Transaction monitoring — complete lifecycle¶
Customer- and partner-facing view of Fraud / transaction monitoring from ingestion through resolution — with repository boundaries and unverified stages marked explicitly.
Evidence boundary
Stages marked Coded (Fraud AI) are implemented in MasteryHive Fraud AI products. Stages marked UNVERIFIED / external depend on the banking application backend or dashboard, which is not in the documentation source tree for this site.
Lifecycle map¶
flowchart TB
ingest[1_Ingest]
baseline[2_Baseline]
score[3_Score_ML_Rule_or_CaseRules]
report[4_Report_compose]
handoff[5_Dashboard_report]
post[6_Post_action]
case[7_Durable_case]
assign[8_Assignment]
approve[9_Approval]
resolve[10_Resolution]
ingest --> baseline --> score --> report --> handoff
handoff -.-> post
post -.-> case --> assign --> approve --> resolve| # | Stage | What happens | Completion signal | Boundary |
|---|---|---|---|---|
| 1 | Ingestion | Accept a batch or event; schedule baseline work | Request accepted; work scheduled — not “all checks finished” | Coded (Fraud AI) |
| 2 | Baseline / profiling | Temporal, location/device, transaction behavioural checks | Profile updated or cold-start skip | Coded (Fraud AI) |
| 3 | Rule / ML evaluation | Authenticated scoring returns a prediction | Prediction returned to caller | Coded (Fraud AI) |
| 3b | Case-rule evaluation | Synchronous case-oriented rules | HTTP prediction | Coded (Fraud AI) — see Case Management Part A |
| 4 | Report | Validate contract, apply policy, attach narrative | Report ready to deliver | Coded (Fraud AI) |
| 5 | Operational handoff | Deliver structured report to banking application | Report accepted by create-report endpoint | Coded (Fraud AI → app) |
| 6 | Post-action | Work after scoring/report (triggers may be published) | — | PARTIAL — publish exists; consumer UNVERIFIED |
| 7 | Durable case | Case record in ops workspace | — | UNVERIFIED in source tree |
| 8 | Assignment | Owner / queue | — | UNVERIFIED |
| 9 | Approval | Maker-checker / approve-reject | — | UNVERIFIED |
| 10 | Resolution / closure | Close, reopen, SAR filing | — | UNVERIFIED (filing also not in code — AML reporting) |
Partial failure semantics (partner-visible)¶
| Situation | Expectation |
|---|---|
| Cold start | Baselines may skip profile persistence; prefer rule scoring for those customers |
| Mixed ML+rule identity keys on report compose | Rejected — see fraud output contract |
| Score/tier conflict | Rejected; do not retry the same invalid body expecting success |
| Narrative model failure | Deterministic summary fallback; decision still from policy code |
| Dashboard handoff failure | Compose surfaces failure so delivery can be retried per environment |
| Case-rule handler errors | Client-visible HTTP error for that request |
| Missing durable case after report | Not a Fraud AI defect by itself — case desk is external / unverified |
Reliability expectations¶
A scored transaction may not always receive a dashboard report
Scoring can succeed while report delivery is delayed, unsuccessful, or rejected by contract validation. ML scoring and report compose currently use different tier labels and score boundaries — some predictions never become reports until owners align policy. See Fraud output contract — Risk tier mapping.
Guidance:
- Do not treat a missing report as “low risk.” Prefer the prediction until a report arrives.
- Resubmitting a scoring call after fixes produces a new prediction and a new report attempt.
- Confirm retry behaviour for your environment with MasteryHive.
- Reconcile scoring responses against received reports by transaction id.
Decision policy¶
Primary decision + escalations on the dashboard payload are produced by fraud-decision-v1 (provisional / non-production). Escalation labels are not regulatory filings.
Separating “case” language¶
| Term | Meaning here |
|---|---|
| Case-rule evaluation | Stateless prediction API — Case Management Part A |
| Durable case | Investigator workspace — UNVERIFIED in this tree — Part B |
| Report | Investigator-facing payload with evidence and narrative |