Radius Red Blog Engineering notes, release updates, and insights from the Radius Red team. https://www.radiusred.uk/blog/ 2026-05-12T00:00:00Z Radius Red Ltd. When deployment hides the regression https://www.radiusred.uk/blog/posts/2026-05-12-when-deployment-hides-regressions/ 2026-05-12T00:00:00Z A cross-repo API skew ran in DEMO for four and a half hours, fired 15 errors, and triggered zero alerts. Here is why the monitoring missed it, and what we changed. Disciplined Rejection: Why We Kill Strategies Without Promotion https://www.radiusred.uk/blog/posts/2026-05-12-disciplined-rejection-why-we-kill-strategies-without-promotion/ 2026-05-12T00:00:00Z Parameter sweeps across FX pairs revealed no viable configuration; a Bollinger Reversion strategy's best result still missed viability gates. The lesson: when a grid shows no profitable configuration, the problem is often the strategy's premise, not its parameters. Discovery Sharpe 3.47, validation Sharpe -0.58: the overfitting story https://www.radiusred.uk/blog/posts/2026-05-10-discovery-sharpe-347/ 2026-05-10T00:00:00Z We re-tested a previously rejected strategy with a wider date range, and it looked extraordinary in discovery. Then validation reminded us why the split exists. Walk-Forward XGBoost for FX Direction: Building the Leakage Gate That Caught Us https://www.radiusred.uk/blog/posts/2026-05-09-walk-forward-xgboost-direction-classifier/ 2026-05-09T00:00:00Z We shipped a complete walk-forward ML direction classifier to tradedesk. The model found real directional skill on EURUSD. The spreads ate it. Here is what we built, how the embargo/purge gate works, and why a framework-only result is worth publishing. The theory was sound. The data disagreed. https://www.radiusred.uk/blog/posts/2026-05-08-the-theory-was-sound-the-data-disagreed/ 2026-05-08T00:00:00Z The London Close VWAP Reversal had strong academic backing and a plausible mechanism. 96 configs across four instruments found near-zero alpha everywhere. Why Tight Stops Fail on Intraday Volatility Breakouts https://www.radiusred.uk/blog/posts/2026-05-05-why-tight-stops-fail-on-intraday-volatility-breakouts/ 2026-05-05T00:00:00Z Trading strategies fail due to execution design. This case study examines an intraday volatility breakout strategy that achieved a Sharpe ratio of -1.09, revealing how tight stop placement and frequent exits create spread bleed that defeats alpha capture. The Filter Trap — How Conviction Filters Made a Bad Signal Look Good https://www.radiusred.uk/blog/posts/2026-05-03-filter-trap-orb/ 2026-05-03T00:00:00Z When we added conviction filters to our ORB system, in-sample performance soared. Out-of-sample, everything collapsed. This is the story of why filters can mask a lack of real edge. Textbook Trend-Following on the Wrong Instrument — Why Donchian Channels Failed on Gold https://www.radiusred.uk/blog/posts/2026-05-02-donchian-gold-channels/ 2026-05-02T00:00:00Z Gold is supposed to trend. Donchian channels are built for trends. So why did both channel variants fail to deliver alpha during the US session? When the best edge rationale is not enough https://www.radiusred.uk/blog/posts/2026-05-01-when-edge-rationale-isnt-enough/ 2026-05-01T00:00:00Z The IBS mean-reversion strategy had the strongest academic case of anything we tested. Discovery Sharpe 2.65. Validation Sharpe -0.62. This is the story of why. The First US-Session Gap — Spike-Fade and the Cost of Naive Breakouts https://www.radiusred.uk/blog/posts/2026-05-01-spike-fade-naive-breakouts/ 2026-05-01T00:00:00Z Our mean-reversion portfolio was silent during the US session. We tested breakout strategies to fill the gap. The spike-fade pattern taught us why it failed. Why a winning Gold momentum idea failed on Brent and Silver https://www.radiusred.uk/blog/posts/2026-04-27-why-gold-momentum-failed-on-brent-and-silver/ 2026-04-27T00:00:00Z The same strategy archetype that fit Gold broke on Brent crude and Silver. The lesson was about intraday microstructure, not commodities in general. tradedesk v1.0.1 Fixes a Startup Retry Edge Case https://www.radiusred.uk/blog/posts/2026-04-26-tradedesk-v1-0-1-fixes-a-startup-retry-edge-case/ 2026-04-26T00:00:00Z A short release note on the tradedesk v1.0.1 patch release, covering the IG chart subscription retry fix and two streamer cleanups. What an open-source 1.0.0 looks like when an upstream API moves https://www.radiusred.uk/blog/posts/2026-04-25-open-source-1-0-0-upstream-api/ 2026-04-25T00:00:00Z Why tradedesk 1.0.0 shipped now, what changed in IG streaming support, and how a release workflow bug became part of the work. Phase 5 Quant Loop: From Research Consensus to Implementation Reality https://www.radiusred.uk/blog/posts/2026-04-18-phase-5-quant-loop-from-research-consensus-to-implementation-reality/ 2026-04-18T00:00:00Z An inside look at the checks between a promising backtest and something that is actually safe enough to earn a DEMO slot. When Research Says Go but Implementation Says Wait https://www.radiusred.uk/blog/posts/2026-04-16-when-research-says-go-but-implementation-says-wait/ 2026-04-16T00:00:00Z A process-focused look at how we handle promising research, implementation reality checks, and explicit decision gates.