Lowes 🧘🏻 | we're hiring retweeted
I built an open-source trading harness for classification models
GitHub: github.com/zadescoxp/calibra…
While testing the latest classification model from @dmind_ai, Kairos on @minara, I built a trading harness around it to see what the model can actually do. Everything ran on my Bybit testnet account with fees and slippage included, which makes it even more interesting.
Across 3 runs I traded BTC, ETH, SOL, XRP, DOGE and HYPE perps, 10.5 hours in total. All three used a looser policy, with these results:
Run 1: −$2,233 (92% of it came from SOL’s thin order book)
Run 2: −$359
Run 3: +$60.19
With the loose policy, the model was taking trades without proper risk management. So I ran it once more with a stricter policy. After 4 hours it had taken zero trades, which means $0 PnL. Still better than losing a big chunk.
What the harness does:
• Asks the model labellable questions on every bar and scores every answer against what actually happened
• Gives the model the market’s own base rates and calibrates its probabilities against them
• Runs a strict, expected-value-gated policy, or an exploration mode for collecting trade data
• Caps entry slippage with IOC orders, rests take-profits, and confirms stops on mainnet prices
• Books P&L, fees and R straight from the exchange’s own records, fill by fill
• Caps risk per trade, total open risk and same-direction exposure, with a daily loss limit and a kill switch
• Has built-in evaluation: Brier skill, AUC, calibration curves, plus mirror and repeat consistency probes
• Tracks every test run separately next to an overall total
• Comes with a Bloomberg-style live terminal over WebSocket, and you can pick any of the top 100 perps
The harness is built so any classification model can take trades in a real market, reading the current market situation and following a proper risk-management strategy.
Next up: I’m integrating it with my order-flow screener (order-flow-zadescoxp.com), so models get richer, cleaner order-flow data to read the market from.
The project is at a very early stage and I’m looking for feedback and suggestions. If you’d like to contribute, let’s #connect
Lowes 🧘🏻 | we're hiring retweeted
[Minara Desktop & Web v1.9.1 업데이트]
미나라 데스크톱 앱과 웹 버전이 v1.9.1로 새로 업데이트되었습니다!
이번 업데이트는 전체적인 플랫폼의 데이터 정확도와 안정성을 높이는 데 집중했다고 하네요.
▪️ 주요 업데이트 내용
- 대화 기록 개선: 이전 채팅 내역을 더 쉽게 찾고 탐색 가능
- 거래 & 지갑 정밀도 향상: 거래 내역 및 지갑 잔고 데이터 정확도 개선
- 전략 & 백테스트 시각화: 전략지표와 백테스트 결과를 한눈에 알아보기 쉽게 개선
- 전반적인 시스템 안정화: 차트 조회 및 데이터 검색 시 한층 더 부드럽고 쾌적해진 성능
PC 앱을 업데이트하시거나 웹 브라우저를 새로고침하시면 바로 이용하실 수 있습니다! 링크는 댓글에~
📌 출처: Minara 공식 디스코드 공지
#MinaraAI @minara
미나라 v2한테 현재 시장 상황 어떻게 보냐고 물어봤더니 비트($77.8K)부터 미 증시, 국채금리까지 표로 한눈에 싹 정리해 주네요 🤖
이번에 PC(v1.7.0), Web, Mobile(v2.4.24) 전 플랫폼 업데이트가 진행되었습니다.
🔹 업데이트 핵심 포인트
- 차트 로딩 속도 대폭 개선
- 전략 구독 시 레버리지 조절 옵션 추가
- 모바일 TP/SL 가격 미표시 버그 수정 및 UI 정돈
성능 개선에 실시간 시장 브리핑까지 더해져서 매매할 때 참고하기 진짜 좋아졌습니다.
📌 출처: Minara 공식 디스코드
#MinaraAI @minara
Lowes 🧘🏻 | we're hiring retweeted
Agentic Trading已经有PMF了
Robinhood CEO @vladtenev 在 @NoPriorsPod 上说，Robinhood MCP推出后，超过15 万用户开通了agent交易账户，每天大约 3000 万次工具调用
他对agentic trading产品的北极星指标是：把对冲基金的能力，装进普通人的口袋，更多用户会成为量化交易者
和AI编程的采用率类似，专业交易员会先用AI agents交易，再慢慢扩散到每个人
这和@minara 的产品拓展思路一致，降低制作交易策略的使用门槛，让AI agent协助做策略，辅助交易者做各种决策
增量在于如何教育散户、切入交易增量最大的资产和平台
Short too early
NEW: We are short Nvidia $NVDA. We believe Nvidia has a significant China problem. Please see our full report and disclosures now available on our website, link in bio.
3
230
What if Ethereum had never abandoned PoW?
Today, AI data centers could potentially use idle GPU capacity to mine ETH, secure the network, and improve capital efficiency.
This could have created a powerful economic flywheel connecting Ethereum’s DeFi and stablecoins with the semiconductor and AI industries.
Instead, PoS largely confines network security to ETH holders, weakening its direct connection to external AI industry and capital.
My opinion:
I think we sometimes overestimate the power of algorithmic innovation while underestimating the compounding value of real-world economic incentives.
When a proven path already aligns capital, resources, and incentives toward a clear goal, don’t abandon it for the promise of a theoretically better algorithm or paradigm.
I know this is a predictable opinion from me but I don't agree with the strong emotional reaction against the anti-data-center movements that I've seen this week.
As I see it, algorithmic progress means that whatever you would be able to do with more and bigger data centers in 2 years, you will be able to do with the data centers we have today in 4 years - probably sooner. So nothing remotely apocalyptically bad will happen from even extreme levels of success by anti-data-center people.
Meanwhile we need all the time we can get to better harden the world against both already-visible cybersecurity issues and any bio and other risks soon to come. Science applications are bottlenecked by many things other than AI, so slowing AI down by one year will slow those down by much less than one year.
Given this, a slowdown in compute hyper-scaling seems quite net good? Notice this is without any controversial "doomer arguments", RSI etc. Even if net-bad, it's certainly not worse than like 10+ other political things.
Old trading advice: no fat finger.
New trading advice: no stupid finger.
Minara Copilot checks the second one for you.
Here's Copilot in the Minara app, on $AAPL perps.
Set margin, leverage and style, then ask long or short. Copilot gives a side, entry, take profit and stop loss, with the reasons.
One Confirm opens the trade:
For security reasons, only some features are available on the web currently.
If you want the full general financial intelligence experience, I’d recommend the desktop app in a heartbeat.
You’ll get a hands-on feel for the future of agentic finance.
Minara V2 is open for early access.
We did extensive internal testing to ensure every user has their own isolated, secure cloud sandbox for every workflow. 🤓
Enjoy and let me know your feedback!
Lowes 🧘🏻 | we're hiring retweeted
Minara @minara がWeb V2を公開。Desktop版と同じUI・機能がブラウザで使えるようになっただけでなく、ユーザーごとに専用のクラウドサンドボックスが付くのが今回の面白いポイント。
サンドボックスは、コードやツールを動かすための独立した実行環境。金融AIにこれが付くと、回答の精度以上に「調べたデータをどう処理して、次の作業にどうつなげるか」が気になってくる。
例えば、保有銘柄のニュースやオンチェーンの変化を追い、前回との差分を比較して「投資判断の前提が変わったもの」だけをピックアップする。こういう作業を継続して任せられるなら、毎回ゼロから質問する従来の使い方とは体験がガラッと変わるはず。
ただ、専用サーバーがあることと、ブラウザを閉じても処理が続くことはイコールではない。実際に触るなら、タスクの継続性、ファイルや調査履歴の保存、外部ツール連携の範囲はチェックしておきたいところ。
自分が金融AIにまず任せたいのは、まさにこの「調査と差分確認」。売買の自動化より先に、日々の確認作業をどこまで減らせるか。今回のアップデートは、その観点で見るとかなり興味深い👀
Lowes 🧘🏻 | we're hiring retweeted
just tested @minara as a complete beginner
had $1.50 USDC on Base, asked the AI agent to find the best trade, and it asked about my timeframe and risk tolerance before executing a spot ETH buy
no leverage or complications, just a simple conversation and the trade was done
pretty seamless experience
Crypto doesn’t really capture everything happening onchain anymore. And as AI meets crypto, we’re going to see things we’ve never seen before.
Tech buzzwords are useful when something is new. But as an industry grows up, we should start talking about what people actually use it for: stablecoins, DeFi, RWAs, perps, prediction market, and more.
Goodbye, crypto. Hello, agentic finance.
