Status
Phases
- SetupTeam, roles, tools, workflow
- DesignArchitecture, layer interfaces, model manifest format
- Market dataCrypto and US stock candles through our API
- PipelinePreprocessing, models, post-processing and backtests in the web app
- Architecture comparisonTime-series architectures trained and compared on the same backtests
- StabilityWhich combination stays stable across market conditions
- SharingPublic experiments, API keys and an MCP server
Done
- —GitHub organization, website and the domain candlestack.tech
- —Staging and production with automatic deploys
- —Market data API for crypto (Binance) and US stocks (Alpaca)
- —Backtest engine, checked against the reference results of the BLESK models
- —Sign-in with GitHub and Google, account settings
- —Model upload (Keras and ONNX) with checks of the input and output shapes
- —Preprocessing: Renko, indicators, normalization and input windows
- —Post-processing: calibration, thresholds, smoothing, k-means, minimum holding, position size, stop loss
- —Backtest builder and result page with metrics and charts
- —Compare backtests side by side
In progress
- —Comparison of time-series architectures (layer 2)
- —Tests of the web app pages
Planned
- —Stability of each combination across market conditions
- —Kagi representation and segmentation
- —Public catalog of published experiments
- —API keys and an MCP server
- —More benchmarks (SPY, QQQ, DIA)
Last updated 02.11.2026