On this page Jump to section
Question and scope
Does editing an existing bank rule in QuickBooks Online change transactions that were already posted under the old rule, or does the edit apply only to transactions downloaded after the change? This question matters to a bookkeeping VA who inherits a client file with months of rule-driven postings and a request to correct a category. The answer decides whether the correction is a rule edit, a journal entry, or a bulk reclassification. The test below separates the rule definition from the transactions it has already touched.
The scope is one QuickBooks Online company file, one bank or credit card feed, and a defined set of transactions downloaded and posted before the edit. It includes transactions posted automatically by auto-add and transactions reviewed and accepted with the rule's suggestion. It excludes rules that were deleted, because a deleted rule cannot be edited and its history needs a different test. The test also excludes transactions that were manually recategorized before the edit, since a manual change is a separate intervention.
Why a rule edit could affect history
A bank rule is a set of conditions and an action, and the action is applied when a transaction is processed. Intuit's help documentation describes rules as applying to downloaded transactions and notes that QuickBooks applies one rule per transaction in list order. The documentation describes forward-looking behavior, but it does not promise that an edit rewrites prior postings. A reviewer therefore has to test the boundary rather than assume it in either direction. That is the gap the audit is designed to close.
Two mechanisms could produce retroactive change. The first is a rule engine that re-evaluates existing transactions against the current rule set when the rule is saved. The second is a background process that reconciles posted transactions with their rule definitions. Either mechanism would leave a visible trace in the audit log or in the transaction's history. The test looks for that trace before and after the edit, so the conclusion rests on evidence rather than on a product description.
Population and unit
Build the population from transactions that were posted under the rule before the edit date. Record the transaction date, the bank description, the amount, the account and class the rule assigned, and whether auto-add or a human confirmed the posting. Include at least one transaction from each month the rule was active, so a later re-evaluation would have something to change in every period. Exclude transactions that the rule never matched, because they cannot show the effect of the edit.
The unit of analysis is one posted transaction, with the rule as the grouping variable. A sample of one transaction per month is enough to detect a broad rewrite, but it cannot estimate how many transactions changed. If the goal is an error-rate estimate, draw a random sample from the full population and record the sampling method. The test in this report is a behavior test, so it uses a targeted sample chosen to make any change visible.
Test setup
Freeze the evidence before touching the rule. Export the transaction list for the affected account, including the date, description, amount, account, class, and last-modified field. Save the rule's current conditions and action as a screenshot or PDF. Record the audit log start point, because the log is the primary evidence that a change occurred. Without a frozen baseline, a later difference can be explained away as a reporting filter.
Create a copy of the rule rather than editing the original for the first pass. A copy lets the reviewer compare the original and the variant without destroying the client's working configuration. If the product does not allow a copy, export the rule first and record the exact settings. Change one condition at a time, starting with the category, and note the time of each save. A test that changes several settings at once cannot show which one mattered.
Procedure
Run the edit in a controlled sequence that starts with a baseline. Record the transaction list and rule settings, then change only the category in the copied rule and save. Refresh the transaction list and compare each baseline transaction's account and last-modified value. Check the audit log for entries tied to the rule or the transactions. Repeat with a condition change, such as narrowing the bank text, to see whether a definition change behaves differently from an action change.
Give the system time to process between steps, and refresh the page from a hard reload rather than relying on a cached view. Record the timestamp of each observation in the same time zone, because a background job that runs near midnight can look like a retroactive change when it is only a delayed write. If a transaction changes, open its history and capture the before and after values. The history panel is more persuasive than a report total because it names the field that moved.
Definitions and measures
A retroactive rewrite means a transaction posted before the edit shows a different account, class, payee, or split after the edit without any human action. A forward-only change means prior postings keep their original values and only transactions downloaded after the save use the new action. A mixed result means some prior postings changed and others did not, which requires a transaction-level explanation. Each outcome gets a separate count and a list of the transactions that support it.
Measure the result at three levels: the transaction, the register total for the affected account, and the rule's match count. A transaction-level change with no register total change may be a reclassification between two accounts that nets to zero, which is still a change the client should know about. The rule's match count is a secondary signal, not proof, because a count can change when the rule is re-saved without altering any posting. Report the three levels separately.
Worked example
Suppose a client has a rule that posts every transfer from its operating account to a savings account as an owner draw. The rule has run for four months, and the client now says the transfers are actually to a tax reserve account. The VA copies the rule, changes the category to the tax reserve, and saves. The test then compares ten transfers posted before the edit with the same ten after a refresh. If all ten keep the owner draw account, the edit is forward-only for that action.
Now suppose the same VA narrows the bank text condition so it no longer matches a second savings account. The test checks whether transactions from the excluded account that were posted under the old rule are re-evaluated and moved, or whether they remain where the old rule put them. A forward-only result means the VA must reclassify the historical transactions manually or with a journal entry. A retroactive result means the correction is already done and the evidence needs to be attached to the close packet.
Evidence standard
The evidence set should include the baseline export, the rule settings before and after, the audit log extract, the transaction history screenshots, and a short memo naming the tester and the time of each step. Each claim in the memo should point to a specific artifact. A statement that the rule did not rewrite history is only credible when the reviewer can show the same transactions before and after with unchanged values. Store the set with the client's rule-change record.
Keep the raw exports alongside the memo, because a summary table can hide a transaction that changed in a field the table omitted. If the audit log is unavailable or incomplete, say so and lower the confidence of the conclusion. Do not fill a gap with an assumption about how the product should behave. The client's tax preparer may rely on this record, so the standard is the same one used for any other bookkeeping evidence.
Interpretation limits
A forward-only result in one company file does not prove that every QuickBooks Online edition or every rule type behaves the same way. The test covers one action, one condition change, and one period. A different result could appear after a product update, a change in the rule engine, or a migration between subscription levels. Record the product edition and the date so a later reviewer can tell whether the result still applies.
The test also cannot see a change that the interface hides, such as a modified timestamp on a child split line that does not alter the parent total. That is why the procedure checks the transaction history and not only the register. A clean result means no visible change in the fields examined, not proof that no database write occurred. State the boundary in the memo so the conclusion is not read more broadly than the evidence allows.
Review cadence
Repeat the test after any product release that touches banking, and after a client migration to a new subscription tier. A quarterly check is reasonable for a firm that relies on auto-add, while a firm that reviews every rule suggestion can test only when a rule changes. The cadence should follow the risk of the action, not the calendar alone. A rule that posts to a loan or tax account deserves a retest sooner than a rule that posts a small subscription.
Close the review with a named owner, the exact rule settings, and a date for the next check. Link the result to the bank rule change review and the bank rules review workflow so the next reviewer starts from the same method. If the client asks for a historical correction, convert the finding into a reclassification plan with an approval step. The test answers what the edit did; the plan decides what the books should show.
