arc-t-defi-pack / README.md
solsticestudioai's picture
Add ARC-T DeFi sample (10K lifecycles) with README, SCHEMA, parquet, JSONL
4c7c761 verified
---
license: cc-by-4.0
task_categories:
- tabular-classification
- tabular-regression
- reinforcement-learning
language:
- en
tags:
- synthetic
- defi
- decentralized-finance
- trading
- risk-detection
- on-chain
- mev
- liquidations
- autonomous-agents
- mcts
- reinforcement-learning
- blockchain
- agentic-ai
pretty_name: ARC-T DeFi Decision Telemetry Pack
size_categories:
- 10K<n<100K
configs:
- config_name: default
data_files:
- split: train
path: arc_t_defi_sample.parquet
---
# ARC-T DeFi Decision Telemetry Pack (Sample)
**A synthetic DeFi risk-to-execution decision-telemetry dataset for autonomous-trading research, on-chain anomaly detection, and reinforcement-learning environments.** Each row is a complete risk-to-execution lifecycle: a triggering market anomaly, market-state snapshot, adversarial-planner reasoning (MCTS branch count + winning strategy), step-by-step execution chain with per-step gas / slippage / latency telemetry, execution summary, and final decision outcome with PnL delta.
Built by [SolsticeAI](https://www.solsticestudio.ai/datasets) as a free sample of a larger commercial pack. 100% synthetic. No real trades, wallets, addresses, or on-chain identities. Safe for model training, competitive evaluation, and public benchmark work.
## What is included
| File | Rows | Format | Purpose |
|---|---:|---|---|
| `arc_t_defi_sample.parquet` | 10,000 | Parquet | Columnar, typed, best for analytics |
| `arc_t_defi_sample.jsonl` | 10,000 | JSON Lines | Streaming / LLM training friendly |
**Source pack:** 100K-lifecycle corpus (production pack scales to 2.5M+)
**This sample:** 10,000 lifecycles, stratified 2,500 per decision outcome
**Outcome classes:** `alpha_captured`, `risk_mitigated`, `slippage_loss`, `execution_failed`
**Chains covered:** `ethereum`, `arbitrum`, `solana`, `optimism`
**Protocols covered:** Aave-V3, Uniswap-V3, Curve-Fi, GMX, MakerDAO, Drift, Lido, Kamino, Jupiter, Orca, Marinade, Synthetix, Meteora
**Risk triggers:** oracle staleness, TVL crash, whale dump, governance attack, bridge congestion, funding/borrow-rate dislocation, liquidity fragmentation, redemption run, liquid-staking dislocation
## Record structure
Each record is one risk-to-execution lifecycle with 7 top-level fields:
| Field | Type | Contents |
|---|---|---|
| `schema_version` | string | Pack schema version (`1.0.0-arc-t-sample`) |
| `event` | struct | `id`, `trace_id`, `timestamp`, `decision_outcome`, `pnl_delta_usd` |
| `risk_context` | struct | `trigger`, `protocol`, `chain`, `impacted_asset`, `anomaly_signature`, nested `market_state` (severity, volatility_regime, liquidity_band, oracle_age, venue divergence, price impact, notional, 1h vol, etc.) |
| `agent_reasoning` | struct | `engine`, `winning_strategy`, `confidence_score`, `mcts_branches` |
| `correlated_telemetry` | list<struct> | Ordered component chain (`ARES`, `FRACTAL`, `ARGUS`, `SENTINEL`, …) with per-step latency, slippage, gas price, priority fee, route hops, node provider |
| `execution_summary` | struct | `strategy`, total execution time, gas cost, average slippage |
| `genetic_optimizer_feedback` | struct | `fitness_score_update`, `parameter_drift` |
See [SCHEMA.md](./SCHEMA.md) for the full nested field breakdown.
## Why this dataset is useful
Most public on-chain datasets are either raw block-level trace data or narrow single-protocol slices. This pack is shaped around what autonomous-trading and on-chain-risk teams actually need to train decision models:
- Complete risk-to-execution lifecycles rather than isolated transactions
- Balanced outcome classes across winning and losing execution states
- Adversarial reasoning trace (strategy + MCTS branch count + confidence) alongside the telemetry
- Per-step gas, slippage, and latency signals to train execution-aware policies
- Multi-chain, multi-protocol coverage for generalization across venues
- PnL-linked outcomes for reward shaping in RL environments
- Stable schema suitable for backtesting, RL gym integration, and dashboarding
## Typical use cases
- Autonomous DeFi agent training (RL and supervised)
- On-chain anomaly detection and risk-trigger classification
- Execution-strategy optimization (gas / slippage / route)
- Arbitrage and cross-venue routing model development
- Market-microstructure and liquidity-state modeling
- LLM fine-tuning on execution narratives and decision rationale
- Benchmarking MCTS strategy selection under uncertainty
- Gym environment authoring for multi-chain trading
## Quick start
```python
import pandas as pd
import pyarrow.parquet as pq
df = pq.read_table("arc_t_defi_sample.parquet").to_pandas()
# Outcome distribution (stratified balanced)
print(df["event"].apply(lambda e: e["decision_outcome"]).value_counts())
# Average PnL by strategy
df["strategy"] = df["agent_reasoning"].apply(lambda r: r["winning_strategy"])
df["pnl"] = df["event"].apply(lambda e: e["pnl_delta_usd"])
print(df.groupby("strategy")["pnl"].mean().round(2))
# Chain vs outcome cross-tab
df["chain"] = df["risk_context"].apply(lambda r: r["chain"])
df["outcome"] = df["event"].apply(lambda e: e["decision_outcome"])
print(pd.crosstab(df["chain"], df["outcome"]))
```
Streaming form:
```python
import json
with open("arc_t_defi_sample.jsonl") as f:
for line in f:
lifecycle = json.loads(line)
# one risk-to-execution lifecycle per line
```
## Responsible use
This dataset is intended for **research, prototyping, and defensive / monitoring** use cases around autonomous DeFi systems: anomaly detection, strategy evaluation, execution-policy training, and RL environments. It contains synthesized market states, execution traces, and decision outcomes — it does **not** contain real wallet addresses, real transaction hashes, real pool reserves, or any live on-chain state. Do not deploy policies trained solely on this synthetic data to real capital without independent validation against live data.
## License
Released under **CC BY 4.0**. Use freely for research, RL experiments, education, and commercial prototyping with attribution.
## Get the full pack
This Hugging Face repo is a **10K-lifecycle sample**. The production pack scales to 2.5M+ lifecycles with expanded chain and protocol coverage, finer-grained market-state regimes, campaign-linked multi-step attack sequences, MEV-aware telemetry, richer step-level component traces, parquet + JSONL + gym-compatible formats, and buyer-specific variants.
**Self-serve (Stripe checkout):**
- [**Sample Scale tier — $5,000**](https://buy.stripe.com/7sY5kD2j85QTfSb5lfeEo03) — ~25K records, one subject, 72-hour delivery.
**Full pack + enterprise scope:**
- [www.solsticestudio.ai/datasets](https://www.solsticestudio.ai/datasets) — per-SKU pricing across Starter / Professional / Enterprise tiers, plus commercial licensing, custom generation, and buyer-specific variants.
**Procurement catalog:**
- [SolsticeAI Data Storefront](https://solsticeai.mydatastorefront.com) — available via Datarade / Monda.
## Citation
```bibtex
@dataset{solstice_arc_t_defi_pack_2026,
title = {ARC-T DeFi Decision Telemetry Pack (Sample)},
author = {SolsticeAI},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/solsticestudioai/arc-t-defi-pack}
}
```