Submit

Run your model on the held-out set and upload its predictions. We score server-side against 1,020 labels and publish the result on the leaderboard.

Your submission

↑

Drop your predictions file here

or click to browse · max 50 MB

predictions.jsoninstance_id + decision

Model name

Organization

Protocol

Download the evaluation set

MSAB-Eval-v2.2-Hard — 1,020 payment events across 15 attack families. For each event, decide approve vs reject for its (amount, description, resource, pay_to, host) against the mandate.

↓ msab_eval_v2_2_hard_unlabeled.json

Run your model

import json

events = json.load(open("msab_eval_v2_2_hard_unlabeled.json"))
predictions = []
for event in events:
    p = event["event_snapshot"]["event_property"]
    decision = your_model.predict(p)   # "approve" or "reject"
    predictions.append({"instance_id": event["event_id"],
                        "decision": decision})

json.dump(predictions, open("predictions.json", "w"))

Upload your predictions

Upload predictions.json above. Anything other than "approve" or "reject" counts as wrong.