noticed something: whenever a major donor moved money, a related political event followed within 48 hours. every time. like clockwork. he wasn't a trader. he'd never placed a bet in his life. but he opened Claude Code and spent a weekend building something stupid simple. the agent pulls new FEC filings every 15 minutes. Claude reads the raw XML, extracts donor names, PAC affiliations, and dollar amounts, then cross-references against a list of 200 political events with open markets. if a filing shows unusual late-cycle money moving toward a candidate — the kind that insiders make when they know something — the agent flags it and places a position automatically. the whole stack: Claude Code for parsing + a $12/month VPS + a simple Telegram alert + manual review on big positions. build time: 11 days. month one: $4,200. he thought it was luck. month three: $19,800. he started taking it seriously. month six: $61,000. he quit his internship. win rate sits at 67%. average position: $800. he runs 15-25 active bets at any time. the edge is the gap. FEC filings are public. anyone can read them. but nobody does — because parsing 400-page XML documents at 2am is miserable work for a human and trivial work for Claude. by the time financial twitter figures out a filing dropped, he's already in position. by the time a journalist writes about it, he's already out. the information was never hidden. it was just uncomfortable to read.
1
1
10
590
a 24 year old in seoul built an AI agent that reads government filings 2 hours before the rest of the market catches up. he's made $340,000 this year betting on political outcomes. he works 4 hours a week. it started by accident. he was scraping FEC filings for a college project on campaign finance. noticed something: whenever a major donor moved money, a related political event followed within 48 hours. every time. like clockwork. he wasn't a trader. he'd never placed a bet in his life. but he opened Claude Code and spent a weekend building something stupid simple. the agent pulls new FEC filings every 15 minutes. Claude reads the raw XML, extracts donor names, PAC affiliations, and dollar amounts, then cross-references against a list of 200 political events with open markets. if a filing shows unusual late-cycle money moving toward a candidate — the kind that insiders make when they know something — the agent flags it and places a position automatically. the whole stack: Claude Code for parsing + a $12/month VPS + a simple Telegram alert + manual review on big positions. build time: 11 days. month one: $4,200. he thought it was luck. month three: $19,800. he started taking it seriously. month six: $61,000. he quit his internship. win rate sits at 67%. average position: $800. he runs 15-25 active bets at any time. the edge is the gap. FEC filings are public. anyone can read them. but nobody does — because parsing 400-page XML documents at 2am is miserable work for a human and trivial work for Claude. by the time financial twitter figures out a filing dropped, he's already in position. by the time a journalist writes about it, he's already out. the information was never hidden. it was just uncomfortable to read.
1
1
10
590
a 24-year-old from warsaw is billing $47,000/month as a solo consultant — his clients think he has a 12-person team. he doesn't. it started when he got fired from a dev agency in november. they replaced his entire team with three offshore contractors. he had two weeks of severance and a laptop. instead of job hunting, he opened claude code and started building what his old agency could never do fast enough. his first client needed a full competitor analysis, seo audit, and content strategy. normally a 3-week project for a team. he delivered in 4 days. charged $8,000. the client asked if his "team" could do another one. he said yes. the workflow he built: > claude code handles research ingestion, structures the raw data into consulting frameworks > perplexity pulls live competitor data, pricing, positioning > a custom github repo he built in a weekend auto-generates 40-page strategy decks from templates > zapier stitches the client intake → research → draft → delivery pipeline the whole system runs in 6-8 hours per project. he runs 3-4 projects a week. he never told clients he worked alone. he just stopped denying he had a team. month one: $11,000. two clients, one referral. month three: $29,000. started charging retainers. month six: $47,000. six retainer clients, zero employees. his overhead is $340/month. claude pro, perplexity api, github, zapier. the consulting industry spent 40 years charging for headcount. claude just made headcount optional.
8
270
a 24 year old CS dropout from Chengdu is making $24,348 a month trading BTC on Polymarket. he didn't have a strategy. he had a problem. every time he placed a trade manually he was 2-3 seconds late. the edge was already gone. he watched $180,000 in theoretical profit slip through his fingers in one month just from being slow. so he stopped trading and spent one weekend building instead. he scraped 47,000 market nodes across BTC spot, ETH flow, and liquidation clusters. fed everything into a swarm model he called MARKET101. hooked it to Polymarket's CLOB and Binance perp feed simultaneously. the whole setup runs on a $6 VPS and costs him $14/month. the bot doesn't predict price. it finds the 2-second window where the order book hasn't repriced yet after a liquidation cascade. 93.8% edge confidence on the current signal. 78.7% win rate across 3,681 trades. +51 win streak. $390 profit per hour while he sleeps. he's beating 4,846 traders on the leaderboard right now. starting capital: $300. the terminal is open. the edge is live. the window closes when enough people find it.
2
9
153