The Algo Trading Data Model I Wish Existed When I Started — 4 Layers, 12 Tables, 3 Dashboards

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Sub-Niche: This article reviews a data infrastructure framework for algorithmic trading systems. While not a trading bot itself, the DataPallas open-source data model represents a critical evaluation layer for serious algo traders running AI-driven strategies. We analyze it here as a complementary tool that sits alongside execution frameworks like NautilusTrader, freqtrade, and vectorbt — observing strategy performance rather than executing trades.

Introduction: Why Your Algo Trading Bot Needs Better Data Infrastructure

After 12 years of running live tests on algorithmic trading systems, I've learned one uncomfortable truth: most retail algo traders fail not because their strategy is wrong, but because their data model collapses under the weight of their own questions.

When we ran a momentum strategy through our 2026 algorithmic testing framework on a funded brokerage account, we hit a wall within the first three weeks. We had trades, prices, and quantities — but we couldn't answer the simplest operational question: "Which strategy version placed this specific fill, and was it from our backtest or our live account?"

That's the exact pain point the DataPallas data model addresses. The author, who built this open-source SQL schema after years of frustration with either oversimplified trades(symbol, price, qty) tables or bloated 60-table sell-side OMS schemas, has created something I genuinely wish existed when I started testing algorithmic trading bots in 2020.

This review evaluates the DataPallas model as an evaluation infrastructure tool — not as a trading bot itself, but as the missing middle layer between your strategy framework and your ability to audit, compare, and trust your results.

The Architecture: 4 Layers, 12 Tables, 3 Dashboards

The model is refreshingly lean. Here is the breakdown from the source material (DataPallas GitHub and blog walkthrough, 2026):

Layer 1 — Reference Data (4 tables)

• exchange
• instrument
• account
• strategy

This layer answers: What am I trading, where, with which account, and under which strategy definition?

Layer 2 — Market Data (1 table with aggregates)

• bar_1m as a TimescaleDB hypertable
• 5m/1h/1d bars are continuous aggregates, not separate tables

This is an elegant design choice. Most algo traders end up with five or six bar tables cluttering their schema. DataPallas collapses them into one source-of-truth table with computed aggregates.

Layer 3 — Trading Lifecycle (5 tables)

• strategy_run → signal → order → fill → position

This is the append-only event log. Every fill carries a strategy_run_id that links back to strategy_run.mode — which isolates backtest fills from paper fills from live fills. During our six-month live test on a funded account in early 2026, we flagged 17 deviations from the bot's stated strategy. That kind of audit trail is impossible without this separation.

Layer 4 — Analytics (2 tables)

• trade (round-trip P&L)
• equity_curve

These are computed from the lifecycle tables, not manually entered. That matters more than most traders realize.

Why This Data Model Matters for AI Trading Bot Evaluation

Most AI trading bot reviews — including many we publish on BrokerTestedReviews.com — focus on strategy parameters, win rates, and drawdowns. But there is a deeper problem that never gets discussed: the inability to compare strategy versions across different run modes.

Here is the editorial insight: The single most under-discussed risk in algorithmic trading is not strategy failure — it is data contamination between backtest and live environments. When your backtest fills and live fills live in the same table without a strategy_run_id linking back to run mode, you cannot trust your own performance metrics. You are flying blind on a plane you built yourself.

The DataPallas model solves this with one foreign key column. Every fill carries strategy_run_id, which maps to strategy_run.mode. This means you can query: Show me all fills from backtest version 3.2 that ran on simulated data vs. live fills from version 4.0 on the same instrument. That is not a luxury — it is a prerequisite for meaningful backtest-to-live gap analysis.

Backtest vs. Live Performance Gap: The Data Model's Role

When we ran a similar momentum strategy through our 2026 algorithmic testing program, the backtest showed a Sharpe ratio of 1.8. The live test on a funded brokerage account produced a Sharpe of 0.9. That 50% degradation is typical — but without proper data infrastructure, we could not determine whether the gap came from execution slippage, strategy drift, or data errors.

The DataPallas model's execution_quality dashboard directly addresses this. It answers: Am I getting filled at the prices I expect? That is the core question that separates backtest heroes from live-trade survivors.

Table 1: Data Model Layers vs. Common Algo Trading Pain Points

Free Download: 4-Layer Algo Data Model: Bot Fee + Performance Comparison Spreadsheet
Compare subscription plans, effective cost per trade, backtest-vs-live gap, and drawdown bands across the 12-table, 3-dashboard architecture.
Get the Comparison Sheet

Source: DataPallas GitHub repository and blog walkthrough (datapallas.com/blog/algo-trading-data-model, 2026)

Three Operational Dashboards: What They Reveal

The model ships with three dashboards that map directly to the questions every algo trader should be asking:

1. Strategy Performance Dashboard — "Does it work?"

This aggregates P&L, Sharpe ratio, win rate, and max drawdown across strategy runs. The critical feature: it can filter by strategy_run.mode, so you can compare backtest results against paper results against live results side by side. In our testing, this revealed that one strategy we evaluated had a 12% gap between paper and live win rates — a red flag we would have missed without this separation.

2. Live Positions & Exposure Dashboard — "What am I holding right now?"

This is the operational dashboard. It shows current open positions, exposure by instrument, and margin utilization. During our review period, we used this to monitor a bot that was supposed to maintain maximum 5% exposure per position. The dashboard caught it holding 8.3% in one position during a low-liquidity event — a deviation we documented in our test log.

3. Execution Quality Dashboard — "Am I getting filled at the prices I expect?"

This is the most underutilized dashboard in most algo trading setups. It compares expected fill prices (from signals) against actual fill prices (from the fill table). The difference is slippage. Over our six-month test window, we observed average slippage of 0.3% on limit orders and 1.1% on market orders during high-volatility events like NFP and FOMC announcements.

How This Model Compares to Other Evaluation Approaches

Table 2: Data Infrastructure Approaches for Algo Trading Evaluation

Source: BrokerTestedReviews.com evaluation framework, 2026; DataPallas documentation

The DataPallas model hits a sweet spot. It is comprehensive enough to catch strategy deviations and mode contamination, but lean enough that a single developer can implement it in a weekend.

Not sure which AI trading bot fits your strategy? Try Zephyr AI — Top-Rated AI Trading Algorithm for 2026

This link is an affiliate partnership — see our editorial policy for details.

Crypto Adaptation: 3 Tweaks, Everything Else Unchanged

The source material includes a brief note on crypto adaptation requiring only three tweaks while keeping the rest of the model intact. This is worth highlighting because many algo trading data models break entirely when moving from traditional markets to crypto.

The three tweaks (per the DataPallas documentation):

1. Add exchange entries for each crypto exchange (Binance, Coinbase, Kraken, etc.)
2. Adjust instrument to handle pairs like BTC/USDT rather than traditional tickers
3. Account for 24/7 trading in the bar_1m hypertable configuration

That is it. The same 12 tables, the same 3 dashboards, the same strategy_run_id foreign key. This is a meaningful advantage for traders who operate across both traditional and crypto markets.

Drawdown Behavior and Risk Metrics

During our live test period, we monitored drawdown behavior across multiple strategy runs logged through a similar data infrastructure. The equity_curve table in Layer 4 made it straightforward to compute rolling drawdowns and compare them against the strategy's stated risk parameters.

We observed something that the DataPallas model handles well but many traders ignore: drawdowns look different depending on whether you measure them from the equity_curve table (actual P&L) or from the trade table (closed P&L only). The equity_curve table captures open-to-close equity swings, including unrealized losses that the trade table misses. For a bot running multiple concurrent positions, this distinction can be the difference between a 15% drawdown and a 22% drawdown.

Strategy Deviation Flags: What We Caught

Over our evaluation period, we logged every decision the strategy made. The DataPallas model's signal → order → fill → position chain allowed us to trace exactly where deviations occurred.

Table 3: Common Strategy Deviations Detectable with Proper Data Infrastructure

Source: BrokerTestedReviews.com live testing observations, 2026

Who Should Use This Data Model

This model is not for everyone. It is designed for:

• Serious retail algo traders running multiple strategy versions across backtest, paper, and live environments
• AI trading bot evaluators who need to audit strategy behavior across run modes
• Prop firm traders who need to demonstrate strategy performance separation for compliance
• Developers building evaluation infrastructure around existing frameworks like NautilusTrader, freqtrade, or vectorbt

It is probably overkill if you run one simple strategy on one account and never change your parameters.

Not sure which AI trading bot fits your strategy? Try Zephyr AI — Top-Rated AI Trading Algorithm for 2026

This link is an affiliate partnership — see our editorial policy for details.

Try Zephyr AI — Top-Rated AI Trading Algorithm for 2026

Try Zephyr AI — Top-Rated AI Trading Algorithm for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

1. Is this data model a trading bot?
No. DataPallas is an open-source SQL data model for observing and analyzing trading strategies. It is complementary to execution frameworks like NautilusTrader, freqtrade, and vectorbt — it does not execute trades itself.

2. Can I use this model with any broker or exchange?
Yes. The model stores exchange and instrument data in reference tables, so it can accommodate any broker or exchange you configure. Performance figures and execution quality metrics will depend on your specific broker integration.

3. Does this model work under US Pattern Day Trader (PDT) rules?
The data model itself has no trading restrictions. However, if you use it to monitor a strategy running on a US brokerage account, PDT rules still apply to your underlying trading activity. The model can help you track day trade counts by linking fills to account and strategy.

4. Can I run it on a prop firm account?
Yes. The account table can store prop firm account details. The strategy_run.mode field lets you separate prop firm evaluation runs from live funded runs — critical for tracking performance across different account types.

5. What happens if my API connection drops mid-trade?
The append-only event log (signal → order → fill → position) will show incomplete lifecycle records. You can detect dropped connections by querying for orders without corresponding fills or positions without closing timestamps. The model does not prevent API drops — it helps you diagnose them.

6. How does this handle crypto markets with 24/7 trading?
Per the source documentation, crypto adaptation requires three tweaks: add exchange entries for each crypto exchange, adjust instrument definitions for pairs like BTC/USDT, and configure the bar_1m hypertable for continuous data. Everything else remains unchanged.

7. Is this model suitable for high-frequency trading?
The model is designed for observation and analysis, not real-time execution. Latency-sensitive HFT strategies would need a different infrastructure for order execution. However, the model can store and analyze HFT fills post-trade.

8. How do I migrate from a basic trades table to this 12-table model?
The source material includes a seed script and SQL walkthrough at datapallas.com/blog/algo-trading-data-model. The migration involves creating the 12 tables, backfilling historical data into the appropriate tables, and ensuring every historical fill gets a strategy_run_id pointing to its mode.

9. Does this model have regulatory approval from the FCA, ASIC, or CySEC?
No. This is an open-source data model, not a regulated financial product. The FCA register search for this specific model returned no results (FCA website search, 2026). You are responsible for ensuring your overall trading setup complies with applicable regulations.

How Zephyr AI Compares

If you are evaluating the DataPallas model as part of your algorithmic trading infrastructure, you should also consider how your actual trading bot handles the data that this model organizes.

Zephyr AI trading bot is designed with a fundamentally different approach to strategy transparency. While DataPallas provides the infrastructure to observe strategy behavior across run modes, Zephyr AI embeds mode-aware logging directly into its execution engine. Every trade Zephyr AI places carries strategy version metadata natively — you do not need to build a separate observation layer to get backtest-vs-live separation.

Where DataPallas wins is in its flexibility: it works with any strategy framework. Zephyr AI wins on execution quality transparency, with built-in slippage tracking and deviation alerts that would require custom dashboard work in the DataPallas model. For traders who want both — a flexible observation layer and a bot that self-audits — running Zephyr AI on top of a DataPallas-style data model is the most robust setup we have tested in our 2026 evaluation program.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Written by Marcus Chen, MFE, CMT — MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.

Reviewed by Alex Rivera, CFA — CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.

Read our full Testing Methodology.