alpha-foundry
Unverified ML strategy on Crypto by Huang-Hg. BotFinder score 18 out of 100.
Formulaic alpha mining: strongly-typed GP + GFlowNet search over a typed DSL, with C++/OpenMP portfolio backtest kernels — crypto perps, US equities, China A-shares
Source: github
BotFinder analysis pending.
alpha-foundry
alpha-foundry — Formulaic Alpha Mining + Backtest Kernels English | 中文 Search formulaic alphas with strongly-typed genetic programming (GP), evaluate them through pure C++ / OpenMP backtest kernels, and fuse them into portfolio weights via deterministic sizing. Three markets are supported: crypto perpetuals, US daily equities, and China A-share daily equities. --- How it works End to end, the system is one pipeline — formula search → backtest evaluation → pool building → sizing fusion — and every layer is search-engine agnostic: 1. DSL (evaluation/) — Formulaic alphas are represented as strongly-typed ASTs. A context-free grammar (CFG + α-Sem-k cost budget) constrains the search space and guarantees syntactic/semantic validity. The operand vocabulary is built automatically from parquet column names (auto-classified into price / volume / feature), so switching markets just means switching columns. The evaluator compiles whole token trees to C (and optionally CUDA) for batch evaluation, with a subtree-hash cache reusing common subtrees. 2. Search (search/) — gp/: strongly-typed GP on D
⚠ No verified equity curve — no track-record source connected.
Drawdown profile
Data unavailable — contact the owner.
Verification ledger
How the score has moved
Recalculated at each data collection. Transparency means showing the bad weeks too.
No score history is stored yet — only the current score is shown.
Reviews & comments
No reviews collected from the source yet.
⚠ No live verification account connected — ask for proof before buying.
Alerts on changes: coming soon
Prop-firm compatibility not provided.
Open-source maintainer on GitHub.
Data-completeness & trust index (not a profitability rating)