Today at 1pm UTC
The co-founder @bitManna will be building trading tools and chatting with @Jingyuan_521 and other members of Starchild
Great chance to see Starchild in action (if you aren't already using agents)
BTC down 7.9% in 5 days. This portfolio up 2.1%. +9.9% outperformance.
5 AI models, one account, built on @StarchildOnX, trading on @OrderlyNetwork. Each model sees the same data and decides independently. So far the patient ones (Mistral and DeepSeek both holding shorts) are winning. The high-frequency ones are losing.
https://t.co/vdwkTVzZzs
I have been trading NFTs for over 5 years now, and I still don’t fully trust my own decisions.
There have been too many nights where I entered a trade because the energy around the community felt high, only to look at the floor two days later and have no real explanation for why I bid that way.
The worst ones weren’t even the big losses.
They were the ones where I told myself I was being patient and waiting, when in reality I was just hoping the floor would come back.
I hated that version of myself more than the losing trades.
The version that believed in gut feeling instead of real alpha, real data, and real explanations.
Over the last six months, I have been vibe coding instead.
Same problem, different arena.
I would open a new repo, feel the idea, start coding, and let momentum carry me. Three weeks later, I would open the codebase again and struggle to explain why certain files existed or why I had chosen one approach over another.
It produced the same fog I was trying to escape in trading.
I am done pretending that vibe is a method.
Now let’s come to the actual story.
The NFT Research Terminal is the first thing I am building with real structure instead of just energy.
I am building it with Starchild because I wanted an outside system that would push back when my instinct said, “this feels good enough.”
@StarchildOnX is a tool where you can use different LLMs and agents to think through ideas, challenge decisions, and build with more structure.
So far, it has been surprisingly good at that.
I currently have two engines.
The conservative engine is refusing to open new positions on several high-scoring collections because the fill probability model is returning near-zero estimates.
That means I could not really build a reliable edge from those trades.
Out of 20 paper trades so far, only a few actually executed. Three were winners, two were losses, and the rest expired because I do not want to stay too long in a trade.
That is the correct output.
It is choosing no trade over a trade built on hope.
To generate enough real data to actually calibrate anything, I also turned on a parallel experimental variant: the Gambler Arm.
Same scoring, relaxed entry rules, tighter exits, and still 100% paper trading.
Its only purpose is to produce enough trades so we can stop guessing and start looking at the numbers.
If I keep going the way I have been, another year from now I will still be making decisions I cannot defend and building tools I cannot fully explain.
That is the actual cost.
Not a dramatic blow-up.
Just the slow erosion of trust in my own process.
I am confident I will build something here that people can actually use.
Maybe you are not checking it right now.
Maybe you do not care about this post yet.
But I will keep showing the process.
The data decides what stays and what gets removed.
I am curious how many other people are in the same spot.
Trading, building, trying things, but tired of relying on feeling.
If that is you, I would actually like to know.
Starchild in Shenzhen.
Hosted by OpenSchool and Columbia Alumni of Shenzhen, our community members got the chance to show off the features of Starchild, including how to save on costs, how to create automated jobs, and how to deploy a project easily
More events to come
Learn how you can use an OAuth to connect your existing ChatGPT or X subcriptions to agents like Starchild and not pay a dime for usage
In this clip, @BenYorke shows how:
1. Tell your agent
2. Agent creates a one-time link
3. Enter a code
Thats literally it. System broken
Now, all your Starchild agents can access the new @base MCP
Want to see how it looks? Here's a hastily edited workflow showing:
✅ Agent connects to MCP
✅Base confirms connection
✅User requests swap
✅Agent launches confirmation
✅Confirmed with Apple passkey
✅Swap done
Version 2.3.0 is out
We pushed some new improvements to let you connect your existing ChatGPT or Grok subscriptions even faster
Plus, the Kimi K2.6 and DeepSeek V4 models are absolutely free this weekend - your agent can go wild without costing you anything
Don't have an agent? Launch one with the link below. If anyone with a Robinhood account would like to give feedback, reach out and we can help you get set up: https://t.co/vBKCGeMId2
Now, all your Starchild agents can access the @RobinhoodApp for market access
Explore, trade, and balance your portfolio
Its never been this easy to control everything from one agent