Maven Release 2025.4

Maven 25.4 General Release Notes 

Land-Mix Model Weight Testing 

A new testing feature has been added for running Model Evaluations in the background.  This will test differing Weights for Comparison Values to determine which settings provide the best results. This test is run as a Scheduled Job, which can be run on demand or a daily or weekly schedule.  

Configuring for the Weight Test is done on the Setup -> Valuation Model Testing -> Weight Test Settings page. 

 From here, the specific Land Mix Model to test when running the Job can be set. A Stop Time to end the test can also be set, which will stop execution of the Evaluations at the given time. 

 

Kicking off a Weight Test is done on the Admin -> Scheduled Jobs page. Clicking start will begin the test using the configured model. 

This Job has the potential to be long-running depending on how many Comparison Values a Model has. The Weight Test runs Evaluations, continuously adjusting the Comparison Value Weights up and down until an iteration does not improve upon a previous iteration, or until a Weight of 0 or 100 is hit. While the test is running, it can be halted by clicking the Stop button, which appears in the Scheduled Jobs table. It will also stop if the configured Stop Time is hit. 

 The next time the Job starts, either manually or on a schedule, it will pick up and resume a stopped Test rather than starting a new one, provided the Model has not been changed since. 

 

Hidden Text Fields 

Updated the Client Secret and API Key fields for the Camo Ag integration and Cost New integration to hide the contents by default. The content can still be made visible to the user by clicking on the button on the right of the field.  

LLM (AI) Comments Generation 

Added the ability for Maven to generate approach comments automatically during the valuation process using an Azure OpenAI Service instance. Currently, comments can only be generated for the approach’s Approach Comments. The user can edit generated comments. 

For comments to be generated, they must be enabled by an admin and on each model/scope where comment generation is desired. The top-level configuration is available via Admin -> API Configuration -> LLM Comments and requires the Azure OpenAI deployment endpoint and API key. 

Each model allows comments to be enabled and requires a template if comments are enabled. The template specified here will be used to generate the prompt that is sent to the LLM. There are several defaults provided that send relevant information for each model. 

Report scopes also allow for comment generation across all models that have a template set. When running a report scope, the Use LLM Comments value overrides that of all models in the scope. 

Once enabled, comments will be generated automatically when a valuation is run. They will be populated into the approach comments at the bottom of each approach tab. 

All generated comments will start with a warning message stating that the comment was generated. This is hardcoded and cannot be disabled, but can be removed by the user manually. 

 All comments are generated with a Top P and Temperature of 0.2 to reduce “creativeness” and randomness of the responses (currently hardcoded). 

  

Azure OpenAI Services have a quota associated with each deployment. Maven will retry comment generation after the quota resets if generation fails. It will, however, only retry 6 times to prevent prolonged valuation times.  

  

Maven will also check token counts when generating comments, and if a prompt contains more tokens than the deployment’s quota, it will not attempt to create comments. 

  

If comment generation fails, the valuation will continue without the comment, and the approach’s approach comments will appear blank. A warning message will also appear at the top of the page stating why the comment generation failed. 

  

For report scopes, comments will be generated for each model that has a template set (ignoring the individual model checkboxes).  If comment generation fails for any model, then a warning will appear stating which comments failed, and why. All successfully generated comments should still appear in their model’s approach comments field. 

  

When the prompt is generated, all data from the valuation will be available. The one difference from the report generation is that only the approach data for the approach being commented on is available during prompt generation. Also, some fields, such as attachments and images, will not have any effect on the prompt because it is passed to the LLM in a plain text format. 

LLM Template Editor 

Added a new template editor for LLM comments with almost identical functionality to the existing editor, except that custom templates are stored in the new LLM-Custom directory and default templates are read from the LLM-Default directory. 

 

LLM Template Tester 

Added a new template tester page for LLM templates. This allows the prompt to be generated using any template on a model from an existing valuation. The template and report index fields function identically to the report template tester. The model field is populated once a report index is set. It will contain all models used in the report. 

Prompts can be generated without configuring an Azure OpenAI Service in Maven; however, in order to create comments, a service must be configured (through Admin –>  API Configuration -> LLM Comments). 

  

There are two ways to trigger comment generation. Enabling Auto Generate Comments will automatically generate and display LLM comments when Show is clicked. Additionally, when a prompt is generated, a button will appear below it. The button triggers comment generation. 

The prompt and comment will appear side by side, with a splitter so that the user can change the sizes. Both the prompt and comment are read-only. 

When the prompt is generated, all data from the valuation will be available. The one difference from the report generation is that only the approach data for the approach being commented on is available during prompt generation. Also, some fields, such as attachments and images, will not have any effect on the prompt because it is passed to the LLM in a plain text format. 

 

Approach Comments 

Approach comments have been added for approaches in a valuation. The comments appear at the bottom of the page under each approach’s tab. 

Subject Details Update 

The Subject Details page has been reworked to now function in the same manner as the updated Valuation Details page. The layout and function of the page have not changed, but the way data is retrieved, updated, and saved has been completely redone. 

  

“Save With Warnings” should no longer appear when saving a Subject while editing. This will prevent issues with cloned Subjects not matching the current data shown on the form if the Subject had been Saved With Warnings prior.

· 

• Grantor & Grantee fields will truncate to 50 characters when importing Sales/Subjects to not throw an error if longer.
• Drop-down options for Custom Fields are now sorted alphabetically.
• “Label” type Custom Fields will now always display on their own row.
• Added “Client” to Appraisal Field Sets

• Download links for Appraisal Files should no longer appear if no data for the file exists in the database.