arc-t-defi-pack / SCHEMA.md
solsticestudioai's picture
Add ARC-T DeFi sample (10K lifecycles) with README, SCHEMA, parquet, JSONL
4c7c761 verified
# ARC-T DeFi Decision Telemetry Pack — Schema
One row = one complete risk-to-execution lifecycle. All records share the same seven top-level fields.
Schema version: `1.0.0-arc-t-sample`
## Top-level fields
### `schema_version` — string
Schema identifier. Constant within a sample release. Used to detect pack version drift.
### `event` — struct
Identifier fields, decision outcome, and the PnL delta for the lifecycle.
| Field | Type | Notes |
|---|---|---|
| `id` | string | Stable event identifier, e.g., `ARCT-100000`. |
| `trace_id` | string (UUID) | Cross-links telemetry steps within the lifecycle. |
| `timestamp` | string (ISO-8601) | Lifecycle start time. |
| `decision_outcome` | string | One of: `alpha_captured`, `risk_mitigated`, `slippage_loss`, `execution_failed`. |
| `pnl_delta_usd` | double | Realized PnL delta attributable to this lifecycle. Can be negative. |
### `risk_context` — struct
Environmental and market context plus the triggering anomaly.
| Field | Type | Notes |
|---|---|---|
| `trigger` | string | Triggering risk event. One of: `oracle_staleness`, `tvl_crash`, `whale_dump`, `governance_attack`, `bridge_congestion`, `funding_dislocation`, `borrow_rate_spike`, `liquidity_fragmentation`, `redemption_run`, `liquid_staking_dislocation`. |
| `protocol` | string | DeFi protocol (e.g., `Aave-V3`, `Uniswap-V3`, `Curve-Fi`, `GMX`, `MakerDAO`, `Lido`, `Drift`, `Kamino`, `Jupiter`, `Orca`, `Marinade`, `Synthetix`, `Meteora`). |
| `chain` | string | L1/L2 chain: `ethereum`, `arbitrum`, `solana`, `optimism`. |
| `impacted_asset` | string | Ticker of the affected asset (e.g., `WETH`, `USDC`, `stETH`, `SOL`). |
| `anomaly_signature` | string | Human-readable signature summarizing the anomaly. |
| `market_state` | struct | See below. |
`market_state` struct:
| Field | Type | Notes |
|---|---|---|
| `severity` | double | 0–1 severity score. |
| `volatility_regime` | string | `calm`, `normal`, `stress`, etc. |
| `liquidity_band` | string | `thin`, `normal`, `deep`. |
| `oracle_age_sec` | int | Age of the oracle feed in seconds. |
| `venue_divergence_bps` | int | Cross-venue price divergence, in basis points. |
| `estimated_price_impact_bps` | int | Expected price impact of a trade, in basis points. |
| `order_notional_usd` | double | Order size in USD. |
| `liquidity_usd` | double | Available pool liquidity in USD. |
| `volatility_1h_bps` | int | 1-hour realized volatility, in basis points. |
| `bridge_delay_sec` | int | Cross-chain bridge delay, in seconds. |
| `open_interest_delta_pct` | double | Change in open interest, in percent. |
### `agent_reasoning` — struct
Planner-side metadata for the autonomous agent selecting the strategy.
| Field | Type | Notes |
|---|---|---|
| `engine` | string | Planner engine label. In this sample: `arct_planner`. |
| `winning_strategy` | string | Selected top-level strategy. One of: `Immediate_Hedge`, `TWAP_Exit`, `Cross_Venue_Sweep`, `Arb_Rebalance`, `Priority_Fee_Exit`, `Collateral_Rotation`, `Delta_Neutral_Reprice`, `Liquidity_Provision`. |
| `confidence_score` | double | Planner confidence at selection time. |
| `mcts_branches` | int | Number of MCTS branches considered. |
### `correlated_telemetry` — list<struct>
Ordered sequence of internal component actions during execution. One struct per step.
Step struct:
| Field | Type | Notes |
|---|---|---|
| `timestamp` | string (ISO-8601) | Step timestamp. |
| `component` | string | Which internal component emitted the step (e.g., `ARES`, `FRACTAL`, `ARGUS`, `SENTINEL`, `EXECUTION_ENGINE`). |
| `action` | string | Action code (e.g., `ARES_WHALE_ALARM`, `FRACTAL_MARKET_IMPACT_REESTIMATE`, `ARGUS_ANOMALY_CONFIRMED`, `SENTINEL_ORDERBOOK_SCAN`). |
| `telemetry.latency_ms` | int | Step execution latency in milliseconds. |
| `telemetry.slippage_bps` | int | Observed slippage in basis points. |
| `telemetry.gas_price_gwei` | int | Gas price observed during the step (gwei). |
| `telemetry.priority_fee_gwei` | int | Priority fee (gwei). |
| `telemetry.route_hops` | int | Number of hops in the selected route. |
| `telemetry.estimated_price_impact_bps` | int | Step-level price impact estimate (bps). |
| `telemetry.venue_divergence_bps` | int | Venue-vs-venue price divergence observed at the step (bps). |
| `telemetry.node_provider` | string | RPC provider observed (e.g., `Alchemy`, `Local_Geth`, `Triton`, `Solana-RPC`). |
### `execution_summary` — struct
Roll-up metrics for the full lifecycle.
| Field | Type | Notes |
|---|---|---|
| `strategy` | string | Final strategy label executed. |
| `total_execution_ms` | int | End-to-end lifecycle duration. |
| `gas_cost_usd` | double | Realized gas cost in USD. |
| `avg_slippage_bps` | double | Average observed slippage across steps, in basis points. |
### `genetic_optimizer_feedback` — struct
Outer-loop optimizer feedback used to tune future planner iterations.
| Field | Type | Notes |
|---|---|---|
| `fitness_score_update` | double | Delta applied to the strategy's fitness score. |
| `parameter_drift` | string | Coarse drift descriptor (`none`, `minor`, `major`, …). |
## Distribution of this sample
- 10,000 lifecycles, stratified 2,500 per decision_outcome across all four outcomes.
- Balanced (within shuffled sampling) across 10 risk triggers (~1,000 each).
- Balanced across 4 chains (weighted toward `ethereum`).
- Broad protocol coverage (13 protocols).
- 8 winning strategies observed.
- Planner engine label constant (`arct_planner`) after sanitization of internal method identifiers.
## Sanitization notes
- Internal identifier prefix (`SIMA-V4-ARC-*`) has been normalized to `ARCT-*`.
- Internal planner engine code name (`QuantumImagination_V4`) has been normalized to `arct_planner`.
- No real wallet addresses, transaction hashes, pool reserves, or on-chain identifiers are present.
## Relationship to the full pack
The production pack scales to 2.5M+ lifecycles with expanded chain and protocol coverage, finer-grained market-state regimes, MEV-aware telemetry, campaign-linked multi-step attack sequences, richer step-level component traces, and gym-compatible delivery. See the pack card for commercial access.