Connect Functional Risk to a Proportionate Assurance Approach
A guide to risk-based software assurance decisions. Follow one synthetic function from intended use and risk reasoning to the chosen activity, the expected evidence and the handling of an unexpected result.
Written by CSV/CSA Maturity Editorial TeamPublished 12 min read
On this page8 sections
A proportionate assurance approach is built one function at a time. For the function in question, record what it does in your process, what would follow if it failed and what else would catch that failure. Then choose the testing or other assurance activity that fits this reasoning, say why it fits, and decide before execution how an unexpected result will be handled. The effort follows the risk reasoning for that function. It does not follow the size of the system or what the team did last time.
This guide is for quality and technical owners who decide how much assurance a software function needs, in teams that practice computer system validation (CSV) or computer software assurance (CSA). CSV here means validation, not comma-separated values. The guide follows one synthetic function through a decision record and sets a second function beside it where a different method is reasonable. It does not classify your systems or decide which regulations apply to them.
Where the regulatory sources stand
FDA's final guidance on computer software assurance, issued on February 3, 2026, gives nonbinding recommendations for software "used as part of production or the quality management system for medical devices" (FDA guidance, PDF [S03]). Its status and history are on the FDA guidance page [S02]. The document gives no recommendations for software that is itself a medical device, and it does not mention pharmaceutical manufacturing, laboratory or clinical systems.
Within that scope, the guidance describes a risk-based approach to establishing and maintaining confidence that software is fit for its intended use. It has four parts: identify the intended use, determine the risk-based approach, determine the appropriate assurance activities and establish the appropriate record (FDA guidance, section V [S03]). FDA presents this as a supplement to its 2002 General Principles of Software Validation [S35], and the 2026 guidance says the underlying regulation requires device manufacturers to validate such software for its intended use. Our reading is that FDA frames assurance as a risk-based way of meeting that requirement, not as a regime that removes validation.
Medicinal products in the EU fall under a different text. Annex 11 of the EU GMP guidelines says risk management should be applied throughout the lifecycle of a computerized system (clause 1) and that there should be evidence of appropriate test methods and test scenarios (clause 4.7) (Annex 11 [S36]). As of November 12, 2026, the EudraLex Volume 4 index [S04] lists Annex 11 as the January 2011 revision.
Our reading is that the pattern both texts point to (reason about the function, fit the activity to the reasoning, keep a record of both) is useful beyond device manufacturing. That is editorial synthesis: the FDA guidance does not say it applies to pharmaceutical or laboratory systems. The comparison of what CSV and CSA each cover goes further into the scope of each source.
Seven steps from one function to a recorded decision
The steps are a working sequence, not a mandated procedure. A synthetic example runs through them: a production-tracking system at an invented medical device manufacturer, a setting the FDA guidance addresses.
1. Identify the function
Start from something a person relies on. A module name is too coarse. A usable statement names the trigger, the action and the result: "When an in-process measurement is recorded outside its limits, the system places the lot on hold and prevents it from moving to the next operation."
The FDA guidance works at the same level: it recommends examining the intended use of individual features, functions and operations (FDA guidance, section V.A.1 [S03]). If the function cannot be stated in a sentence or two, documenting intended use and acceptance expectations comes first.
2. Explain the potential effects in context
Ask what happens in your process if the function fails, and name the failure mode: the hold is not applied, it is applied to the wrong lot, or it is applied too late. Then ask what stands between that failure and a consequence.
In the example, a missed hold lets a nonconforming lot continue, and the only later check is a final inspection that does not repeat this measurement. Write both facts down. The same function in a process with an independent downstream check would have a different effect.
3. Document the risk reasoning
Turn the effects into a short rationale that a reviewer can argue with. The FDA guidance presents risk in two categories, "high process risk" and "not high process risk". It describes a function as high process risk when its failure may result in a quality problem that foreseeably compromises safety, and it says risk sits on a spectrum, with the two categories used for presentation (FDA guidance, section V.A.2 [S03]).
Whatever scale your procedure uses, record the reasoning and not only the label. "High" gives a reviewer nothing to test. "A missed hold could let a nonconforming lot continue, and no later step repeats the measurement" tells them what the assurance activity has to show.
4. Select suitable assurance activities
Choose activities that would reveal the specific failure if it existed. For the hold function, that means exercising the limit boundaries, the hold itself and the block on moving the lot, with expected results fixed in advance so that a pass means the same thing whoever runs the test. The team in the example adds one unscripted session in which a tester tries to get around the hold, because predefined steps find only the failures someone thought to write down.
5. Capture the rationale
Record why these activities fit this function, in the same record as the decision, so that a reviewer who was not in the room can reconstruct the choice.
The FDA guidance lists what the record should capture: the intended use, the result of the risk-based analysis, a description of the testing, any issues found, a conclusion, who performed the work and when, and review and approval when appropriate (FDA guidance, section V.A.6 [S03]). It says the record "need not include more evidence than necessary" for the risk identified, and it recommends digital records such as system logs and audit trails in preference to paper, screenshots or duplicated results (same guidance [S03]). Our reading is that this concerns the amount and form of evidence. It does not say that a lower-risk function needs no record.
The record also has to stay findable, which following one need to its evidence version tests.
6. Address unexpected results
Decide before execution what happens when a result is not the expected one: who records it, who owns it, who assesses the impact and who may approve a disposition. An unexpected result is information about the function or about the test, and either way it belongs in the decision.
7. Review when the function or its use changes
The rationale belongs to a function used in a particular way. A change to the limit logic, a new route by which lots can move, or a decision to drop a manual check because the hold exists each reopens the reasoning. List those triggers in the record.
A synthetic decision record for two functions
| Record field | Case A: out-of-limit hold | Case B: open-lot count for planning |
|---|---|---|
| Function | When an in-process measurement is recorded outside its limits, the lot is placed on hold and cannot move to the next operation. | A planning view shows the number of open lots per work center, refreshed when the page loads. |
| Intended-use boundary | Relied on as the control that stops a nonconforming lot. Covers measurements entered at the three in-process stations. Final inspection is a separate function. | Used by shift supervisors to plan staffing. Not used to release, hold or disposition product; those decisions are made from the lot record. |
| Risk rationale | If the hold fails, a nonconforming lot could continue, and no later step repeats this measurement. Under its procedure the team treats the function as high process risk. | If the count is wrong or stale, a supervisor may misjudge a shift's workload. The lot record is unaffected and the error is visible on the floor. The team treats the function as not high process risk, provided the boundary holds. |
| Chosen activity | Scripted tests with predefined expected results at, just inside and just outside each limit. A scripted check that no role can move a held lot. One recorded exploratory session aimed at getting around the hold. | Unscripted scenario testing against a written objective: compare the displayed count with the lot list for three work centers, including one with no open lots. |
| Expected evidence | Executed test record TP-014 with actual results. System log extract showing each hold event. Session notes on what was tried and found. Conclusion statement. | Session record TS-022: objective, what was compared, the result, any observation, tester and date. Conclusion statement. |
| Reviewer | Quality lead (role), independent of the tester. Approval required before release. | System owner (role) reviews the session record. Quality lead (role) confirms the boundary statement. |
| Unresolved issue | ISS-007: a lot can be moved while an out-of-limit value is entered but not yet saved. Owner assigned, impact assessment in progress, release on hold. | None from testing. Open question for the reviewer: do supervisors use the count for anything beyond planning? If so, the boundary and the rationale are reopened. |
| Review trigger | A change to limit logic, lot routing or the stations in scope. Any decision to drop a related manual check. | Any use of the count in a product or quality decision. A change to how the count is calculated. |
What the two cases show
Case A is where additional rigor is justified. The function is the control, nothing downstream repeats it, and a failure reaches the product. Predefined expected results, a check across every role and independent approval are proportionate to that reasoning.
Case B is where a different method is reasonable, subject to qualified review. A knowledgeable tester comparing the view with the lot list can cover the ground without a step-by-step script, and the session still leaves a record with an objective, a result, a name and a date. The method depends on the boundary, which is a claim about how people use the view. If supervisors turn out to hold or move lots on the strength of the count, the function is doing a different job and the reasoning for Case A applies.
Handling the unexpected result in Case A
During the exploratory session the tester entered an out-of-limit value at one station and, before saving, moved the lot from a second workstation. The hold is evaluated when the measurement record is saved, not when the value is entered.
The observation and the steps to reproduce it were recorded before anyone discussed whether it mattered. Three dispositions were then considered: a configuration change so that the hold is evaluated on entry, a procedural control requiring a save before leaving the station, and a narrower statement of what the hold is relied on for. The method does not presume which is right. It requires that an authorized person chooses one and writes down why, and that the release decision waits for that approval.
| ISS-007 field | Entry (synthetic) |
|---|---|
| Observation | A lot can be moved from a second workstation while an out-of-limit value is entered but not saved. |
| Owner | Production systems lead (role) |
| Impact | Bears on the intended use: a nonconforming lot could move before the hold applies. |
| Disposition | Configuration changed so that the hold is evaluated on entry. Affected steps of TP-014 re-executed and the bypass attempt repeated without success. Approved by the quality lead (role). |
| Follow-up | Hold events sampled from the system log after the first weeks of use. No movement between entry and hold found. Result attached to ISS-007. |
A justified, approved disposition is a controlled outcome. The assessment model's guidance for its issue-handling question says not to equate maturity with having no reported issues.
Scripted and unscripted testing as contextual methods
Neither method outranks the other. The FDA guidance describes both: it gives scenario testing, error guessing and exploratory testing as forms of unscripted testing, says more rigor such as scripted or hybrid testing may suit high process risk functions, and adds that its examples are not exclusive to either risk category: unscripted testing may be better suited even for some high process risk functions, and scripted, automated testing may be more efficient for some that are not (FDA guidance, section V.A.4 [S03]).
| Consideration | Scripted testing tends to fit when | Unscripted testing tends to fit when |
|---|---|---|
| What a failure would look like | You can name the failure and the exact condition that would reveal it. | You suspect failures that nobody has listed yet. |
| Repeatability | Another person must be able to repeat the step and get the same result, for example after a change. | The value comes from a knowledgeable tester following what they observe. |
| Consequence of a missed failure | Serious enough that predefined coverage and independent review are warranted. | Limited, visible in use, or caught by another control. |
| Record produced | Steps, expected results, actual results, pass or fail. | Objective, what was tried, what was found, conclusion, tester and date. |
Case A uses both. "Unscripted" describes how the tester works. A record is still expected, and a session with no objective and no notes supports no decision.
Testing is also not the only assurance activity. Depending on your procedures, a team may give weight to supplier evidence it has checked against its own intended use, to a downstream control that would catch the failure, or to monitoring after release. Whether these count is for your quality system and qualified reviewers to decide, and the decision record is where the reasoning goes.
How the record maps to the assessment
In model version csa-maturity-2.0.0, the risk-based assurance practice area has three questions. Each is answered with one of four practice statements scored 0 to 3, or with "I do not know yet".
| Question | What a 2 describes | Where the decision record shows it |
|---|---|---|
| Q05 | "Function-level risk reasoning is documented and used to determine assurance effort." | Risk rationale and chosen activity |
| Q06 | "The approach is justified for the function, risk and available evidence, with appropriate records." | Chosen activity, expected evidence and reviewer |
| Q07 | "Issues have an owner, an assessed impact and a documented resolution or justified disposition." | Unresolved issue and the ISS-007 record |
A 3 on each question describes review over time: the rationale is revisited when use changes, activity choices are checked for whether they find meaningful issues, and repeated issues are examined for patterns. One good record cannot show that.
Q05 is one of three priority questions in the model. If it is answered 0, the profile carries a priority review prompt for that practice whatever the overall score. In a synthetic profile where every answer is 2 except Q05 at 0, the risk-based assurance area scores 44.4, the overall self-reported score is 63.5 (Defined), and the prompt remains. If Q05 is answered "I do not know yet", it is not counted as zero: the profile is partial, numeric coverage is 94.4%, no overall score or band is shown, and the possible range is 63.4 to 68.3.
You can answer the risk-based assurance questions for a scope you choose. The assessment asks for a scope, takes you through the seven practice areas and a review step, then asks for your first name, last name, work email and company before it shows the profile, which you can download as a PDF. The result is a self-reported profile for planning, not an audit or a certification.
What this guide does not establish
- The two cases show a way of reasoning, not the right answer for any real function.
- Calling a function high process risk in your own record is your organization's judgment under its procedure. It is not a regulatory classification, and this guide cannot tell you which regulations apply to a system.
- The FDA guidance cited here is nonbinding and addresses software used in medical device production and quality management systems. Annex 11 is EU GMP guidance for medicinal products. Carrying reasoning from either into another setting needs review by qualified specialists.
- A well-kept decision record shows that the reasoning can be reconstructed. It does not show that the reasoning was correct, that testing found every defect or that a control is effective.
Related reading
- Review validation traceability with one evidence chain: checks whether a decision record like this one can still be followed to the right evidence version after a change.
- CSV vs CSA: scope, reasoning and evidence: read this before carrying reasoning from the FDA guidance into a setting it does not address.
- Review software changes and supplier releases: covers the change review that step 7 hands off to.
This article describes assessment model csa-maturity-2.0.0.
Sources and references
- S02Computer Software Assurance for Production and Quality Management System Software. U.S. Food and Drug Administration.
Final guidance (nonbinding), dated February 2026; covers software used in medical device production and quality management systems.
- S03Computer Software Assurance for Production and Quality Management System Software: Guidance for Industry and Food and Drug Administration Staff. U.S. Food and Drug Administration.
Final guidance document (PDF), issued February 3, 2026; nonbinding; supersedes the final guidance issued September 24, 2025.
- S04EudraLex - Volume 4 - Good Manufacturing Practice (GMP) guidelines. European Commission.
Official index of EU GMP guidelines; listed Annex 11 as 'Computerised Systems (revision January 2011)' when checked on November 12, 2026.
- S35General Principles of Software Validation. U.S. Food and Drug Administration.
Final guidance (nonbinding), January 2002; FDA's 2026 computer software assurance guidance supersedes its Section 6.
- S36EudraLex Volume 4, Annex 11: Computerised Systems. European Commission.
EU GMP guideline, revision 1, with a stated deadline for coming into operation of June 30, 2011; the version linked from the EudraLex index on November 12, 2026.