Sanctions Desk gathers and checks the evidence for every sanctions-screening alert and recommends an outcome; a person with the right entitlement approves it. One alert, start to finish · The problem it works on

Replay of recorded runs. Nothing runs live here. Every value below is read from files committed in the repository. To run the desk yourself, follow the README quick start.

What happened: The decision step proposed a close 3 times; the critic sent it back 3 times; the policy guard raised it to ESCALATE_L2 (rule: critic_unresolved). Outcome: Escalate to level 2 (ESCALATE_L2), as the label expects.

Outcomes: CLOSE_FALSE_POSITIVE = close as a false positive · ESCALATE_L2 = escalate to level 2 · NEED_MORE_INFO = ask for more information · RECOMMEND_BLOCK = recommend a block (two people must confirm). Raise-only: the policy guard can make an outcome more cautious, never less.

An ownership chain: the decision step proposed a close three times, the critic sent it back each time, and the guard escalated it because the critic's objection was still open.

Outcome

Desk recommendation

Escalate to level 2 (ESCALATE_L2)

Label in the case file

Escalate to level 2 (ESCALATE_L2) (eval/cases/exam.json, case EX-043)

Compared with the label

matches the label

Code's first finding

Close as a false positive (CLOSE_FALSE_POSITIVE)

Critic

3 round(s), final: escalated

Model calls

7 (2 local, 5 cloud); 7 with a signed receipt, all verified

Run

routed-20261023T133311Z, policy version 5, 46 steps in 152.8 s

Case note from the case file: The holding company is 60 percent owned by the listed party, so under the 50 percent rule it is blocked and its 70 percent counts in full; effective ownership is 42 percent. Relabelled from CLOSE_FALSE_POSITIVE on 24 Oct 2026 (see eval/cases/README.md).

When this run was scored, the case file's label was Close as a false positive (CLOSE_FALSE_POSITIVE). The case was relabelled afterwards; the change and its reason are in eval/cases/README.md. The comparison above uses today's label.

Why the close was stopped

The decision step proposed a close 3 times; the run ended at Escalate to level 2 (ESCALATE_L2). What each check said, in order:

• critic (step 31): rejected: Rationale states 'Per the effective ownership definition, below threshold means close as false positive' but effective_ownership_threshold definition does not mention false positive; false_positive definition requires name similarity plus DOB conflict, nationality mismatch, and no document match.; c
• critic (step 37): rejected: cited fact does not hold: unknown table "ownership". Cite only rows of these tables: account (key account_id), alert (key alert_id), beneficial_owner (key ownership_id), business_definition (key definition_key), customer (key customer_id), customer_alias (key alias_id), customer_identifier (key iden
• critic (step 43): rejected: definitions_met.false_positive is false (comparison block); false positive definition requires name similarity, DOB conflict, nationality mismatch, and no document match; evidence shows customer name Juniper Software GmbH vs listed party Open Joint Stock Company Russian Electronics, no DOB, no docum
• policy guard (raise-only rules in code): CLOSE_FALSE_POSITIVE raised to ESCALATE_L2. Checks that raised or flagged: critic_unresolved: block

Steps in order

Each line is one event from the run's trace (trace.jsonl.gz). The model is the one that served the call, as the signed receipt records it.

Model calls and receipts

Each signature was checked when this page was built, with the same code as npm run evidence:verify (Ed25519 over the receipt body in receipts.jsonl.gz); the build stops if one does not verify, is missing, or names a different model from the trace. The Nightshift runtime's receipt key is pinned in config/default.json, so signature valid, key pinned shows the receipt was not changed and that this runtime signed it; a receipt signed by any other key stops the build.

The whole run passes the same check: 260 receipts with valid signatures, 260 model calls in the trace that name a receipt, none missing and none naming a different model from its receipt.

Signing key fingerprint (first 16 hex of SHA-256): 1835d1eab9211f97.

Zetaris queries

This run's export keeps the table, the row count and the Zetaris query id of each read, not the statement text; the statements are not reconstructed here.

Check it yourself

Clone the repository and run npm run evidence:verify -- eval/runs/routed-20261023T133311Z. What the export left out: tokens, local paths, hostnames and addresses other than the local machine, usernames, e-mail addresses and secret environment values were removed from results.json and trace.jsonl.gz; receipts.jsonl.gz is untouched.