Skip to content

AI uses for records

An accrual reversal temporal dataset for the month a reversal distorts

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

An accrual reversal temporal dataset should link every accrual to its reversal and to the invoice that settles it, with each entry's posting date, creation date and the period it serves, then grade how a model reads the following month. In eight constructed cases, seven showed an October reader a misleading figure in the ledger, a flash report or a department view, while seven of the eight three-month totals still tied.

Value drivers

Accrual reversal temporal

This page covers an accrual reversal temporal dataset for the month a reversal distorts.Illustrative

Key takeaways

  • Store the chain, not single entries: accrual, reversal, invoice and the reports a reader saw, joined by IDs across at least three periods.
  • The reversal date is a system setting: NetSuite can enter it at once or create it on the date, and Axelor offers today, tomorrow, a chosen date or the original entry's date.
  • In eight constructed cases, seven showed an October reader a misleading figure somewhere, and seven still tied over three months.
  • IAS 34 leaves earlier interim periods unadjusted and asks for disclosure of changes in their estimates, so later periods are expected to carry earlier costs.
  • Grade linkage, period attribution, reversal mechanics, reporting context and restraint separately, and include genuine swings so blame-the-reversal answers fail.

What should an accrual reversal temporal dataset capture?

It should capture the chain, not the entries one by one: the accrual, the reversal it triggers, the invoice that finally carries the cost, and the reports that showed each month along the way. A single month's ledger cannot tell a reader whether a credit is a saving or the echo of last month's accrual; the link can.

The question this dataset answers is narrow. How large an accrual should have been is an estimation problem with its own evidence. Here the issue is timing: on which date the reversal lands, which month it lands in, and what that does to the figures someone reads for the following month. The problem is built into an auto-reversing accrual, which cancels itself on a fixed date whether or not the invoice has arrived.

Every case therefore carries four dates per entry (service period, posting date, creation date and, for reversals, the rule that set the date), the IDs that join accrual to reversal to invoice, and dated snapshots of the reports a reader actually saw. The label is a reading: what the following month's figure means, and how much of it belongs to an earlier month.

Where the reversal date comes from, system by system

From a setting chosen when the accrual or the cut-off run is created, and that setting decides whether the next month's first report includes the reversal at all. Accounting systems differ on three points that matter to a dataset: who picks the date, whether the reversal exists before that date, and whether it stays tied to the original.

Oracle's NetSuite documentation has the user enter a Reversal Date on the journal. With Defer Entry checked, the reversal is a memorized transaction created automatically on that date, so it does not exist in the ledger before then; with Defer Entry cleared, it is entered immediately with that transaction date. The documentation adds that a reversing entry is permanently linked to the original and that any change to the original affects the reversal.

The open-source Axelor Open Suite (commit e7cbea5, 16 October 2026) shows the same choices in code. Its accounting cut-off batch stores a move date and a separate reverse move date with an automatic reverse flag. A manual reversal offers four dates: today, the date of the original move, today plus one, or a chosen date. The core reversal service refuses a date before the original entry, copies the original's origin reference and description, and keeps each line's cut-off start and end dates, but in the code we read it sets no field that points back to the original entry. Xero's repeating-journal setup offers an optional auto-reversing date.

The options below, all documented in those sources, produce different next-month readings from the same accrual:

Where the reversal date comes from, system by system
Reversal date optionWhere documentedHow the date is setWhat the next month's first report shows
Fixed date, entered at onceNetSuite, Defer Entry clearedEntered on the journal; the reversal posts at once with that dateThe reversal, whether or not the invoice has arrived
Fixed date, created on that dateNetSuite, Defer Entry checkedEntered on the journal; the reversal is created automatically on that dateNothing until the date; early flash reports miss it
Separate reverse date in a cut-off runAxelor cut-off batchA batch setting apart from the move date, with an automatic reverse flagWhatever month the reverse move date falls in
Date of the original entryAxelor manual reversalChosen at reversal; posts on the accrual's own dateNo reversal; the earlier month is rewritten instead
Today, today plus one, or a chosen dateAxelor manual reversalChosen at reversal; the day of the action or a picked dateDepends on when someone acted, not on the invoice

Eight constructed cases: what October showed and what it cost

In seven of eight constructed cases, a reader looking at October saw a figure that misstated October's cost, in the ledger, the first report, a flash or a department view, and in seven of the eight the September to November total still matched the cost incurred. The cases are deliberately simple: one service costing 100 units a month, an accrual at each month-end for any month not yet invoiced, a reversal on the first of the next month and the invoice on the tenth, with exactly one mechanic changed per case.

A short script recomputes every figure from the entries, using integers only and no random inputs. October's view counts as misstated when the ledger, the figure first reported on 3 November, a 1 to 12 October flash report or a department's October figure differs from the cost that view should show (for the flash, what a first-of-month reversal would show). No real company data is involved; the point is the mechanism, so the numbers are illustrative.

Read the October column against the incurred cost of 100 (120 for September in case 8):

Eight constructed cases: what October showed and what it cost
Case (constructed)Ledger Sep / Oct / NovWhat October seemed to sayCorrect readingField that settles it
1. Control: invoice arrives inside October100 / 100 / 100Flat costFlat; reversal and invoice cancel in OctoberReversal linked to its accrual
2. September invoice slips to 5 November100 / 0 / 200October was free, November doubledSeptember's reversal ran with no invoice and no re-accrualOpen-accrual check at 31 October
3. Accrual of 90 against an invoice of 10090 / 110 / 100Cost up about 22 percentFlat; 10 of October's figure belongs to SeptemberAccrual-to-invoice link and true-up
4. Reversal back-dated to 30 September on 6 October0 / 200 / 100 (Sep first reported 100)September was free, October doubledFlat; a reported month was rewrittenCreation date versus posting date
5. Manual September accrual never reversed100 / 200 / 100October doubledDuplicate cost; total overstated by 100Accrual with no reversal but an invoice
6. Invoice coded to Dept B, accrual and reversal on Dept A100 / 100 / 100 (Oct: A 0, B 100)Dept A saved, Dept B overspentSeptember's cost moved between departmentsCost center on all three entries
7. Reversal deferred to 15 October100 / 100 / 100 (flash to 12 Oct: 100)Early October already at a full monthThe flash counted September's invoice before the reversalScheduled reversal date, flash date
8. September accrual edited on 8 November120 / 100 / 100 (first reported 100 / 120 / 100)October came in at 120October was 100; the edit moved 20 back to SeptemberEdit log on the accrual and its reversal
Four small column charts of September, October and November ledger cost against a flat incurred cost of 100: the control case is flat, a slipped invoice shows October at 0 and November at 200, a back-dated reversal shows September at 0 after first reporting 100 and October at 200, and an accrual never reversed shows October at 200.
Illustrative: constructed cases 1, 2, 4 and 5 from the casebook; values in cost units.

Why the month after carries an earlier month's cost

Because the reversal is a separate entry dated in the new period, it carries the old month's cost into the new month's totals. When the invoice lands in the same month, the two cancel and nobody notices. When it does not, the new month shows a credit, a double charge or a true-up that belongs elsewhere, and a variance review can attribute it to the wrong month.

IFRS interim reporting already assumes that later periods absorb changes to earlier estimates. IAS 34 measures interim figures on a year-to-date basis, notes that this may involve changes in estimates of amounts reported in earlier interim periods of the year, and states that the amounts reported in prior interim periods are not retrospectively adjusted. It requires disclosure of the nature and amount of such changes. A reader of the later period is meant to be told what part of the number is about the past.

A back-dated reversal does the opposite, as case 4 shows: it edits the earlier period after its figures were reported, which leaves the later reader comparing against a number that no longer exists in the ledger. A linked edit, as in case 8 and as NetSuite's documentation describes, has the same effect on both months at once. A dataset that stores only today's ledger cannot tell these apart.

Reporting context: what the reader held on each date

The reading depends on what the reader had, so the dataset has to keep it. For each case, store the report snapshots as they were issued: the month-end figure when first reported, any flash or forecast that used partial-month data, and the ledger as of the label date. Case 7 is right at month-end and wrong on 12 October; only the snapshot date tells a grader which question was asked.

Each snapshot needs its own ID, the timestamp it was produced, the filter it used and whether the period was open, closed or reopened at the time. Keep the creation timestamp of every entry beside its posting date, because a back-dated or deferred reversal is visible only in the gap between the two.

Version chains for revised figures are a related but separate structure: they record each new value of one reported number. Here the unit is a chain of entries across months, and the label is how one month should be read given the others.

Reviewer rubric for reading the month after an accrual

Grade a model's reading of the following month on five points, each pass, partial or fail, reported separately. A model can name the right total and still attribute it to the wrong month, which a single score would reward.

At a glance

Reviewer rubric for reading the month after an accrual

Two accountants label each case without seeing each other's work, a third resolves conflicts, and the same five points are then applied to a model's answer:

  • Linkage: the answer connects each reversal to its accrual and to the invoice that settles it, citing IDs rather than matching amounts.
  • Period attribution: the answer states how much of the month's figure belongs to that month and how much to an earlier one.
  • Reversal mechanics: the answer identifies the rule that dated the reversal (fixed date, deferred, manual, back-dated) and whether it ran without an invoice.
  • Reporting context: the answer says which snapshot it is reading and flags figures that changed after they were first reported.
  • Restraint: the answer does not call a reversal-driven swing a trend, a saving or an overrun without naming the entries behind it.
  • Automatic fail: the answer recommends deleting or re-dating a reversal in a reported period without stating the approval and disclosure that requires.

A three-month total that ties is not a correct October

No. Seven of the eight constructed cases tie over three months, and six of those seven still hand the October reader a wrong figure, a wrong department or a figure that later changed. Arithmetic that closes over a long enough window says nothing about whether each month was presented correctly, or whether each reversal was supported and authorized.

Regulators have treated reversal timing as a judgment, not bookkeeping. In 2015 the SEC charged Bankrate and former executives, saying the then-CFO reversed a marketing accrual in the second quarter of 2012 to help meet analyst estimates. Reducing an accrual lowers the related expense in the period the reversal lands, which is exactly why the label must record the evidence and the approver behind each reversal, not only its date.

The strongest alternative explanation for a swing is real activity: October may really have been cheaper. The dataset should therefore include cases where the swing is genuine, with invoices and service records to prove it, so a model that blames every swing on the reversal fails too.

Linked task packets from real ledgers

Build each case as an authorized, anonymized task packet that spans at least three periods: the month before the accrual, the month it reverses and the month the invoice settles. Shift dates by a constant per packet if needed, but keep intervals and month boundaries intact, since the test depends on which side of a month-end each entry falls.

At a glance

Linked task packets from real ledgers

Replace vendor, employee and cost-center names with stable codes, keep entry IDs and their links, and have a qualified accountant confirm a sample of period labels against the company's close policy before the packets are used. Each packet should hold:

  • Entries: accrual, reversal, invoice and any true-up, each with entry ID, account, cost center code, amount or approved scaled amount.
  • Dates: service period, posting date, creation timestamp and, for reversals, the dating rule and scheduled date.
  • Links: accrual-to-reversal and accrual-to-invoice IDs, plus the matching basis where the system kept only a text reference.
  • Edits: the edit log for any entry changed after posting, with old and new values and the editor's role.
  • Period status: open, closed or reopened on each entry's creation date, from the period-control history.
  • Snapshots: the month-end, flash and forecast reports a reader saw, each with its ID and timestamp.
  • Labels: the correct reading of each month, the share belonging to earlier months, and the reviewer record.

Asking for multi-period accrual cases, and the limits of a constructed test

Buyers who need this should ask for chains rather than entries: the system and its reversal setting, the number of consecutive periods per packet, the share of cases with late invoices or edits, and an acceptance test such as every reversal joining to its accrual by ID. SourceX sources datasets to order from U.S. companies, working from the scope, volume, language, format, timing and licensing terms a buyer sets out; no finance data sits waiting in inventory, and no records move until the supplying company has approved scope and terms and signed.

What this page does not establish: how often each case occurs in real ledgers, which reversal settings companies actually use, or how current models read them. The eight cases are constructed to isolate one mechanic each and say nothing about frequency. Results from real packets are not yet published.

Methods note: the constructed figures come from a script that builds each case's entries and sums them by posting month, with the first-reported view taken as of the third day of the following month and the flash as of 12 October; it uses integers only and no random numbers. System behavior comes from NetSuite's documentation, Xero's help, and Axelor Open Suite's source at commit e7cbea53bae88d2540f59a70b09bf0f326f1022e, read on 27 October 2026. US GAAP interim rules (ASC 270) were not analyzed.

Frequently asked questions

What is an auto-reversing accrual?

It is an accrual entry set to post an equal and opposite entry on a chosen date, such as the first day of the next period, so the expense recorded when the invoice arrives is not counted twice. The reversal runs on its date whether or not the invoice has come in, which is why the following month can look wrong.

Why does a late invoice make the next month look cheap?

The reversal removes last month's estimated cost on its scheduled date, but nothing replaces it until the invoice posts. If the invoice slips past the next month-end and the cost is not accrued again, that month shows a credit and the month the invoice finally lands shows the cost twice.

Should a reversal be dated on the first day of the next period?

That keeps the reversal and the invoice in the same month when the invoice arrives during that month, and it fails when the invoice slips further. Whatever rule a company uses, the dataset should record it, along with the date each reversal was created and posted.

Is a back-dated reversal an error?

Not necessarily; the right treatment depends on the company's policy and on whether the earlier period was already reported. A dataset should keep the first-reported figure, the creation date of the reversal and the approval, so a reader can see that a reported month changed afterward.

How is reversal timing different from labeling accrual estimates?

An estimate label asks whether the accrued amount was reasonable on the evidence available then. Reversal timing asks which month carries which cost and how the next report should be read, which can go wrong even when the estimate was exact.

Sources

  • IAS 34 measures interim figures year to date, does not retrospectively adjust amounts reported in prior interim periods, and requires disclosure of the nature and amount of changes in estimates of amounts reported in prior interim periods. Source
  • NetSuite's Reversal Date sets when the reversing entry posts; Defer Entry creates it as a memorized transaction on that date; reversing entries are permanently linked and change with the original. Source
  • Axelor's accounting cut-off batch stores a move date, a separate reverse move date and an automatic reverse flag. Source
  • Axelor's core reversal service rejects a date before the original entry, copies the origin reference and description, and keeps each line's cut-off start and end dates. Source
  • Axelor's manual reversal offers four dates: today, date of the original move, today plus one and a chosen date. Source
  • Axelor's move record has a cut-off origin move field linking a generated cut-off entry to its source. Source
  • In 2015 the SEC charged Bankrate and former executives; it said the then-CFO reversed a marketing accrual in the second quarter of 2012 to help meet analyst estimates. Source
  • Xero's repeating journal setup includes an optional auto-reversing date. Source

Tell us what data you need

Share scope, volume, format, timing and licensing needs. Datasets are sourced on request.

See if your company qualifies