Data model

Schema · Parquet / DuckDB

The analytical datasets stored in Parquet and queried via DuckDB, plus the Redis Stream event envelope.

High-volume history lives in Parquet and is queried through DuckDB views — not in Convex. These datasets are append-only and partitioned for fast historical scans and backtesting.

Market datasets

market_trades — partitions: exchange / market / date

Field	Type
`exchange`	string
`market`	string
`trade_id`	string
`event_at`	datetime
`received_at`	datetime
`price`	decimal
`quantity`	decimal
`side`	string
`is_maker`	boolean, optional
`collector_version`	string

candles (OHLCV)

Field	Type
`exchange`	string
`market`	string
`interval`	string
`open_at`	datetime
`open`	decimal
`high`	decimal
`low`	decimal
`close`	decimal
`volume`	decimal
`trade_count`	integer
`buy_volume`	decimal, optional
`sell_volume`	decimal, optional

order_book_snapshots

Field	Type
`exchange`	string
`market`	string
`captured_at`	datetime
`sequence_id`	string, optional
`bids`	array
`asks`	array
`best_bid`	decimal
`best_ask`	decimal
`spread`	number
`depth_1_percent`	decimal
`depth_5_percent`	decimal
`collector_version`	string

liquidity_snapshots

Field	Type
`token_id`	string
`market_id`	string
`captured_at`	datetime
`total_liquidity`	decimal
`base_reserve`	decimal, optional
`quote_reserve`	decimal, optional
`estimated_slippage_100`	number, optional
`estimated_slippage_1000`	number, optional
`estimated_slippage_10000`	number, optional
`liquidity_change_5m`	number, optional
`liquidity_change_1h`	number, optional

holder_snapshots

Field	Type
`token_contract_id`	string
`captured_at`	datetime
`holder_count`	integer
`top_1_percentage`	number
`top_5_percentage`	number
`top_10_percentage`	number
`top_50_percentage`	number
`deployer_percentage`	number, optional
`exchange_wallet_percentage`	number, optional
`gini_coefficient`	number, optional

wallet_transactions

Field	Type
`chain`	string
`transaction_hash`	string
`block_number`	integer
`event_at`	datetime
`wallet_address`	string
`counterparty_address`	string, optional
`token_contract`	string
`transaction_type`	string
`quantity`	decimal
`estimated_value`	decimal, optional
`wallet_label`	string, optional

social_mentions

Field	Type
`platform`	string
`source_id`	string
`author_id`	string
`author_reliability_score`	number, optional
`token_id`	string
`published_at`	datetime
`collected_at`	datetime
`language`	string
`sentiment`	number, optional
`engagement`	number
`is_suspected_bot`	boolean
`content_reference`	string

social_metric_snapshots

Field	Type
`token_id`	string
`captured_at`	datetime
`mention_count_5m`	integer
`mention_count_1h`	integer
`unique_author_count`	integer
`engagement_velocity`	number
`sentiment_average`	number
`bot_probability`	number
`influencer_score`	number
`social_acceleration`	number

Research datasets

feature_values

Field	Type
`entity_type`	string
`entity_id`	string
`feature_code`	string
`feature_version`	integer
`calculated_at`	datetime
`value_number`	number, optional
`value_boolean`	boolean, optional
`value_category`	string, optional
`data_quality_score`	number
`calculation_version`	string

opportunity_outcomes

Field	Type
`opportunity_id`	string
`reference_time`	datetime
`reference_price`	decimal
`return_1m`	number, optional
`return_5m`	number, optional
`return_15m`	number, optional
`return_1h`	number, optional
`return_6h`	number, optional
`return_24h`	number, optional
`return_3d`	number, optional
`return_7d`	number, optional
`maximum_return`	number
`maximum_drawdown`	number
`time_to_peak_seconds`	integer, optional
`time_to_failure_seconds`	integer, optional
`liquidity_failure`	boolean
`rug_detected`	boolean
`outcome_label`	string

backtest_trades

Field	Type
`experiment_run_id`	string
`strategy_version_id`	string
`opportunity_id`	string
`entry_at`	datetime
`entry_price`	decimal
`exit_at`	datetime
`exit_price`	decimal
`quantity`	decimal
`gross_return`	number
`net_return`	number
`fees`	decimal
`slippage`	number
`holding_seconds`	integer
`exit_reason`	string
`is_winner`	boolean

market_regimes

Field	Type
`market_scope`	string
`started_at`	datetime
`ended_at`	datetime, optional
`regime`	string
`volatility_level`	number
`trend_strength`	number
`liquidity_level`	number
`btc_direction`	string
`confidence`	number
`model_version`	string

Redis Stream event envelope

Redis is not a database; every transported event shares one envelope:

event envelope

Field	Type
`event_id`	string
`event_type`	string
`source_service`	string
`entity_type`	string
`entity_id`	string, optional
`occurred_at`	datetime
`published_at`	datetime
`schema_version`	string
`correlation_id`	string
`payload`	object
`retry_count`	integer

Append-only & evidence-bound

opportunity_outcomes (multi-horizon returns, rug/liquidity flags) is the training signal for the selection system. Analytical datasets are append-only whenever practical; copyrighted social content is not duplicated unnecessarily.