Question and scope
This protocol measures how often an approved cross-team exception continues past the date its approval expires. An exception is any temporary permission to skip, replace, or delay a standard step, such as a manual approval while a system is down or a waived review during a peak period. Exceptions are legitimate management tools, but an exception without a live expiry date slowly becomes the new process. The measure tests whether the organisation still knows which exceptions are active.
Scope is one process family handled by one coordination team during a defined window, such as all exceptions opened in the last two quarters. Include exceptions granted verbally if they were later written down. Exclude ordinary one-off corrections that never changed a required control. State the process family, the source system, and the window so another reviewer can rebuild the boundary.
What counts as an exception
An exception is a documented departure from an approved step that carries a named approver and an end condition. That end condition may be a date, a system recovery, or a completed remediation. A request that only clarifies how to apply the standard is not an exception. A request that permanently changes the standard belongs in a change review, not in the exception register.
The distinction matters because exceptions and permanent changes have different risks. An exception concentrates risk in a short window and depends on someone remembering to close it. A permanent change should update the procedure, the training, and the audit test. Counting both as one event hides the population that most often drifts. Write the definition before coding so a minor workaround is not treated as a formal waiver.
Unit of analysis and exclusions
The unit is one exception approval with one process step, one approver, and one end condition. Split a bundled approval when each step has its own expiry. If the steps share a single end condition and approver, keep one unit and record that the scope was shared. Use the approval as it was recorded at the time, not a rewritten version created during the audit.
Exclude duplicate entries, test records, and requests that were rejected before any work changed; duplicates are common when a register and a ticket system both record the same request. Exclude exceptions that were closed on the same day they opened, because they carry little drift risk and can distort the age distribution. Report the exclusion count and one reason for each excluded unit. The denominator should be defendable without rereading every message.
Start and stop clocks
Start the clock on the date the exception was approved, not the date it was first discussed. If the approval date is missing, use the first date the exception appears in the work record and mark the start as estimated. An estimated start is weaker evidence and should be reported separately. Do not use the date the exception was entered into the register if that happened later.
Stop the clock on the earlier of two events: the exception was formally closed, or the approval expiry date passed. For units still open at the cut-off, record the age at the cut-off and mark them as censored. Censoring keeps open exceptions in the analysis without pretending they have ended. Report censored units as a count so a long tail is visible.
Evidence and data dictionary
Each unit needs an identifier, the process step, the standard requirement, the reason for the departure, the approver, and the approval date. Add the stated end condition, the actual close date, and the evidence that the standard step resumed. Link to the ticket, message, or meeting note where the approval was granted. The evidence should let a reviewer confirm that the exception existed and that it ended.
Define the terms before coding begins. Distinguish an open exception, an expired exception, a renewed exception, and a closed exception. Decide whether a renewal creates a new unit or extends the same one. Record one example of each category in the dictionary. A short shared vocabulary prevents two reviewers from counting the same record differently, which is the most common source of disagreement.
Sampling and test design
Use a census when the window contains fewer than roughly one hundred exceptions. Above that volume, draw a random sample from the register export and document the method. Stratify by process step or requesting team so a high-volume category does not dominate the result. Freeze the sample before reading close dates to reduce selection bias, and keep the frozen list with the analysis.
Have a second reviewer recode at least ten exceptions and compare the results. Resolve disagreements by returning to the source record rather than by majority vote, and record the resolution so a later reviewer can see how the definition was applied. If the disagreement rate is high, revise the definitions and recode the full sample. A measure that depends on one reviewer's memory will not survive a change of reviewer or a new quarter.
Decision rules before coding
Decide how to treat an exception that was renewed before its expiry date. One option is to count the renewal as a new unit and keep the original record closed on time. Another is to treat the pair as one continuous exception and measure total days. The choice changes the headline share, so write it down before the first row is coded.
Decide how to handle an exception that expired but was later approved retroactively. Count it as expired at the original date and record the retroactive approval separately. Also decide whether an exception with no stated end condition counts as expired or as permanently open. Both choices are defensible, but the report must say which one was used and why it was chosen.
Analysis and reporting
Report the share of exceptions that passed their approval date while still in use, the median days past expiry, and the share with a recorded close date. Add the share that were renewed at least once and the share still open at the cut-off. Show the raw count beside every percentage, especially when the sample is small. A share based on five exceptions should be presented as a count, not a rate.
Break the result down by process step, approver group, and reason for the exception. Suppress groups with fewer than five units so one unusual record does not look like a pattern. Pair the expiry result with the time to close after the expiry was noticed. An exception that is expired but closed quickly is a different problem from one that stays open for months, because the second kind quietly becomes a priority claim in the weekly blocked-work review.
Limits and sources of error
The cited public sources describe internal control and risk management in general terms; they do not provide a benchmark for a specific exception register. A missing close date may mean the exception ended without a record, or that nobody closed it. The measure also depends on whether the register preserves the original approval date. A system that overwrites the date will understate age.
Verbal approvals and backfilled records are common sources of error. Check a sample against the original messages and mark any row entered after the fact. Exclude a unit when the approval or the end condition cannot be verified, but report the exclusion count. A precise number built on guesses is worse than a range with a clear caveat, especially when the register is known to be incomplete.
Review cadence
Review the result at a fixed point and end with one change, one owner, and one check date. Repeat the test after two full cycles before setting a target, because a single quarter may reflect an unusual system outage. Keep the definition and the sample method with the result so the next reviewer can compare like periods. Mark any process change on the same timeline as the data.
Retire the measure if it stops informing a decision, and use the priority dispute rulebook when an old exception is really a contested claim on capacity. If the team keeps it, publish the expired share and the median days past expiry together, because a low expired share can still hide a few very old exceptions. The purpose is a register that reflects reality, not a perfect count. Stop when the team can show that active exceptions are known and owned.
Sources
- U.S. Government Accountability Office, Standards for Internal Control in the Federal Government, reviewed November 10, 2026.
- HM Treasury, The Orange Book: Management of Risk - Principles and Concepts, reviewed November 10, 2026.
- National Institute of Standards and Technology, SP 800-39: Managing Information Security Risk, reviewed November 10, 2026.
