Within the expected cadence
Crypto is checked against calendar days. Stocks use a weekday-aware check so weekends do not automatically make data stale.
NeuralTrend is designed as a market-research and simulation platform. Here you can find the information needed to evaluate its signals, historical results, assumptions, and limitations.
Historical simulations are not brokerage statements, real-money results, or guarantees of future performance. A separately labeled Forward Record will begin from a clearly stated date when NeuralTrend is released publicly; pre-launch testing will not be imported.
This is the same historical simulation framework available in Signal Overview, fixed to the 1Y horizon for four representative crypto assets. It is not a separate validation dataset and should not be confused with the approved, date-bounded Forward Record.
Explore all assets and horizons in Signal Overview →| Asset | Coverage | AI strategy | Buy & Hold | Return spread | AI max drawdown | Data freshness | Latest signal |
|---|---|---|---|---|---|---|---|
| BTC-USD 366 observations | 2025-09-07 – 2026-09-07 | +12.9% | -29.3% | +42.3 pts | -28.7% | Current Through 2026-09-07 | HOLD |
| ETH-USD 366 observations | 2025-09-07 – 2026-09-07 | +17.7% | -42.8% | +60.5 pts | -35.4% | Current Through 2026-09-07 | HOLD |
| SOL-USD 366 observations | 2025-09-07 – 2026-09-07 | +54.9% | -49.9% | +104.8 pts | -12.7% | Current Through 2026-09-07 | HOLD |
| XRP-USD 366 observations | 2025-09-07 – 2026-09-07 | -6.8% | -51.9% | +45.0 pts | -6.8% | Current Through 2026-09-07 | HOLD |
Generated from the currently deployed data files using the 1Y Signal Overview horizon. Coverage and data-through dates are shown per asset.
NeuralTrend intentionally avoids using incomplete same-day market information when producing the daily signal shown to users.
Daily price and volume data are collected after a market day is complete. Technical and other indicators are derived from that information.
The asset’s model evaluates information available through the previous completed day. It does not use today’s price.
The resulting BUY, HOLD, or SELL state is associated with the current signal date and is used by the site’s historical and live simulations.
NeuralTrend displays a data-through date, the time the deployed site data file was updated, and a simple freshness state. These indicators describe the website data available to the model; they are not real-time exchange timestamps or trade confirmations.
Crypto is checked against calendar days. Stocks use a weekday-aware check so weekends do not automatically make data stale.
The latest completed row is one update beyond the normal tolerance. Users should verify the data-through date before acting.
The latest completed row is several expected updates behind. The signal remains visible for transparency but should not be treated as current.
Stock-market holidays are not embedded in the lightweight freshness calculation, so an exchange holiday can occasionally make a current stock file appear delayed.
Historical and current live simulations use the daily closing price stored for the relevant signal date. They are not real-time exchange executions.
The current fixed assumptions are 1.00% per executed crypto side and 0.10% per executed stock side. Actual venue costs may differ.
Positions still open at the end of a selected horizon are marked to market. No artificial final SELL or exit fee is added without a SELL signal.
The benchmark uses the same period and includes an entry cost. Residual cash is retained when a whole-share stock purchase cannot invest every dollar.
Slippage, bid–ask spread variation, DEX gas costs, liquidity-dependent price impact, market impact, taxes, borrow costs, and exchange-specific fee tiers are not modeled separately. Actual trading results can differ materially.
Shows how the signal series would have behaved under the website’s stated assumptions. The complete chart can include periods used during model development as well as later evaluation and production inference.
Calculates performance from compact administrator-approved Date, Close, and signal rows beginning on each asset's stated public start date. Earlier approved rows cannot be silently rewritten through the publication workflow.
Model quality is monitored regularly. Individual asset models are periodically retrained when performance and data-quality checks indicate that an update is appropriate.
Each asset can be updated independently. A model may remain deployed for weeks, months, or longer depending on its observed behavior and available data.
Previously published signal rows are normally preserved when new daily signals are appended. Internal release checks flag unexpected historical changes.
If a correction or data cleanup changes earlier signals, NeuralTrend will identify the affected asset, period, reason, and scope in the public change log.
Optional details for users who want a deeper explanation
The independent market inputs are daily open, high, low, close, and volume observations. NeuralTrend derives technical and other indicators from those inputs. Each currently supported asset uses its own model.
During model development, observations are ordered by time and divided into training, validation, and testing partitions. The current process uses a combination of fixed chronological holdout and continuous walk-forward testing.
NeuralTrend does not publish the complete feature list, model architecture, optimization search space, label construction, retraining thresholds, or source code. Those elements form part of the proprietary product methodology.
Material methodology changes, historical-signal revisions, and data corrections are recorded here. Routine appending of a new daily signal does not require an entry.
NeuralTrend added a public methodology page, clarified the difference between historical simulations and a future immutable live track record, and introduced a public change-log policy. This website release did not rewrite existing market data signal history.