gavindiazandClaude Code
3a005ddcc6
Full project import for the self-hosted code hosting (https://code.aifunny.ltd/gavindiaz/my-NexusTradingForexBot.git). Includes the live-freshness + scaler-hash fixes for the permanent STALE dashboard (tick_pipeline now_utc stamps; infer_probabilities computes model_input_hash on the actual scaled 70D tensor), the data-fetch default-path fix, the g29 web-auth test fixes, and a Chinese user manual (使用手册.md) for the web dashboard. Scratch diagnostic scripts (check*.py, contain a live web-auth token) and the .kilo/ agent-worktree directory are gitignored, not committed. Co-Authored-By: Claude Code <noreply@anthropic.com>
3.0 KiB
title, description, lang
| title | description | lang |
|---|---|---|
| FAQ | Honest answers to the questions newcomers actually ask. | en |
FAQ
What is Nexus? A research-driven quantitative trading platform: a hexagonal, event-driven scalping engine for MetaTrader 5 (primary: XAUUSD M1) with causal features, deep models, an invariant risk engine, deterministic research tooling and forensic observability. See Vision.
Is this a live trading bot? It can execute live trades on MT5 — but PAPER is the default mode, SHADOW carries zero order authority, and LIVE requires explicit interactive confirmation. The repository's own evidence shows rejected candidates; this is a research platform with a runtime, not a money printer.
Can I run it without MT5? Yes — PAPER mode and the full test suite run without a broker. Docker works out of the box in PAPER mode. SHADOW and LIVE need an MT5 terminal (Windows x64).
What is 70D?
The canonical research feature contract: 50 base dimensions + 10 news + 10
liquidity. It is not live — the live contract is 50D (scalp_v1). The 70D
candidate is rejected so far on OOS evidence. See
Status.
What is the difference between research and runtime? Research produces falsifiable evidence about candidates (datasets → walk-forward → OOS → shadow). Runtime executes decisions with hard clamps. Research has zero order authority — the two never mix.
How is leakage (lookahead) prevented? Purged + embargoed walk-forward, strictly causal features (confirmation bars, completed HTF buckets), REPLACE+ALIGN history handling, replay bit-exactness tests, and recorded effective purge/embargo per run (BUG-183). See Walk-Forward.
How are models identified? Artifact-first manifests: dataset ID, feature-schema hash, scaler identity, git commit — validated by the 10-gate load gate at every attach. A model without provenance cannot attach.
How does replay differ from backtesting? Backtest: score a strategy on history. Replay: prove the same code path as live behaves identically on history (bit-exact vs dataset). See Replay.
How is a feature contract enforced? Schema registry + SHA-256 hash over canonical feature JSON + inference validator (dimension, bounds, finite, hash, scaler-dim match). Violations are loud rejections, never silent padding.
What is certified? See the Capability Matrix — each row names its evidence. Certified ≈ formal forensic acceptance with reproducible artifacts (e.g. release verification, provider-gate live smoke, OOS gate behavior).
Is it profitable? No claim is made — that is the point. Published evidence includes negative results (70D OOS rejection, counterfactual stratification). Judge from the evidence, not from promises.
Why is the bug ledger public?
Because agents/bugs.md (root causes, evidence, regression guards) is the
engineering memory. Hiding it would destroy the property that makes the
platform trustworthy.