{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp09_archetype_breadth150_20260804","research_id":"eoa_inverse_innovation_exp09_light_prior_art_20260804","cell_id":"predictive_residual_processing__accounting_auditing","search_lanes":{"direct_problem_and_intervention":{"queries":["journal entry anomaly detection machine learning recurring journal entries audit review","predictive analytics journal entries expected values residual audit anomalies"],"source_ids":["SRC2","SRC4"],"no_result_note":null},"synonyms_and_historical_terms":{"queries":["continuous auditing journal entries concept drift fallback human review","journal entry anomaly detection autoencoder reconstruction error accounting primary research"],"source_ids":["SRC2","SRC3"],"no_result_note":null},"products_practices_and_standards":{"queries":["ISA 240 journal entries testing unusual entries recurring entries audit","site:pcaobus.org AS 2401 journal entries unusual selection characteristics","journal entry testing software anomaly detection product continuous auditing"],"source_ids":["SRC1","SRC4"],"no_result_note":null},"component_combination":{"queries":["audit anomaly detection explainable journal entry field level anomaly score reconstruction error","predictive journal entry reconstruction error threshold model drift continuous auditing","journal entry anomaly detection versioned model random audit sample fallback"],"source_ids":["SRC1","SRC2","SRC3","SRC4"],"no_result_note":"No retained source described the complete combination of versioned structured predictions, field-level residual packets, independently sampled non-alerted entries, synchronization checks, and automatic restoration of full review."}},"sources":[{"source_id":"SRC1","title":"AS 2401: Consideration of Fraud in a Financial Statement Audit","publisher":"Public Company Accounting Oversight Board","url":"https://pcaobus.org/oversight/standards/auditing-standards/details/AS2401","source_type":"OFFICIAL_STANDARD","claims_supported":["Auditors must select journal entries from the general ledger for testing and examine their support.","Unusual accounts, atypical preparers, period-end timing, sparse descriptions, and round numbers are recognized selection characteristics.","Recurring standard entries are generally subject to internal controls, while additional emphasis may be placed on nonstandard entries.","Computer-assisted techniques may identify entries for testing, but required audit judgment and evidence procedures remain authoritative."]},{"source_id":"SRC2","title":"Detection of Anomalies in Large-Scale Accounting Data using Deep Autoencoder Networks","publisher":"arXiv","url":"https://arxiv.org/abs/1709.05254","source_type":"PRIMARY_RESEARCH","claims_supported":["Large journal-entry populations contain a small fraction of anomalous attribute patterns.","Handcrafted fraud rules may fail to generalize and can be circumvented.","An autoencoder can reconstruct structured journal-entry attributes and use reconstruction error, attribute probabilities, and a threshold as an adaptive anomaly assessment.","The method was evaluated on two real-world SAP journal-entry datasets and substantially narrowed the flagged population in the reported experiments."]},{"source_id":"SRC3","title":"Continual Learning for Unsupervised Anomaly Detection in Continuous Auditing of Financial Accounting Data","publisher":"arXiv","url":"https://arxiv.org/abs/2112.13215","source_type":"PRIMARY_RESEARCH","claims_supported":["Journal-entry distributions are nonstationary because organizational processes are introduced, redesigned, or discontinued.","Stationary audit models can miss relevant distribution changes, while sequential updating can overwrite previously learned information.","A continual-learning anomaly framework was evaluated on designed audit scenarios and two real-world datasets, providing initial evidence of reduced false-positive and false-negative decisions."]},{"source_id":"SRC4","title":"AI-Powered Journal Entry Analysis Software: Journals Risk Analyser","publisher":"BlackLine","url":"https://www.blackline.com/products/financial-close/journals-risk-analyser/","source_type":"FIRST_PARTY_PRODUCT","claims_supported":["A commercial product already applies AI-driven anomaly detection to manual journal entries.","The product presents monitoring of an entire journal-entry population and irregularity-focused analysis as a way to improve visibility and reduce misstatement risk.","Population visualization and accelerated analysis of manual entries are established product capabilities adjacent to the proposal."]}],"problem_evidence":{"status":"PARTLY_SUPPORTED","finding":"The sources make the underlying problem visible: journal-entry populations can be large, consequential anomalies may occupy only a small fraction, fixed rules have known limitations, and posting distributions can change over time. Standards and commercial practice already focus attention on unusual entries. However, the retained evidence does not directly quantify reviewer time spent specifically on predictable recurring close entries or demonstrate that such workload delays material investigations in the proposed setting.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"closest_prior_art":[{"name":"Deep-autoencoder journal-entry anomaly assessment","source_ids":["SRC2"],"overlap":"Reconstructs multi-attribute journal entries, computes actual-versus-reconstructed error, adjusts the assessment using attribute probabilities, and flags entries crossing a threshold for examiner attention.","remaining_difference":"It reports an anomaly assessment rather than the proposal's governed review protocol: explicit human-readable signed field residual packets, frozen model-version handshakes, independent review of non-alerted raw entries, declared bypass classes, and automatic full-review fallback are not described."},{"name":"Continual anomaly detection for continuous financial auditing","source_ids":["SRC3"],"overlap":"Directly addresses nonstationary journal-entry data, model adaptation, retained historical knowledge, and false-positive and false-negative behavior.","remaining_difference":"Its continual updating differs from a frozen close-cycle model with controller-approved post-close replay, and it does not supply the proposed residual-packet reconstruction audit, synchronization rejection, bypass, or full-review fallback regime."},{"name":"BlackLine Journals Risk Analyser","source_ids":["SRC4"],"overlap":"Commercially establishes whole-population journal analysis, AI anomaly detection, and prioritization of irregular manual entries.","remaining_difference":"The public product description does not establish explicit structured predictions and signed field residuals, model-version checks, independent sampling of entries producing no alert, or automatic suspension of residual-only triage."},{"name":"PCAOB risk-based journal-entry selection practice","source_ids":["SRC1"],"overlap":"Uses entry attributes, workflow identities, timing, account relationships, and normal-versus-nonstandard status to focus journal-entry testing.","remaining_difference":"It is an authoritative risk-selection framework, not a predictive reconstruction channel with model-relative field residuals, uncertainty weighting, replay-buffer governance, or technical fallback controls."}],"prior_art_disposition":"ADJACENT_PRIOR_ART","contrastive_claim_remaining":"For a bounded recurring-entry population, a frozen and checksummed model can make an explicit structured prediction for every entry and communicate human-readable field residuals as the review message, while independently sampled non-alerted raw entries test reconstruction fidelity and declared synchronization, drift, missingness, bypass, or fidelity failures restore full review. The retained sources establish most neighboring elements but not this integrated, testable control architecture.","contrastive_claim_falsifier":"The contrast is falsified by a directly documented pre-existing system or practice implementing this integrated architecture, or empirically if a chronological shadow replay shows that residual packets cannot reconstruct decision-relevant entry state within the declared tolerance, miss a consequential deviation or bypass entry, fail synchronization or drift controls, or yield no net reduction in reviewer effort after sampling, reconciliation, and model-control costs.","gates":{"adequate_source_search":{"status":"PASS","rationale":"Four search lanes covered direct terminology, older terms such as CAATs and continuous auditing, an official audit standard, a first-party product, primary research, and combinations involving reconstruction error, drift, sampling, versioning, and fallback. Exactly four opened sources from three publishers were retained.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"supported_problem":{"status":"PASS","rationale":"The problem is partly supported: large journal populations, sparse anomalous patterns, rule limitations, and changing distributions are documented, although the specific close-review delay and time burden remain unquantified.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"distinct_testable_claim":{"status":"PASS","rationale":"Adjacent systems reconstruct or score journal entries and monitor anomalies, but the retained art does not disclose the full combination of field-level residual communication, frozen version synchronization, independent raw-channel sampling, bypasses, and declared restoration of full review. Each element has observable pass/fail behavior.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"bounded_next_test":{"status":"PASS","rationale":"A chronological, read-only replay of two recurring-entry classes from one completed close is bounded. Comparison against the already required full review can measure queue size, reviewer minutes, reconstruction disagreements, documented-findings recall, bypass coverage, version mismatches, and total control effort without changing audit conclusions.","source_ids":["SRC1","SRC2","SRC3"]},"no_obvious_safety_or_authority_stop":{"status":"PASS","rationale":"The proposed first step is read-only, preserves the ledger and existing reviews, and leaves audit scope and evidence decisions with authorized professionals. This is compatible with the standard's requirement that auditors select entries and examine support. Production suppression or replacement of required procedures would require separate authorization and validation.","source_ids":["SRC1"]}},"screen_survival":true,"world_novelty_boundary":"This bounded four-source screen supports only an adjacent-prior-art disposition and a researchable residual claim. It cannot establish world novelty, patentability, freedom to operate, market size, expert acceptance, compliance across jurisdictions, operational effectiveness, or realized value; unindexed products, patents, proprietary audit methods, and differently worded systems may contain a closer match."}