MAM: Fraud: Journal Entries Testing
ISA: ISA 240
Module: operations.audit.jedi
Mission
This programme is an ABI module with its own agent and Nexus app. Production tooling for each sub-solution is not built yet. The list below is the mission contract from the digested catalog.
Sub-solutions (4)
JEDI.RuleScoreRule-Based Journal Entry Scoring (4 tasks) · Deterministic Rule Engine (non-AI)JEDI.StatScoreStatistical Anomaly Detection (3 tasks) · Unsupervised Statistical ModelsJEDI.ClusterScoreUnsupervised Clustering & Outlier Detection (8 tasks) · ML (unsupervised): k-means, DBSCAN, Isolation ForestJEDI.InsightFindings & Follow-up Assistant (3 tasks) · Agentic LLM (generative)
Task-skill pairs (18)
| ID | Sub-solution | Task | Skill |
|---|---|---|---|
| JR-01 | JEDI.RuleScore | Extract and structure the full GL journal entry population from client ERP/accounting system | GLExtractionSkill |
| JR-02 | JEDI.RuleScore | Apply ISA 240 non-standard entry criteria: outside business hours, round amounts, unusual account combinations, entries by unusual users, reversing entries, entries near period-end | NonStandardEntryDetectorSkill |
| JR-03 | JEDI.RuleScore | Flag entries posting to sensitive account combinations (e.g. revenue-to-liability, expense-to-equity, intercompany with unusual counterparties) | SensitiveAccountComboSkill |
| JR-04 | JEDI.RuleScore | Score each entry with a composite rule-based risk score and produce a ranked population for auditor selection | RuleScoreAggregatorSkill |
| JS-01 | JEDI.StatScore | Apply Benford's Law analysis to first and second digit distributions of journal entry amounts | BenfordAnalysisSkill |
| JS-02 | JEDI.StatScore | Detect distributional outliers in amounts, frequencies and timing using Z-score and IQR methods | DistributionalOutlierSkill |
| JS-03 | JEDI.StatScore | Analyse posting frequency patterns by user and account to detect unusual activity spikes | FrequencyPatternSkill |
| JC-01 | JEDI.ClusterScore | Feature engineering: build multi-dimensional entry feature vectors (amount, timing, user, account type, description embedding, system/manual, reversal flag) | FeatureEngineeringSkill |
| JC-02 | JEDI.ClusterScore | Run k-means clustering to identify natural posting behaviour clusters and find entries far from cluster centroids | KMeansClusteringSkill |
| JC-03 | JEDI.ClusterScore | Run DBSCAN to identify low-density entry groups that do not fit any natural cluster | DBSCANSkill |
| JC-04 | JEDI.ClusterScore | Run Isolation Forest to score each entry by how easily it can be isolated from the population | IsolationForestSkill |
| JC-05 | JEDI.ClusterScore | Apply ensemble voting: only flag entries anomalous in at least 2 of the 3 clustering methods to reduce false positives | EnsembleVotingSkill |
| JC-06 | JEDI.ClusterScore | Contextual enrichment: cross-reference flagged entries with authorisation logs and ERP workflow data to filter explainable outliers | ContextualEnrichmentSkill |
| JC-07 | JEDI.ClusterScore | Apply relative rarity scoring: compute outlier score relative to entity's own baseline, not external norms | RelativeRarityScorer |
| JC-08 | JEDI.ClusterScore | Human-in-the-loop calibration: feed auditor accept/reject decisions back to recalibrate cluster boundaries for subsequent years | SemiSupervisedCalibrationSkill |
| JI-01 | JEDI.Insight | Translate cluster/statistical anomaly outputs into plain-language audit findings | AnomalyNarratorSkill |
| JI-02 | JEDI.Insight | Populate the journal entry testing working paper template with flagged items, scores and descriptions | WPDrafterSkill |
| JI-03 | JEDI.Insight | Recommend next audit steps based on identified anomaly patterns (e.g. cluster of year-end revenue entries → consider additional revenue cut-off procedures) | NBASuggesterSkill |