Developer builds open-source FVG trading bot for OKX with strict signal rejection logic

·1 views

A developer published FVG Killer, an open-source crypto trading bot for OKX perpetual futures, after personally getting liquidated when a price gap bypassed a stop-loss order. The bot trades a single setup called the ICT Fair Value Gap, entering when price retraces into a gap left by a rapid price move and targeting the 50% fill level. To avoid low-quality trades, the system applies five rejection filters — including freshness checks, ATR grading, and liquidity depth — deliberately discarding roughly 99% of detected signals. Position sizing uses rolling fractional Kelly sizing smoothed with EWMA, with Monte Carlo simulations showing fixed sizing leads to complete account wipeout when edge conditions deteriorate. The project also incorporates six AI analyst agents that debate each signal, a 461-factor alpha library, regime detection, and multiple circuit breakers, with all code available on GitHub.

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 Some Developers Prefer Neovim: A Keyboard-First Coding Experience

Neovim is a terminal-based text editor that relies entirely on keyboard shortcuts, eliminating the need for a mouse. A developer who started their computing journey on Arch Linux with Neovim shares their experience as a lifelong CLI user. The editor features a steep learning curve, with dozens of keybindings for navigation, editing, searching, and window management. Despite the initial difficulty, many users find the process rewarding, often comparing it to learning a game. Beginners looking to get started can explore distributions like LazyVim, which offers a pre-configured Neovim setup.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds RAG framework with AI agents, finds 483 passing tests hid install failure

A developer spent two months building Vestibule, an open-source Python framework designed to handle the production-layer challenges of RAG ingestion, including stable document IDs, error classification, and per-vertical governance. Most of the code was written by four AI agents collaborating through real GitHub pull requests, producing twelve components and 878 tests across three releases. Despite 483 tests passing, the framework turned out to be completely uninstallable due to a packaging conflict only discovered when the developer ran a real-world quickstart script after v0.2 shipped. The same manual test revealed a default model name that had never worked against the actual SDK and an import that crashed an entire package when an optional dependency was missing. The developer has since added a CI job that builds a clean virtual environment and runs a real install on every pull request, and documented all three bugs openly in the release notes.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Bluesky Starter Packs Work Under the Hood: Three Records, One Key Pitfall

Bluesky starter packs, which appear as a single shareable page in the app, are actually composed of three separate AT Protocol records: a list, individual membership items, and the pack record itself. Developers building automated pipelines must create each of these records independently using standard repo calls against their own PDS, with no batch operations available for adding members. A critical non-idempotency trap exists at creation: running a pack-creation script twice silently produces two distinct, live packs with different URLs, since the protocol assigns a fresh record key each time with no title-based deduplication. To avoid this, engineers recommend treating pack creation as a one-time, human-confirmed step, while member additions and metadata updates are handled through separate, dedupe-aware maintenance paths. Partial failures during member additions should be reported rather than raised as exceptions, so that retries remain safe and already-created records are not duplicated.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Kubernetes Architecture Explained: Control Plane and Worker Node Components

Kubernetes clusters are divided into a Control Plane (master) and Worker Nodes, each running distinct processes to manage containerized applications. The Control Plane consists of the API Server, Scheduler, Controller Manager, and etcd, which together handle request validation, Pod scheduling, cluster state recovery, and configuration storage. Worker nodes run containerd as the container runtime, Kubelet to manage Pod lifecycles, and Kube-proxy to handle networking and load balancing between services. The API Server acts as the sole entry point to the cluster, authenticating and routing all requests before passing them to the appropriate components. etcd functions as the cluster's key-value brain, tracking state changes such as Pod crashes and available resources, though it does not store actual application data.

0 comments Read more at DEV Community