dsh-quantstats — performance tearsheets for DeepSeek Harness, honesty included
Sharpe, Sortino, Calmar, drawdown, VaR — computed deterministically from your returns, rendered as a zero-dependency SVG tearsheet, with the assumptions disclosed and the sample-size alarm on. The report that states its own limits. 夏普/索提诺/卡尔玛/回撤/VaR——从收益序列确定性计算,零依赖 SVG tearsheet 呈现,假设显式披露、样本量警报常开。会声明自身边界的报告。
Implements the core concept of QuantStats (7.9k★, by the yfinance author) natively for DSH. Not affiliated. Family: dsh-TradingAgents · dsh-ai-hedge-fund · dsh-Vibe-Trading · dsh-OpenBB · dsh-CCXT · dsh-nautilus_trader · dsh-hummingbot · dsh-jesse.
Why (QuantStats exists — what does the DSH-native version add?)
| QuantStats library | dsh-quantstats | |
|---|---|---|
| Runs where | Python lab / notebooks | DSH in-session — agents tearsheet any family engine's output |
| Report | matplotlib HTML (CDN deps) | zero-dependency inline SVG — offline, shareable |
| Assumptions | stated in docs | stated on every report: annualization factor, historical-VaR method |
| Small samples | — | sample-size alarm: n<30 flags Sharpe/Sortino as noise |
Quick start
git clone https://github.com/vicfei/dsh-quantstats && cd dsh-quantstats
python3 scripts/tearsheet.py --data <klines.json> --save # buy & hold
python3 scripts/tearsheet.py --data <klines.json> --strategy sma --fast 20 --slow 60 --save
Real smoke (binance BTC/USDT 365d, 2026-11-17) — the bear-year in one table:
| buy & hold | SMA 20/60 | |
|---|---|---|
| total | −29.9% | +1.6% |
| vol(ann) | 44.3% | 25.9% |
| max DD | −39.5% | −19.3% |
| Sharpe | −0.58 | 0.19 |
Assumptions disclosed on both (annualization=365, historical VaR, n=364).
Status & roadmap
v0 tearsheet engine smoke-tested on live data. ROADMAP: M1 rolling metrics + benchmark-relative mode; M2 plugin form; M3 report standard.
License & disclaimer
MIT; concept inspired by Apache-2.0 upstream — no upstream code used. Not investment advice. Past performance ≠ future results.
No comments yet. Be the first to write one.