One line of code. The best-supported artifact for any named task.
from reason_py import ReasonClient
client = ReasonClient()
artifact = client.resolve("reason://finance/fraud/anomaly-detection")
for tx in my_transactions:
score = compare(tx.features, artifact.pattern)
if score > artifact.thresholds.high_confidence:
flag_for_review(tx)
Live Walkthrough: Fraud Detection Across Silos
A regional bank with no existing fraud model resolves a reason:// artifact and instantly gains high-accuracy detection on its own transaction stream — without receiving any raw records.
[1] Resolving artifact from live node...
Score: 0.888677
Agent: xport-seed-agent-0
Evidence cases: 92,000
[2] Applying artifact to transaction stream...
(All transactions stay local — only the pattern was received)
TXN ID SIMILARITY TIER GROUND TRUTH
TXN-8821 0.995 high_confidence fraud ← FLAGGED
TXN-4412 0.720 moderate_confidence legitimate
TXN-9034 0.993 high_confidence fraud ← FLAGGED
TXN-2201 0.895 high_confidence legitimate
TXN-5577 0.990 high_confidence fraud ← FLAGGED
TXN-3309 0.673 moderate_confidence legitimate
[3] Summary:
Transactions evaluated: 6
Flagged for review: 6
Correctly flagged: 3 of 3 frauds
The bank now has fraud detection capability backed by 92,000 labeled cases from an institution it has never interacted with.
No transaction records were received from the depositing institution.
One API call. No data crossed any boundary.
Named reasoning capabilities
Every reason:// URI addresses a specific, bounded reasoning capability:
reason://medicine/records/longitudinal-maintenance-prediction
reason://finance/fraud/anomaly-detection
reason://cybersecurity/network/port-scan-classification