XTester Alpha 7: research, portfolios, and Paper recovery

XTester 1.0.0-alpha.7 was released on October 20. It improves chat-based research, portfolio strategy management, statistics calculations, and Paper robot recovery. The update helps you return to a task with its results preserved: open the log of a completed run, see how a research specification relates to the code, or continue simulated trading after restarting Runner.

In our September overview, we discussed separate AI Lab conversations and plans for portfolio management. New builds have appeared since then. Below, we focus specifically on the changes in Alpha 7 compared with Alpha 6: what you can now do and what to check when testing your own strategy.

Refine a research specification in conversation

AI Laboratory now supports configuring research through chat. You can specify periods and calculation steps in the conversation, and the research graph updates automatically. The accepted specification is saved together with the strategy and a digital fingerprint of its code.

This is useful when an idea changes during discussion. For example, you may initially want to test an entry filter on one part of the historical data, then choose another period and refine the calculation conditions. The saved specification lets you return to the agreed research setup and its associated code version. When examining the result, you have a concrete reference: what you agreed to investigate and which source code the task concerned.

You can now start improving a strategy directly from the portfolio report. If the overall result needs investigation, you can move from the report to work on the strategy. You do not have to postpone the observation until another separate research session.

A saved specification does not, by itself, establish that an algorithm is robust. The selected variant still needs testing on data that was not used to tune it.

Bring strategies and results together in a working portfolio

In Alpha 7, you can create a strategy by name, select several strategies for bulk actions, and add them to or remove them from a portfolio. These everyday operations become important once you have more than one variant: you need to change the research set and return to individual strategies' results.

Completed runs and full strategy logs remain available after restarting the application. If you stopped at a suspicious trade yesterday, you can reopen the log today and continue investigating. The release also adds code-version-aware cleanup of outdated backtests.

Keeping logs matters when checking conclusions. Summary statistics show the overall picture, while the log helps reconstruct the sequence of events: what happened before entry, how the strategy handled the position, and where to look for a discrepancy. Saving a result does not mean that every old run will be kept indefinitely; account for outdated-backtest cleanup when maintaining your own archive of experiments.

More calculation conditions are explicit

For simulation, you can choose bar-based or tick-based stepping. A backtest can start from a restart snapshot; per-trading-pair restart information is also used. These capabilities help when revisiting a prepared run and understanding the state from which it continues.

Worst-day statistics and the Sharpe and Sortino ratios now use UTC calendar days. Unavailable daily metrics are explicitly marked for short runs. If there is not enough data for a daily metric, the report shows that limitation.

When comparing results before and after the update, examine daily statistics separately. A metric may change because its calculation rules changed. That alone does not mean the trading idea has become better or worse. For a meaningful comparison, keep the code, period, data, and execution conditions the same.

Execution diagnostics are also more specific: XTester reports when risk rules reject a position size or reduce the calculated quantity. Duplicate open client order IDs are rejected. This helps investigate cases where a signal occurred but the expected order did not appear in the form the strategy author intended.

A Paper robot retains its state between runs

Launcher, XTester's robot-launching component, now has an order book next to the robot chart and an option to keep the computer awake while robots are running. Its behavior on closing, stopping, and restarting has been refined. Paper supports optional position closing with an estimate shown beforehand.

Paper state and accumulated results survive a Runner restart; Runner is the process that executes the robot. After the data stream is restored, the system first catches up on missing data and only then resumes execution. Separate safeguards cover replay during warm-up. Paper fills are retained until the strategy acknowledges processing them.

For users, this is a reason to check a familiar scenario: open a simulated robot's state, restart Runner, and compare its positions and accumulated result. Paper remains simulated trading. Preserving and restoring its state does not establish how a real exchange account will behave or guarantee future returns.

For external AI agents using MCP, application launch, attachment to a project and window, and completion of long-running tasks have been improved. After reconnecting, a client can regain access to its own tasks. Task status shows the owner, and remembered agent connections can be reviewed and revoked. If the connection drops during a run, check the existing task's state before starting another one.

Before updating

Alpha 7 is available on the download page. Packages have been released for Windows x64, Linux x64, and Macs with Apple Silicon or Intel processors. This release lists macOS 14 or later for both Mac architectures.

This is an alpha release with signing limitations: the Windows installer does not yet have a trusted publisher signature, and the macOS applications are ad-hoc signed, without Developer ID signing or Apple notarization. The release description explicitly notes that native Mac launch and Gatekeeper acceptance were not tested. An available archive for a platform must be distinguished from a verified launch on that platform.

Before upgrading, back up your working project. After updating, repeat a familiar run and compare trades, daily metrics, and the log. For AI Lab research, check the saved specification and its link to the code; for a Paper robot, check state recovery separately. The full changelog and verification limits are published with the release.

Historical results and Paper trading results do not guarantee future returns.