Quant Trading Firm Open-Sources Gray-Scale Risk Model to Replace Binary Trade Vetoes

·1 views

A quantitative crypto trading team has published its approach to replacing binary risk management with a dynamic 'gray-scale' scaling system. Traditional rule-based engines issue hard vetoes when recent win rates drop, potentially blocking high-confidence signals that still carry statistical edge. The new method instead reduces position size and tightens stop-losses proportionally, keeping strong signals active while limiting capital exposure. In a worked example involving a 90/100 confidence SHORT signal on NEARUSDT, the system cut position size by 30% and tightened the stop-loss by 10% rather than cancelling the trade outright. The team argues that hard vetoes destroy long-term expected value by conflating normal statistical variance with genuine model failure.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why AI Projects Need a Behavior Contract Before Any Code Is Written

Software engineer Derek Wang argues that AI-driven projects require a foundational 'constitution' — a strategy charter that defines the system's identity, boundaries, and human-AI collaboration rules before any work begins. Unlike code conventions, which govern the artifact, this behavior contract shapes how the AI checks, verifies, and handles mistakes. Wang outlines ten model-agnostic rules for his own projects, covering principles such as reading before writing, making minimal edits, and replacing tools after repeated failures. He warns that without such a charter loaded upfront, an AI model is effectively improvising without guardrails, and no downstream prompting can compensate. The approach treats the constitution not as documentation but as a binding contract every participant — human or AI — must internalize at the start.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Koi Editor Uses Handwritten Lexers for Faster Syntax Highlighting and Folding

Koi Editor takes a direct approach to syntax highlighting by using small, language-specific handwritten lexers instead of declarative grammar systems or syntax trees used by editors like Sublime Text and Zed. Each lexer is purpose-built code that decides how text should be styled and folded without a generic engine in between. This design also enables syntax-aware folding, allowing code blocks to remain foldable even when indentation alone cannot determine structure. In performance benchmarks, Koi was able to apply syntax highlighting to files with millions of lines, while Sublime Text and Zed reportedly became unusable at much smaller scales. The developer attributes this efficiency to the lexers doing minimal work per token, tracking only token type, carry-forward state, and fold level.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds ASCII-Based Python Calculator That Runs in the Terminal

Developer Dilesh Zingare has shared a tutorial on building a visual Python calculator that runs entirely in the terminal using ASCII. The calculator accepts two numbers and an operator from the user, then performs one arithmetic operation at a time, including addition, subtraction, multiplication, division, and exponentiation. The logic is handled using Python's if-elif-else conditional statements to differentiate between operations. Zingare acknowledges the tool is basic in its current form but plans to expand its functionality in future updates. The project's source code is publicly available on GitHub for anyone looking to follow along or contribute.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Completes Restaurant Website Built Purely with HTML in Four-Part Series

Developer Dilesh Zingare has published the fourth and final part of his beginner frontend project series on DEV Community. The series documents the step-by-step creation of a restaurant website called 'The Bite of Soul' using only HTML. Each article in the series was accompanied by a lecture video to help readers follow along. The project source code has been made publicly available on Zingare's GitHub profile. Zingare encouraged readers to build their own versions and share the results in the comments.

0 comments Read more at DEV Community