dsh-Kronos — K-line scenario forecasting for DeepSeek Harness
A seeded, reproducible scenario fan over future K-lines from real history — the statistical baseline that any neural foundation model will have to beat. Zero dependencies, determinism by seed, disclosure on every payload. 从真实历史生成可复现的未来 K 线情景扇——任何神经基础模型都必须先打赢的 统计基线。零依赖、按种子确定性、披露随载荷。
The concept slot of Kronos — the 47.9k★ "foundation model for the language of financial markets" — natively in DSH. v0 ships the statistical baseline (stationary block bootstrap), not the neural model: honest disclosure rides on every payload, and when the real model lands in M1 (
--via kronos, HF weights), this baseline becomes its control anchor — the network must beat bootstrap to count. Not affiliated. Family: dsh-TradingAgents · dsh-OpenBB · dsh-CCXT · dsh-nautilus_trader · dsh-jesse · dsh-quantstats · dsh-QuantLib · dsh-hummingbot · dsh-Vibe-Trading · dsh-ai-hedge-fund.
Why (Kronos is a neural model — why ship a bootstrap first?)
| Kronos (neural FM) | dsh-Kronos v0 (baseline) | |
|---|---|---|
| Deps | torch + HF weights (~GB) | zero (stdlib only) |
| Determinism | sampling seeds | seeded bootstrap — bit-exact reproducible |
| Where | Python/GPU lab | DSH in-session — agents get scenario fans in the loop |
| Role | learned conditional structure | the anchor any learned model must beat — and the honest fallback when weights/network aren't there |
Quick start
git clone https://github.com/vicfei/dsh-Kronos && cd dsh-kronos
python3 scripts/scenario.py --data <klines.json> --horizon 30 --paths 1000 --seed 42 --save
Real smoke (binance BTC/USDT 365d, 2026-11-17, seed 42): 30-step fan with terminal P10/P50/P90 = 53.1k / 62.6k / 73.2k and P(above current) = 38.9% — the market's own history prices the downside wider than the upside. Reproduce bit-exact with the same seed.
Status & roadmap
v0 baseline engine smoke-tested on live data. ROADMAP: M1 the real
model (--via kronos, HF-mirror weights, CPU inference) + the head-to-head
benchmark vs this baseline.
License & disclaimer
MIT; concept slot of an MIT upstream — no upstream code used. Scenarios are not predictions; not investment advice.
No comments yet. Be the first to write one.