nifty-session-logs

Nifty 50 AI Agent — Kaggle Session Logs & Accuracy Telemetry

This dataset repository contains structured session telemetry, real-time trade signals, risk gate evaluations, agent decisions, and accuracy improvement reports generated by the Nifty AI Brain Agent running autonomously on Kaggle dual T4 GPU environment.

Repository Structure

nagarhimanshu37/nifty-session-logs
├── README.md                      ← Telemetry overview & session log index
├── latest_report.json             ← Latest session accuracy & weakness report
└── sessions/
    └── <session_id>/
        ├── session_meta.jsonl     ← Session start/end timestamps & runtime metadata
        ├── session_summary_raw.json ← Quick session totals
        ├── accuracy_report.json   ← Computed accuracy, win rates & weakness analysis
        ├── improvement_notes.md   ← Human-readable action items for next prompt/code edit
        ├── signals.jsonl          ← Every generated AlphaSignal before risk gate
        ├── trades.jsonl           ← Every approved trade + execution/outcome
        ├── vetoed_trades.jsonl    ← Every trade rejected by QuantitativeRiskAgent
        ├── agent_decisions.jsonl  ← Sub-agent cycle steps and market scans
        ├── heartbeats.jsonl       ← 24/7 supervisor health snapshots
        └── errors.jsonl           ← Anomaly reports & runtime exceptions

Purpose

1. Continuous Learning: Track which setups (CPR Breakout, VWAP Rejection, Trap Hunter) perform best.
2. Defect & Weakness Isolation: Automatically identify why signals fail (e.g. low VIX chop, opening 15-min noise, wrong delta).
3. Agent Tuning: Use improvement_notes.md to refine strategy thresholds, risk weights, and LLM reasoning prompts.

Automatically synced from Kaggle dual T4 runners via HFSessionPusher.