This issue has been created
There is 1 update.
 
 
LLM AI Integration / cid:jira-generated-image-avatar-0bd4f6f9-621a-46a2-bf8a-228fa3f1a00b LLMAI-192 Open

A nonsense query returns dozens of low-score hits with nothing to tell the agent it matched noise

 
View issue   ยท   Add comment
 

Issue created

 
cid:jira-generated-image-avatar-2718184b-ba2b-4714-bb17-38e929862308 Paul Pantiru created this issue on 30/Sep/26 17:21
 
Summary: A nonsense query returns dozens of low-score hits with nothing to tell the agent it matched noise
Issue Type: cid:jira-generated-image-avatar-916aaadd-7338-4544-9911-8afd9e1e5bde Bug
Affects Versions: 0.10
Assignee: Unassigned
Created: 30/Sep/26 17:21
Priority: cid:jira-generated-image-static-minor-aebd6c55-c88d-4531-bdb0-60db6fb2457e Minor
Reporter: Paul Pantiru
Description:

query="zxqv-nothing-matches-this" returned "about 57" documents because the hyphenated string is tokenized and OR-ed; the only signal is the score band (4.9 and below, against ~24 for a real title hit).

 

Expected / proposal: One of: (a) bind a minimum-should-match (mm) so that a majority of terms must match; (b) note the best score relative to a "strong match" threshold in the footer ("top score 4.9; scores under ~8 usually mean weak matches"); (c) add a sentence to the man page NOTES teaching the agent to compare scores across calls.

Actual: A confident list of 57 results with a footer identical to a real search.

 
 

1 update

 
cid:jira-generated-image-avatar-2718184b-ba2b-4714-bb17-38e929862308 Changes by Paul Pantiru on 30/Sep/26 17:22
 
Issue Type: Bug Improvement