An analyst approved a case last month. Today, a source changes. Which part of the decision needs another look? A useful audit trail should make that question easier to answer without asking the team to rebuild the whole investigation.
The opportunity is to retain the connection between a finding, the evidence that supported it and the version of the assessment that used it. A reviewer can then distinguish an old decision that was supported at the time from a current conclusion that needs new work.
8DB, our omnimodal database, provides structures for connected evidence and conclusions. Our bounded ownership research has exercised one important part of that idea: identifying which conclusion loses support when a source is withdrawn, while preserving a broader conclusion supported elsewhere. Turning that into a customer's complete audit workflow requires a separate evaluation.
Three questions that should not share one answer
Historical inspection asks what information and rules were recorded when a decision was made. Deterministic replay asks whether a defined procedure, run again with its exact inputs and environment, produces the same result. Current reassessment asks what the available evidence and applicable rules support now.
These can produce different answers without contradicting one another. A faithfully reproduced old decision may no longer be appropriate. A changed result may reflect a legitimate source correction. A stored digest may confirm that bytes are unchanged without establishing that the source was accurate or the decision lawful.
Well-designed logs and event-sourced systems can support these jobs. The design question is whether the record captures the inputs, versions and dependencies needed for the review, and whether the relevant procedure can actually be rerun.
Decision history · Three distinct investigations
Same case. Different questions.
Keep the old assessment available while investigating what changed.
What supported the decision then?
Read the recorded sources, rule versions and reviewer action.
Output: an evidence historyDoes the same procedure agree?
Rerun the defined computation with its exact inputs and required environment.
Output: a reproducibility checkWhat remains supported now?
Apply the current evidence and rules, keeping alternative support visible.
Output: a successor assessmentPreserve the reasons with the result
For a repeatable review, define a decision record that includes:
| What to retain | What it helps the reviewer answer |
|---|---|
| Source identities, versions and coverage | What evidence was available, and what could it establish? |
| Rules and effective period | Which criteria applied to this assessment? |
| Procedure version and required environment | Which computation can be reproduced? |
| Intermediate findings and unresolved questions | Why was this conclusion reached, and what remained unknown? |
| Reviewer action and successor assessments | Who decided, and what changed afterwards? |
Retaining this material is a design requirement, not proof that every current integration captures it. A replay test should deliberately change a source or rule version and refuse to label the new execution as an exact reproduction of the old one.
OFAC's compliance framework identifies risk assessment, internal controls, testing and auditing among the elements of a sanctions compliance program. The record design here is our engineering proposal for supporting review. The framework does not make this particular database design a regulatory requirement or certify a product that uses it.
A useful part we have already exercised
Our OpenSanctions evidence-update experiment distinguishes a dated ownership transition from a broader ownership link. When a dated source was withdrawn, the detailed chronology lost support while another source still supported the broader link. Removing that second support changed the link's status too.
That is more useful to an analyst than a single valid-or-invalid flag for the entire case. It identifies the specific conclusion to reconsider and keeps alternative support visible. The recorded experiment compared incremental updates with rebuilding the tested evidence set; it did not establish real-world ownership, legal clearance or production-wide replay of every decision.
The potential operational benefit is less repeated investigation. A customer pilot must measure that benefit against the team's existing workflow, including missed affected conclusions and unnecessary rechecks. Agreement with a software reference does not replace the analyst's adjudication.
Keep disclosure and judgment explicit
A review package may contain information that every recipient is not entitled to see. Define the authorized view, the evidence it includes and what that view cannot establish. An integrity proof can authenticate a disclosed commitment without deciding who has a lawful entitlement to the underlying material.
The same care applies to retention and erasure. Their treatment depends on the data, dependencies and applicable obligations. This article makes no claim that an immutable record or a deletion certificate automatically resolves those obligations.
Our current Sovclear counterparty-review pilot is a practical setting for testing better ownership and sanctions case evidence. It supports analyst decisions and can begin alongside existing sources and review tools. Automatic clearance, live policy enforcement and settlement integration require separate qualification. We are not presenting that pilot as a regulated clearing service or a guarantee of compliance.
Turn the next correction into a smaller investigation
Use this three-question check on one existing case:
- Which exact conclusion depended on the changed source?
- Which other evidence still supports it, or a less specific conclusion?
- What is the smallest next check that could settle the remaining question?
That is the practical value of connected evidence: direct the analyst toward the work that changed, rather than automatically restarting the case. The ownership experiment supplies a bounded example; a pilot establishes how much effort the same pattern saves in your workflow.
Test the difficult case your team already has
Choose a historical assessment that was hard to explain or had to be revisited after an evidence change. Agree which sources and versions can be used, identify the decisions to compare, and keep sensitive material out of the initial enquiry.
Then ask whether the review makes the affected conclusion, surviving support and next research step clearer. Measure analyst effort and material findings against the existing process. If deterministic replay is part of the requirement, test exact inputs, procedure versions and output agreement as a separate acceptance condition.
Bring us the case that keeps coming back. Start with a non-confidential description of the review problem. We can agree a focused pilot before requesting case data or proposing integration.
Lead illustration: conceptual artwork, not a product screenshot or a deployed customer system.
Evidence scope Design guide · Bounded evidence-update research · Analyst-review pilot. Explore the evidence.