One offline launch path for financial-agent reliability: lifecycle traces, execution realism, hard market rules, crisis-scene visuals, and cached model replay metadata.
Open one replayable observe-plan-risk-act-reflect trajectory with fills, rejections, memory, and reproducibility state.
Inspect intent versus executed weights, slippage attribution, and the risk intervention timeline.
Inspect representation trajectory, correlation heatmap, feedback curves, and exposure waterfall snapshots from tracked diagnostic artifacts.
See how A-share history is normalized into the same OHLCV CSV boundary used by every other TreLLM data provider.
See T+1, price-limit, and board-lot constraints become clipped or blocked risk-gate outcomes.
Compare a risk-aware realistic agent against buy-and-hold under the same execution frictions.
Check provider/model coverage, relative timestamp masking, prompt modes, and parsed signal counts without shipping raw prompts.
Inspect the tiny CSV adapter example for optional news, macro, filings, and alternative-data fields.
Generated by python scripts/run_launch_demo.py. None of these examples calls DeepSeek, Poe, OpenAI, or live market-data APIs.
python scripts/run_launch_demo.py