← All tools

Tool

Journal Entry Test (JET)

46 criteria run together over every journal. Each criterion is individually checked in a self-test, and 47 built-in test cases can be re-run at the push of a button.

Journal Entry Testing

Analysis of 100,000 journal lines in 2 to 3 seconds.

The JET is the core of every journal entry review. Audit Chen runs every criterion directly in your browser, on your machine. The journal is not uploaded for the analysis.

2–3 sanalysis of 100,000 journal lines, without import
46test criteria, each with a reason
8audit test groups, from timing to statistics
LocalThe analysis runs in your browser on your machine

Measured with synthetic test data (100,000 lines, 5 MB) on an Apple M1 with 8 GB RAM in Chrome, two runs on 3 November 2026: JET analysis 2.0 to 3.0 s, import 0.7 to 0.8 s. Your hardware may differ.

Journal (synthetic)Analysis in browserImport
100,000 lines2.0 to 3.0 s0.7 to 0.8 s
1 M lines17.5 s4.7 s

Guide value: JET has been tested several times on real journals with around 300,000 lines.

Honest note: at 1 M lines the browser becomes slow (same machine, one run). That is why we are building the desktop variant. An earlier approach searched only the relevant population instead of loading everything. That is more efficient.

In progress: the desktop variant is meant to test 100,000,000 journal lines. We publish results once they are measured.

Only conspicuous entries in the list

The hit list shows the entries that trigger at least one criterion, not the whole journal. The cut-offs for the rating can be adjusted.

Traceable for the regulator

Audit steps are recorded in the audit log and can be exported.

The AI proposes, you decide

Every AI finding is marked as a draft. Sign-off rests with the people on the audit team who are authorised to give it.

What the JET actually tests

46 test criteria. Every hit comes with a reason.

Every line of the journal is tested against all active criteria. The criteria are grouped by theme, and each can be switched on or off.

7

Timing

Weekend, public holiday and off-hours postings, period-end, backdating, postings after closing.

7

People and approval

Broken segregation of duties, special accounts, rare preparers, missing approval, manual general ledger postings.

2

Management influence

Signs of management instructions or overrides, corruption risk in consulting and commissions.

11

Amount and pattern

Round amounts, just below approval thresholds, split series, duplicates, cent clustering, micro-amount series, circular account flows, outliers.

8

Accounts

Unusual account combinations, suspense and estimate accounts, postings that bypass the income statement.

5

Text and documentation

Vague or missing posting texts, placeholders, missing documentation, unbalanced documents.

4

VAT

Input VAT without expense, wrong tax rate, incomplete reverse charge, missing tax coding.

2

Statistics

RPCA and Isolation Forest find outliers that no fixed criterion covers.

From the journal to the audit step

  1. Lock the import. Every imported file is logged with a SHA-256 checksum. Later changes to the source file become visible.
  2. Score the whole population. Every posting receives points for each criterion it triggers.
  3. Classify. High, medium-high, medium, low or none. High needs at least 60 points and two strong signals (standard cut-offs). The calculator below shows the rules.
  4. Give reasons. For each document: an explanation in auditor language, the likely audit step, and the document to request.
  5. Select. Conspicuous documents are listed one by one. The JET draws a sample for the rest. With a seed the draw is reproducible.
  6. Project. The error rate from the testing feeds back into the planning of the internal control system. At 5 % or above, the planning is raised one step.

Try the rating

This is how the JET derives the rating from points and strong signals. The standard cut-offs are 60 points for high and 30 for medium. The app lets you adjust them. Simplified: the threshold below which nothing is flagged is left out.

Medium
  1. High: at least 60 points and 2 strong signals
  2. Medium-high: 60 points and 1 strong signal, or 30 points and 2 strong signals
  3. Medium: 30 points and 1 strong signal
  4. Low: at least 1 point, but none of the conditions above
  5. None: 0 points

Projection and planning

The error rate is the number of failed items divided by the number of tested items. From 5 % the planning is raised one step.

6.7 %

Conspicuous does not mean wrong

The analysis tells you which documents to look at individually. It is not evidence of a misstatement.

More than rules

On request, a Benford analysis, round-trip detection in the network, a behaviour analysis and a micro-forensics check also run.

Checked against built-in test cases

The criteria are checked against built-in test cases, including combined fraud scenarios. We only publish a hit rate once it has been measured on real, independent data. Clean test postings rarely reach a medium or higher rating in the tests. The JET is a selection tool and does not find everything. The PDF export documents every tested posting.

Data cleaning

What a real journal looks like before and after.

Sample data with typical everyday errors. Every change is marked and recorded in the audit log.

Raw file (ERP export)

Doc noDateAccountTextAmountD/CEntered by
BN-1023110.03.2026"4000 "Rent March1.250,00Smueller
BN-1023110.03.2026"4000 "Rent March1.250,00Smueller
BN-102322026-03-116010Salary January 3.480,50HHuber
BN-1023311/03/20264711Office supplies-120,00S
BN-1023414.03.20261200Customer Müller GmbH12.400,00S

Cleaned (analysis basis)

Doc noDateAccountTextAmountD/CEntered by
BN-102312026-03-104000Rent March1250.00Smueller
BN-10231Duplicate removed, noted in the audit log
BN-102322026-03-116010Salary January3480.50Hhuber
BN-102332026-03-114711Office supplies-120.00Smissing – flagged
BN-102342026-03-141200Customer Müller GmbH12400.00Smissing – flagged

Green = standardised automatically. Yellow = the auditor decides. Struck through in red = duplicate, not deleted but recorded.

Tempo

From raw journal to analysis — in seconds.

Sample run with invented demo data. The rule is shown in simplified form and is not Audit Chen's program code.

rule.txt

        
0 postings cleaned