Product Update August 2026: Backtesting Engine V2,
Core Progress in August 2026
In August 2026, TradeMates focused on two major engineering initiatives: institutional-grade historical data accuracy and infrastructure cost reductions for scheduled precomputations. In parallel, the Community hub was overhauled with a dense, telemetry-focused interface (TradeMates Architecture 2026).
Key Takeaways
- Backtesting Engine V2: Transitioned to split- and dividend-adjusted EOD data via Financial Modeling Prep (FMP) with mandatory 90-day cohort maturity gates.
- Per-Ticker Learning Loop: Direct injection of historical post-mortems into future evaluation prompts for the same equity.
- Hetzner Compute Pipeline: Offloaded heavy background prewarming workflows to dedicated infrastructure, slashing external LLM costs.
- Community Redesign: Dense instrument-panel layout with volume sparkbars, sentiment distributions, and uncapped stock research counts.
- Score Journey Card: Transparent visual timelines demonstrating gains captured and capital losses averted by TradeMates signals.
- Retention Automation: 14-day personalized re-engagement journeys and curated daily market intelligence digests.
Institutional Rigor: Backtesting Engine V2
Initial backtesting operations highlighted the necessity of comprehensive corporate action handling. Version 2 deployed key enhancements:
- Dividend & Split Adjustments: Historical candle loaders integrate adjusted closing prices, eliminating artificial price drops caused by stock splits.
- Cohort Maturity Gating: Marketing and UI surfaces withhold accuracy claims until cohorts complete their full 90-day trajectory.
- Administrative Control Center: Real-time diagnostics monitor error distributions and model divergence across thousands of archived snapshots.
Furthermore, the Per-Ticker Learning Loop extracts historical miss lessons and supplies them into subsequent prompts as grounded counter-heuristics.
Self-Hosted Compute: Dedicated Hetzner Prewarms
Maintaining instant report access requires background computation for high-volume market tickers. To eliminate external token overhead, TradeMates deployed hybrid compute topology:
- Dedicated Inferenz Workers: Scheduled prewarms execute on self-hosted infrastructure.
- Shadow-Scoring Parity: Telemetry streams compare self-hosted outputs against cloud frontier models to ensure scoring parity.
- Transactional Publishing: Results commit to the production cache only after passing strict validation schemas.
Community Hub: NASA-Grade Telemetry
The community research surface evolved away from bloated card grids into an information-dense workspace:
- Rank & Momentum Indicators: Highlights top tickers experiencing surges in quantitative scrutiny.
- Volume Sparkbars & Sentiment Bars: Visualizes collective Buy/Hold/Avoid sentiment at a glance.
- Uncapped Discovery Totals: Upstream limits were lifted to report the genuine count of community-audited assets.
Frequently Asked Questions (FAQ)
Are cash dividends incorporated into alpha calculations?
Yes. Engine V2 calculates total returns factoring in cash dividend payouts and corporate recapitalizations.
Does self-hosted inferencing reduce user-requested report quality?
No. User-initiated live analyses continue running on premier cloud reasoning models. The dedicated cluster services scheduled caching.
How does the architecture mitigate survivorship bias?
Snapshots freeze at analysis execution time. If an analyzed firm delists, its terminal drawdown remains permanently recorded in the backtesting dataset.