Has anyone encountered field misalignment in WorkBuddy when summarizing contract attachments and search materials?

The issues I'm encountering with WorkBuddy are concentrated on merging materials from different sources into a single ledger. The field boundaries aren't locked down, causing columns like contract number, counterparty entity, signing date, dispute clauses, and source attachments to shift into each other's places. The contract attachments and search materials themselves can be read correctly.

I have a batch of supplier framework agreement attachments, similar case search materials, and some reimbursement screenshots. I've been using WorkBuddy for document comparison and semantic comparison for about a month; identifying individual documents is generally okay. These past two days, I tried the contract summary feature, wanting to aggregate contract attachments and search materials into one ledger. I only set up fields for: contract number, counterparty entity, signing date, dispute clauses, and source attachment. After merging, things often go wrong: the contract number ends up in the counterparty column, the signing date becomes an attachment filename, the source attachment is unstable, and there are quite a few duplicate rows. At one point, a prompt popped up saying something like "field conflict, unable to parse header."

I also tried creating a template first and having the contract summary fill it according to the template, but the template fields didn't match the auto-summarized content. The more I tweaked it, the more it felt like manually cleaning up after the model. Later, I tried separating semantic comparison and document comparison, extracting key clauses first, then merging, but field misalignment still occurred.

Now I'm unsure: should I not let WorkBuddy process raw contract attachments and search materials all at once? Instead, should I lock down source boundaries first—like generating intermediate tables for each type of material separately, then merging them via field mapping? Or do the entry rules themselves need to be set to "source priority" rather than merging directly by field name? Has anyone encountered similar situations?