Which entries that someone typed in by hand deserve a second look, and why?
Most entries in your books come from an invoice, a bill, a payment or a payroll run. A person types a few entries by hand. This skill reads every entry in the period. It screens each entry against 21 warning signs. The hand-typed entries are the focus. The few entries that need a question go to the top.
You give it a period, your approval limit and a few facts about your books. It reads every entry and proves that it read all of them. It scores each entry and shows the patterns first. It lists every test that it could not run. A flag is a question for a person. It never says that fraud occurred, and it never changes your books.
1,284 transactions at Northbrook Trading, and 7 flags
Northbrook Trading keeps its books in QuickBooks Online, in US dollars. The skill reviews the quarter from 1 July to 30 September 2026. The run is on Wednesday 14 October 2026.
Why look at the hand-typed entries? A good control system makes sure that at least two people take part in most transactions. Then a second person checks the work of the first1. An invoice or a payment goes through those normal checks. An entry that one person types in by hand can skip them. The skill says it in these words: "the ones typed in by hand bypass the normal checks, and those are the focus."
The tie-out comes first. The skill adds the debits of every entry that it read and compares the total with the movement in the general ledger. The difference is 0.00, so the skill can call the screen complete. If the difference is not zero, some entries are missing, and a clean result means less.
| Entry | Dated | Login | Accounts | Amount $ | Warning signs (points) | Score |
|---|---|---|---|---|---|---|
| JE-202609-027 | Wed 30 Sep, typed in Mon 12 Oct | bookkeeper | DR 6310 Utilities, CR 2110 Accrued expenses | 1,263.40 | Back-dated (3), after the close (3), into a closed period (3), period end (1) | 10 |
| JE-202609-019 | Sat 26 Sep | owner | DR 2450 Warranty Reserve, CR 5000 Cost of Goods Sold | 9,800.00 | Reserve (2), a little below the $10,000 limit (2), login that does not usually post (2), description "adjustment" (1), weekend (1), period end (1) | 9 |
| JE-202609-023 | Wed 30 Sep | bookkeeper | DR 4000 Revenue, CR 2400 Deferred revenue | 33,000.00 | Revenue (2), at or above materiality (2), round amount (1), period end (1) | 6 |
| JE-202609-022 | Wed 30 Sep | bookkeeper | DR 6410, CR 2110 | 9,000.00 | Round amount (1), period end (1) | 2 |
| JE-202608-014 | Mon 31 Aug | bookkeeper | DR 6310, CR 2110 | 2,040.00 | Set to reverse on 1 Sep, but no reversal (2) | 2 |
| JE-202609-024 | Wed 30 Sep | bookkeeper | DR 2110, CR 6310 | 2,040.00 | Period end (1) | 1 |
| JE-202609-026 | Wed 30 Sep | bookkeeper | DR 6150, CR 6520 | 640.00 | Period end (1) | 1 |
| Total | 10 + 9 + 6 + 2 + 2 + 1 + 1 | 31 |
Some low scores need no review. JE-202609-022 is a round accrual at the quarter end. The skill says that "a round-number accrual at period end is a description of most accruals." The history of the last six quarters shows that such accruals are routine at Northbrook. JE-202608-014 did not reverse, but JE-202609-024 already corrects it.
One entry looks fine, and three quarters together do not
Look at JE-202609-019 alone. It is $9,800, below the $10,000 limit, so it needs no approval from the owner. A review of one entry at a time can see nothing wrong. The skill also looks across periods, and it puts the patterns at the top of the report.
One entry at a time
JE-202609-019 by itselfThe pattern across three quarters
The same account, the same login, on a Saturday each timeThe skill calls this "an indicator worth investigating", because the reserve moves in the direction that makes results smooth. It is a question, and it is not a finding. Maybe the warranty claims went down. Maybe the owner shares the login. The skill asks for the warranty claims analysis. It does not guess the answer.
Why ask at all? In a large international study of fraud cases, a lack of internal controls contributed to 32% of cases. An override of existing controls contributed to 19%3. Small organizations are much less likely to have anti-fraud controls3. A screen like this gives an owner-run company a second look after the fact. It finds entries that deserve a question. It does not find fraud.
Three reviews: one error, one correct entry, one open question
| Entry | What the review found | Result |
|---|---|---|
| JE-202609-027 (10) | The September utilities bill came in late. On Mon 12 Oct the bookkeeper typed in a true-up dated 30 Sep. The close was on Fri 9 Oct, so the entry went into a closed quarter. | Error. The date is wrong. The entry must move to the open period |
| JE-202609-019 (9) | The question goes to the outside accountant, with a due date of Fri 23 Oct 2026. The support asked for is the warranty claims analysis. | Open. With no support, it goes to the chair of the audit committee |
| JE-202609-023 (6) | The support is the Pinnacle sales receipt and the service contract. The entry defers 11 months of a payment for a full year. | Correct |
The skill does not correct JE-202609-027. A person does that, or the journal entry builder proposes a correcting entry for a person to post. That skill prepares entries before anyone posts them. This skill looks at entries after a person posts them.
If an item goes to the audit committee, the skill sends no approval request to anyone in the scope of the escalation. The skill says: "escalation is not an accusation." A fraud investigation is outside this skill.
Four approval tests not run, and the report says so
The owner approves hand-typed entries over $10,000 by email. Nobody keeps an export of who approved what. So there is an approval step, but no log. The skill cannot run the four tests that need approval data. It marks them NOT RUN, and it does not leave them out.
Separation of duties is hard in a small company, because a few people do all the work2. One option for that risk is more review after the fact2. That guidance is for US federal agencies. We use it here because it states the control idea clearly. It is not a rule for Northbrook.
The $9,800 release needed no approval at all under a $10,000 limit. That is why the skill tests for amounts a little below the limit. The skill also states that adjustments in the consolidation workbook are not in this population. MosoFin has no connection to that workbook.
Nine questions, in one message
The skill asks these first, in one short batch. It asks only for what it cannot already see. The letters are the skill's own. Question I comes before H.
- Use MosoFin for this run, or upload the data yourself?MosoFin is the connection to your books and other linked platforms. The checks are the same on both paths. The skill asks this only if your account has the MosoFin tools. If not, it says once that the run is manual.
- Which period is under review?Required. The period sets which entries it reads and when the period ends. It screens nothing without it.
- What are your approval limits?Required for the test for amounts a little below the limit. The skill never assumes a limit. Without one, it names that test as not run.
- What size of amount is important enough to matter? Accountants call this materiality.It tests large amounts with it. The skill never assumes a figure. It recommends that you give one.
- Does anyone make adjustments outside the books, for example your accountant at year end, in tax software or in a spreadsheet?Required. Those entries carry more risk, and they may not be in the books that it reads. The report must say if they are in the population.
- Who may see the output?Required. The output names logins and entries.
- Must someone approve a hand-typed entry? If yes, can you export who approved each one?Recommended. Four tests depend on it. With no approval step, the report says "Control not in place". With a step but no log, those four tests are NOT RUN.
- Through which date are the books closed or locked, and when was the close done?Required for three tests: entries after the close, entries into a closed period, and back-dated entries. Without it, those tests are NOT RUN.
- Which accounts are sensitive, and do you know of any risk areas?Recommended. The answer focuses the review. The skill can find revenue, reserves, equity and suspense accounts by their type. You name the rest.
Then it reads the plan back and waits for your yes. Here is the skill's own read-back for Northbrook, word for word:
"Reviewing every journal entry posted by Northbrook Trading for 1 July–30 September, against a baseline from the six prior quarters, read from your Mosofin connection. Authorization threshold $10,000, materiality $25,000. Topside entries are made in the consolidation workbook, which is not connected, so they are not in this population. No approval log, so the four approval-based tests will be reported as not run. Output for the CFO and the audit committee. Shall I go ahead?"
The skill's sample has no line for question I. For this example we add one: "Books closed through 30 September, close done 9 October."
Connect once, or upload the full journal
Say yes, then give it what the books do not hold
- It reads back the workspace and the company file by name, and waits for your yes
- It checks for a connection to a consolidation or reporting platform. With no connection, it says that those adjustments are not in the population
- It reads every entry of the period and of several earlier periods. It also reads the chart of accounts, the general ledger, and the profit and loss and balance sheet
- It finds out what is "normal" from your own history
- You still give it the approval log and the limits (questions C, D and G). The connection does not show them
Bring the journal and the trial balance
- Journal for the period, all transaction types, plus the first month after the period, with the transaction date, create date, created by and transaction type. CSV or XLSX, required
- Opening and closing trial balance, CSV or XLSX, required. It proves that the population is complete
- Entries for several earlier periods, recommended. They show what is normal
- Chart of accounts with account types, recommended
- Approval log, recommended. Without it, four tests are NOT RUN
- Profit and loss and balance sheet by period, CSV, XLSX or PDF, recommended, for the reserves test
- Period, company name, currency, period ends, where outside adjustments live, and who may see the output, typed in the chat
Nine sheets, patterns first
The header gives the population size, the tie-out and whether outside adjustments are in it. The skill asks before it makes an Excel file. Here the name is "JE_Review_2026-Q3_Northbrook Trading.xlsx". If you say no, you get the same sheets as tables in the chat.
After the results, the owner asks: "Save this quarter's flags so we can check them next quarter." The skill refuses. It keeps the method and never the results. On a yes, it saves the sensitive accounts, the limit, the baseline, the weights and rules such as "quarter-end accruals to 2110 are routine". It never saves flagged entries, scores, user names or explanations.
Six rules that never bend
- Never reads MosoFin without asking. A connected account is not permission to read it.
- Never reads anything before you answer the intake questions and say yes to the read-back.
- Never writes, updates or deletes anything, in any system. It makes no entries. If you ask it to reverse or re-date an entry, it declines once, in plain words.
- Never treats a flag as a finding. In the skill's words:
"A flagged list is a list of questions, not of suspects; a user name is not a person; escalation is not an accusation."
If you ask "Who is fiddling the books?", it names nobody. - Never skips the proof or hides a gap. It screens the whole population, ties it to the ledger, and reports every test it could not run as NOT RUN.
- Never assumes an approval limit, a materiality figure or a conclusion about an entry. It never saves flagged entries, risk scores, user names, explanations or anomalies.
Seven terms, in plain English
- Journal entry
- A record that moves an amount between accounts in your books. Some come from invoices and payments. A person types others by hand.
- Population
- All the entries in the period under review. The skill reads all of them, and it does not take a sample.
- Tie-out
- The proof that the entries it read add up to the movement in the ledger. Here the difference is 0.00.
- Back-dated entry
- An entry typed in later than the date that it carries. The test needs both dates.
- Reserve
- An amount set aside for a known future cost, for example warranty claims.
- Topside entry
- An adjustment made outside the books, for example in a consolidation spreadsheet.
- Segregation of duties
- Different people authorize, keep and record a transaction, so one person cannot do all three2.
Where the facts on this page come from
The sources are an accounting textbook, a US government standard and a study of fraud cases. We found each one on 28 October 2026 in search results. We did not open the pages, so follow each link for today's wording. What the skill does comes from the skill's own files.
- OpenStax: Principles of Accounting, Volume 1, 8.3 Describe Internal Controls within an OrganizationA good control system makes sure that at least two people take part in most transactions. Then a second person checks the work of the first. Examples are placing and receiving an order, and writing and signing a check.
- U.S. Government Accountability Office: Standards for Internal Control in the Federal Government (the Green Book), GAO-25-107721Segregation of duties separates authority, custody and accounting. A smaller entity has more difficulty with it, because a few people hold the duties. The 2024 exposure draft (GAO-24-106889) lists more review levels and the review of randomly selected transactions as options. Written for US federal entities.
- ACFE: Occupational Fraud 2024: A Report to the NationsA lack of internal controls (32%) or an override of existing controls (19%) contributed to more than half of the cases. Small organizations are much less likely to have anti-fraud controls. Limited staff usually means fewer checks and less segregation of duties.
No outside authority backs the 21 warning signs, the NOT RUN and "Control not in place" labels, or the nine sheets. They are the skill's own rules. The points for each sign, the 5-day period-end window and the $9,500.00 to $9,999.99 band are this example's choices. The skill tells you to state weights, but it defines none. We made up Northbrook Trading and every figure in the example.
This page explains journal-entry-review, from the MosoFin finance skills library, MosoFin/mosofin-finance-skills on GitHub. Read the skill on GitHub → The skill is in the card format on the main branch. It merged in PR #1 on 26 October 2026, and a correction from the bookkeeping audit followed on 27 October 2026. This page describes that version. Want the worked examples? Learn it step by step →