Continuous due diligence programmes break in three ways that no re-scheduling adjustment can fix.
A client is cleared at onboarding. Three months later, their UBO appears in a sanctions update. The next periodic re-run is still four weeks away. The gap is not an edge case — it is a structural property of any batch process, and regulators treat it as a programme failure, not a timing issue.
A periodic re-run produces a different result. Was it because the client changed? Because a list updated? Because the algorithm was retuned? Without an evidence delta per evaluation, you cannot answer the question — and you cannot explain to a regulator why a previous clearance was correct.
A client is onboarded as "Zhang Wei" in your CRM. A new OFAC SDN entry names the same person as "張偉" in Traditional Chinese. A string-match pipeline returns zero overlap. The link is invisible until a manual review surfaces it — weeks after the designation went live.
A standing query registered at onboarding re-evaluates automatically whenever new evidence ingested for the client warrants it — a watchlist mention, an adverse-media passage, a corporate-registry filing. There is no batch window — and no gap.
Each re-evaluation returns the delta between the current evidence state and the prior one — which attributes changed, which list entries were added or amended, and what merge_confidence was before and after. A programme manager can see exactly what changed and why a result shifted.
Every clearance and escalation records the graph version current at decision time. Open a session against that version and re-run the query: you get the same result against the same evidence basis, replayed from the append-only audit log rather than reconstructed. Replay holds for as long as the version is retained — long enough to satisfy regulatory lookbacks.
A client onboarded in Latin script is matched against watchlist entries in Chinese, Arabic, or Cyrillic at query time — not translated at index time. Name variants, nickname conventions, and transliteration differences are handled by the matching engine, not by manual review queues.
A standing query registered at onboarding fires automatically whenever newly ingested evidence — a watchlist update, an adverse mention — warrants re-evaluation. The alert carries the evidence delta — what changed, and what merge_confidence was before and after — not just a new verdict.
The original onboarding clearance is independently reproducible within the retention window. When a regulator asks you to prove what you knew at Day 0, you do not reconstruct — you replay, against the same pinned evidence graph.
"A regulator asking you to replay a clearance gets an exact answer — not a reconstruction."
Entity "Meridian Holdings BV" and UBO "陳偉 (Chen Wei)" resolved and enrolled. Name Intelligence query run against OFAC SDN, EU consolidated, UK, and UN lists — all clear. Standing query #KYC-0441 registered. Session #S-0441 pinned to graph v_20260115_0900.
Operator's sanctions feed ingests 14 new designations. Standing query #KYC-0441 re-evaluates the client's entity against the updated list. No change detected — coherence thresholds not met for any new entry. Evidence delta recorded. Session #S-1829 pinned.
A regulatory filing naming the UBO is ingested via the operator's adverse-media pipeline. UBO "陳偉" resolved to entity #ent-4412 with merge_confidence 0.94 and merge_coherence "likely". A new mention attributing "subject of investigation" resolves to the entity. Standing query #KYC-0441 re-evaluates — threshold met. Alert #KYC-ALT-0019 fired with full evidence payload.
Compliance officer reviews the alert payload — source document, passage excerpt, and resolution scores. Decision: ESCALATE TO ENHANCED DUE DILIGENCE. The original onboarding clearance (session #S-0441) remains independently reproducible — same result against the same evidence state at Day 0.
Cross-script parse–retrieve–rerank. Matches client names against watchlists in the script they arrived in — no index-time transliteration.
standalone librarySession pinning against a versioned evidence record. Onboarding clearances remain independently reproducible regardless of subsequent list updates or re-evaluations.
Append-only record of every evaluation event. Evidence delta per re-evaluation — which attributes changed, which list entries fired, before-and-after scores.
Continuous reshaping as UBO structures and corporate hierarchies evolve. No destructive merges — every split and merge recomposable from the evidence record.
A demo runs against a sample client portfolio, registers standing queries, and walks through a live re-evaluation cycle — including the evidence delta your compliance team would receive and the session record your regulator can replay.