Been cooking something the past weeks. Keywords: rh, nfts, 1 token=1nft, on-chain casino... Anyone that wants to get involved send me a dm. This tweet will auto-destruct in two hours ;)

Is there any way you can filter out all AI generated content on YouTube?

locked in

I've added support for these crime wallets in the Tradecraft signal lab. It keeps the list up to date and through the "Crime Wallets" node you can incorporate this as a condition in your strategies. In the example we signal every token that has at least 5 individual crime wallets as a buyer. The 24h backtest shows 238 tokens with a 14% win rate (which is very low), that definitely means these were bad tokens you would not have wanted to buy. In practice you would use this in the opposite direction, meaning filter out any tokens that these wallets are involved in. Great job @GualaGualahh!

Follow-up on this. The WOFL token that we picked had a high above 800k and now sits at a dip now at 400k. That other one? Topped at 200k and now sits at 20k. The point is - the data you have available in the Tradecraft radar mode does help when picking tokens within a narrative.

Here's an example of how you can use the radar mode in Tradecraft. In the promising column we have two WOFL tokens. You see in the cards that they are part of a narrative. You can zoom in and see data about all the 77 other tokens within this narrative. Comparing the two, DTGWeDVPAvr7TdY7sdbfjz2bi4QihVMMn9KhyKH1pump looks better as it has a higher 5m (and 1h) volume, more Telegram users that have potentially seen the contract (324k vs 213k). I put in a paper trade on that token, let's see how it turns out.

That's all good, but what if you dont want to buy when there are too many holders that sit with big unrealized profits? Then just add an MVRV node to your strategy. In this example we've set it to max 1.5, which roughly means on average we dont want the holders to sit on more than a 1.5x unrealized profit. Well what do you know, 26 signals over 24 hours with a median multiplier of 2.21x. Win rate (at least 2x) of 61%. Definitely good enough to be profitable.

With the QualifiedWallets node we attribute a score to all the wallets on Solana and give them a ranking from 0-100 depending on recent performance. In this strategy we signal whenever 4 wallets with a score over 75 buys a token within 1 minute of each other. To the left are the backtest results. Should be good enough to trade profitably on those signals.

Built something cool over this past week that leans more into a semi-automated style of trading. I know that the token discovery in the gmgns and axioms of the world are popular so did something similar but adapted to the Tradecraft workflow, signalling strategies and data pipeline. On top I added something I dont think any other platform has - a replay mode where you can go back in time and see the tokens appear as the replay progresses. Kind of like watching a video on Youtube. You can pause, accelerate, go back and forth. The intent there is to eventually introduce a paper trading so that you can experiment with different approaches and see what would have worked. Someone, I think it was @blknoiz06, talked about having a Duolingo for trading, and I guess it goes in that direction for new pairs. The video shows both the live and replay modes. What's important to understand is that the condition for a token to appear in the first lane is that it was triggered by a user defined Tradecraft strategy. In the case of the video it's when a certain number of TG groups have mentioned one of those tokens.

Asked my friend Codex to go over the Tradecraft codebase and list the features. It gives a pretty accurate view. As mentioned previously I've closed the beta for now to focus on private use within a small group. Having said that, if you are an advanced new-pairs trader on Solana and want to check it out, send me a DM. One year into this (well four if we count the previous ETH based project) and it has grown into something very unique. Signal Lab - Visual binary tree signal strategy builder - Live evaluation against the Solana transaction firehose - Historical signal strategy simulation and replay - Exact execution-path inspection for every evaluated token - Live and historical token charts - Strategy heatmaps and node-level diagnostics - Importable and exportable strategy definitions - Time-of-day analysis with timezone support - Historical token lookup and signal comparison - Performance analysis by win rate, best return, average, median, P70 and P80 - Automated strategy-configuration optimization Supported on-chain indicators - Market cap and trading volume - Volume acceleration against historical baselines - Token age and launch platform - Emerging narratives and copycat detection - On-chain token-burn events - Specific wallet tracking - Qualified-wallet activity - Independent buyer groups - Fresh and dormant wallets - Fresh-wallet supply concentration - Coordinated wallet-funding clusters - Funding-source analysis - Observed-holder MVRV - RSI, EMA, MACD and Fibonacci conditions - External signal-source conditions - ML launch scoring and price-multiple predictions Backtest simulations - Distributed simulation workloads - Replay of Signal Lab graphs over historical Solana data - JavaScript strategy backtesting - Configurable date ranges, data sources and starting balances - Complete simulated trade and position histories - Starting balance, ending value and net-P&L calculations - Per-signal and per-trade result inspection - Simulation comparison and result export - Historical execution-path reconstruction JavaScript trading strategies - Programmable strategy(data) execution contract - buy, sell and hold decisions - Access to prices, indicators, signals, configuration and position state - Custom position sizing and decision reasoning - Multi-stage entries and exits - Dip-entry conditions - Partial profit-taking - Global stop losses - Time-windowed entries with timezone handling - Visual easy mode or full JavaScript code mode - AI-assisted strategy creation and code editing - Isolated strategy execution environments - Worker-thread execution pools - Persisted strategy state Strategy debugger - Line breakpoints - Conditional breakpoints - Enable and disable individual breakpoints - Step into - Step over - Continue and stop - Live variable inspection - Execution-state inspection - Breakpoint-aware historical simulations - Detailed strategy, position and trade logs - End-to-end trade execution-path inspection Machine learning - XGBoost classification models - Separate 2x, 3x, 5x and 10x outcome targets - Lifecycle-specific models for nascent, young, maturing and established tokens - General models used as inference fallbacks - Point-in-time feature snapshots - Raw and derived market features - Market-cap-relative volume and wallet features - Wallet-behavior and token-lifecycle features - Recency-weighted model training - Automated retraining cycles - Model validation and feature-compatibility guards - Native XGBoost inference inside the Rust signal executor - Hot model reloading without restarting signal processing - Signal outcomes fed back into future training data Launch intelligence and ML data engineering - First-tradable-swap launch anchors - Dense launch-relative snapshot schedules - Price, volume, wallet and market-state capture - Launch-relative and trajectory feature generation - Price, market-cap, volume and wallet-growth slopes - Drawdown, volatility, pullback and recovery features - Sustained-winner and trap labeling - Multiple prediction checkpoints after launch - Versioned feature contracts - Versioned labels and training datasets - Data-quality eligibility gates - Missing-feature and snapshot-coverage auditing - Model registry, promotion and scoring pipelines Trading and operations - Manual and fully automated trading - Attach an automated strategy to an existing manual position - Multiple embedded or self-managed Solana wallets - Per-wallet strategy assignment - Position sizing, slippage and priority-fee controls - Accelerated transaction execution - Partial buys and sells - Live position and P&L monitoring - Detailed trade history and execution paths - Pause, resume or remove individual monitored tokens - Automatic source activation based on measured performance - Configurable win thresholds, evaluation windows, sample sizes and cooldowns Signals and integrations - Telegram channels, groups, users and bots - Discord channels - Webhook and API signal sources - Private and public data sources - Signal-source marketplace - Direct generation of Signal Lab or trading strategies from a data source - Telegram and Discord alerts for signals, trades, positions, errors, simulations and activation changes This is the scope of Tradecraft: a complete research, simulation, ML, execution and observability stack for automated trading on Solana.

Building /launch-scorer. At the surface: a score for whether a fresh launch is worth paying attention to. Behind it: launch-time feature snapshots (hundreds of on-chain features per token), delayed labels, model versioning, offset-specific decisions, XGBoost, signal emission. Tons of on-chain data and continuous outcome tracking. Simple output. Complicated machinery. The 900s model looks decent, 44% win rate. This is live for all the Tradecraft beta testers and can be integrated into the signalling strategies.

Spending time on trading and improving strategies has paid off big time. The small group of guys that believed in this from the start will continue to work with me in a private group. The Tradecraft open beta is officially closed and no more members will be accepted. What you see there is the win rate of signal over the past 3 days. A win in this context means at least 2x from the signal.