Skip to main content
When evidence is uploaded to a request item on one obligation, the system searches for semantically similar request items on other active obligations in the same organization. If it finds a match (82%+ cosine similarity on the item embeddings), it creates an evidence link — a lightweight pointer that surfaces the existing evidence for human review.

​
How it works

Evidence linking operates in two directions:
DirectionTriggerWhat happens
Forward propagationEvidence is uploaded to a request itemSystem searches other active obligations for similar items and creates links
BackfillA new obligation is parsedSystem searches all existing evidence from other obligations for matches against the new items
Both directions use pgvector cosine similarity on the request item embeddings. The 82% threshold ensures only strong semantic matches surface — reducing noise while catching genuine overlaps.
Evidence linking works across regulation types. Evidence from a BSA/AML examination can link to a third-party oversight consent order if the underlying request items are semantically similar.
On any request item that has linked evidence from another obligation, you’ll see an amber-bordered banner above the manual evidence section. Each link card shows:
  • The evidence file name
  • The source obligation and request item (e.g., “EX-2026-001 / RI-005”)
  • The match confidence percentage
  • Confirm and Reject buttons
Confirmed links turn to a stone-colored card with a “Linked from EX-2026-001” badge. Rejected links disappear. The evidence count in the item header reflects both manual and confirmed linked evidence (e.g., “3 evidence (1 linked)”).

​
”Also linked to” indicators

Below each evidence attachment on the source side, you’ll see where that evidence has been linked: 
Also linked to:
· Consent Order (CO-2024-001) · CO-007 — Confirmed
· Third-Party Review (TP-2026-001) · TP-003 — Pending
Click any entry to navigate to the target obligation’s request item. Obligations with pending evidence links (auto-applied, not yet reviewed) show an amber dot next to their ID in the sidebar. This helps your team prioritize review — the badge disappears once all links are confirmed or rejected. 
auto_applied → confirmed
auto_applied → rejected
Links are created automatically when the system detects a match. Your team reviews each link and either confirms (evidence is relevant) or rejects (false positive). There is no manual link creation — the system handles discovery, humans handle validation.

​
Limits

  • Maximum 5 cross-links per evidence upload
  • Only active (non-archived) obligations are considered for matching
  • Links are scoped to the organization — no cross-tenant matching