Aletheia V4 is scheduled for release next week. The final adjustments are being made mainly around the web search component and the system is entering its last preparation phase before the next public testing cycle.
This isn’t the full technical breakdown. That will come with the official release. The goal here is to give a clear overview of what changed, why those changes were necessary, and what traders should expect from V4.
What Previous Versions Taught Us
The earlier versions confirmed that the foundation works.
Aletheia is capable of:
-
Monitoring and analyzing market events
-
Building structured reasoning around them
-
Generating trade setups
-
Passing those setups to EAs for execution and management
In many cases, strong events produced solid reasoning, clean entries, and targets that were reached quickly.
At the same time, earlier iterations exposed weaknesses.
Some low-quality or noisy events were still entering the pipeline. When that happened, the reasoning layer sometimes had too much room to interpret incomplete context. That led to setups that, in hindsight, should not have been generated.
This reflects the reality of using AI in financial analysis.
It can process more information than any human and operate across multiple assets and news streams simultaneously. But it is still a probability-driven system. Without strong rules, high-quality inputs, and proper constraints, it can misclassify weak signals as meaningful ones.
V4 was built around a simple objective:
Increase consistency and reduce room for unreliable reasoning.
A Full Rework of the Agent Structure
V4 is not a minor update. Every agent in the system was rebuilt.
-
Prompts were rewritten
-
Tools were refined
-
Logic flows were redesigned
-
Decision constraints were tightened
Each agent went through repeated testing and controlled reruns. The aim was straightforward: if a decision is based on clear, structured reasoning, it should remain stable across executions.
To support this, dedicated backtesting scripts were developed for every agent. These scripts allow full traceability:
-
What data the agent received
-
What rules were applied
-
How reasoning developed step by step
-
What output was produced
This made iteration more grounded and significantly improved the development process. Instead of evaluating only outcomes, the entire decision path could be inspected and improved.
Removing Structural Weak Points
Some earlier gatekeeping steps turned out to be unnecessary or even counterproductive.
One example was analyzing events too early before enough surrounding datapoints were gathered. This forced the system to reason without proper context, increasing the likelihood of weak conclusions.
In V4, the event lifecycle was adjusted so that analysis happens only once sufficient information is available.
At the same time:
-
Reasoning gates were tightened
-
Event filtering became more selective
-
Hypothesis generation was expanded
-
The web search layer was upgraded to deliver higher-quality datapoints
The objective is not to process more information, but to process the right information.
Extracting More Value from Fewer, Higher-Quality Events
As the reasoning layer was tightened, fewer events qualified for deeper analysis. The focus shifted toward extracting maximum value from high-quality events rather than processing large volumes of mixed signals.
To support this, event coverage was expanded so the system can capture more of the possible ramifications around those selected events. This includes second-order effects and cross-asset implications.
At the same time, the web search stage was significantly strengthened to ensure that broader coverage does not introduce noisy datapoints or weak reasoning.
Better inputs, combined with stricter reasoning, lead to more reliable trade selection.
Introducing Risk Types
Testing showed that event impact is rarely binary.
A single development can:
-
Strongly affect one company or asset
-
Partially affect another
-
Have minimal impact elsewhere
Treating outcomes as simply “relevant” or “not relevant” was too limited.
V4 introduces structured risk types:
-
Low Risk – Higher confidence setups
-
High Risk – Lower confidence setups
This is not a quality label but a confidence classification.
Traders remain fully in control:
-
Choose which risk types to receive
-
Opt in or out entirely
-
Define risk allocation per type
-
Configure partial targets per type
The system provides structure, while execution preferences stay in the user’s hands.
New Trade Lifecycle Agent
User feedback highlighted a gap between automated management and manual trade handling. Trailing stops and partial targets were effective but did not always capture the adaptability of active trade management.
V4 introduces a dedicated lifecycle agent designed to operate on top of the existing management logic.
This agent:
-
Activates only after TP1 is hit
-
Focuses on reducing profit give-back
-
Adjusts management decisions during the trade’s later stages
The delayed activation is intentional. Early intervention increases the chance of closing positions prematurely. Waiting until the first target is secured allows the trade to develop before additional management logic is applied.
As with other features, this is optional. Traders can choose whether to use it.
Improvements to the Web Search Layer
The web search component plays a central role in data quality.
For V4, it was reworked to:
-
Retrieve more relevant and contextual sources
-
Improve data filtering
-
Strengthen alignment between events and analysis
This ensures that expanded event coverage does not reintroduce noise into the reasoning process.
Final refinements to this part of the system are currently being completed.
What Comes Next
Once the last modifications are finalized, Aletheia V4 will move into its next public testing phase.
Earlier versions demonstrated that the framework is capable of generating valid analysis and trade setups. V4 focuses on making those outcomes more consistent, structured, and reliable.
A broader technical overview will follow at release, including how the system operates, how agents interact, and how traders can configure its behavior. The internal reasoning pipeline itself will remain private.
The web application is also undergoing a rework and is expected to launch either alongside V4 or shortly after.
As always, real market conditions will be the true test. Feedback during the upcoming testing phase will play a key role in refining the system further.
Join the Next Testing Phase
Access to Aletheia testing has always been free, and V4 will follow the same approach.
Bringing more testers in is important at this stage. A wider group helps us observe how the system behaves across different environments and configurations, which is essential for refining it further.
In particular, broader participation helps us:
-
Validate execution across a wider range of brokers (and fine-tune the symbol matching logic)
-
Observe performance under different risk configurations
-
Compare behavior across accounts receiving different trade types
-
Evaluate setups using and not using the trade lifecycle agent
-
Test different partial closure structures depending on risk types
This kind of diversity is difficult to simulate internally and becomes far more valuable when tested in real trading conditions.
Anyone interested in participating is welcome to join the community and take part in the upcoming phase. The more feedback collected, the easier it is to identify weaknesses, refine behavior, and improve the system before wider deployment.
Community access: https://discord.gg/r5j7jNtfBF
