𝕂𝕠𝕜𝕚

Building @VoidScanner — Solana AI trading agent.
Real code. Real PnL. Real failures. 🔴
Claude AI as my senior engineer.
Paper → Live soon. 🇭🇺 builder

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

1/6
VoidScanner was finding good tokens.

But it was managing positions badly.

Today I fixed that. 🧵

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

@kalpperisii an autistic person with a lot of money

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

@Borg_Cryptos 2-1 Ronaldo 🫂

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

@blknoiz06 What eth? what is that

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

@DanielSmidstrup There is a West and there is an East that do not want a strong Europe

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

@mert According to the development plan, will the upgraded WS reliably transmit https://t.co/8a9xbtWwp1 Migrate events in real time?

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

Running a Solana memecoin bot on Helius free tier. WS Migrate events not reliable effectively scanner-only 5-15min delay after graduation late entries.
Considering Developer plan for Enhanced WS. Question for anyone who's tested it:
Does transactionSubscribe actually deliver https://t.co/JPAuZjPknO graduation events fast enough to get EARLY entries <2min after grad?
Real latency numbers appreciated, not marketing copy.

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

6/6
What's next:

Not a new scanner.
Not a new AI plugin.

Stable data collection.
50-100 trades. Same config.

Let the system prove itself.

This is how you build something real. 🔴

#Solana #BuildInPublic #AIAgents

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

1/6
VoidScanner was finding good tokens.

But it was managing positions badly.

Today I fixed that. 🧵

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

5/6
Results — old vs new:

Old losers: -15%, -20%, -95%
New losers: -0.4%, -2.6%, -3%

Winners still alive: +7%, +18%, +96%

First live validation:
[MomentumExit] triggered on a
125sec / -11% / 0 momentum trade.

The lifecycle layer works.

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

6/6
Honest assessment:

30.7% win rate is low.
The liquidity_gone exit trigger needs work.
Rugpull protection needs work.

But the system is running.
The data is real.
And every loss is a lesson I can fix.

This is day 1 of live data collection. 🔴

#Solana #BuildInPublic #AIAgents

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

1/6
First real data day for VoidScanner.

13 closed trades.
Win rate: 4/13 = 30.7%
Net: -0.0678 SOL

Here's what actually happened. 🧵

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

5/6
What broke today:

WS Watchdog — zombie mode 10h50m. Fixed.

Validator — https://t.co/8a9xbtWwp1 API 530 blocked ALL tokens.
Fixed. First trade after: PALESTINE opened.

Telegram alerts — now live.
No more silent failures.

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

Just to flag this isn't an Anthropic release. It's a chapter from someone's course or book. The math is solid (Kelly Criterion and VaR are legit) but the $300-1500/day claim comes from a 90-day simulated backtest which is basically meaningless without live trading results to back it up. Anyone can curve-fit a backtest to look good.
Good concepts though. Wrong headline.

[𝕂𝕠𝕜𝕚 @Kokiahegyrol]

Great breakdown of the three leak paths. Most people only worry about Claude directly opening .env and completely miss the runtime output angle. Had a moment on a Solana bot project where a failed RPC call dumped a full connection string into stdout and Claude was watching. Not fun.The settings.json deny rule is the one thing I'd tell every Claude Code user to add before anything else. One line. Five minutes Done.