How to Assess CSV/CSA Maturity Without Turning It into a Compliance Score
A seven-step method for a scoped, evidence-led self-assessment of validation and assurance practices, with a one-page workshop worksheet and a synthetic example of two teams whose practices differ.
Written by CSV/CSA Maturity Editorial TeamPublished 12 min read
On this page7 sections
To assess computer system validation (CSV) and computer software assurance (CSA) maturity, fix one boundary, bring in the people who hold the evidence, sample a few records for each practice area, and select only the practice descriptions those records support. Everything else is marked "I do not know yet". What comes out is a self-reported profile and a short list of owned review actions. It is not a compliance percentage, an audit result or a certificate.
The method below is written for the CSV/CSA Maturity Assessment, model version csa-maturity-2.0.0: 18 questions in seven practice areas, each with four practice descriptions scored 0 to 3 and an explicit unknown. It is meant for a quality lead or system owner preparing a working session with colleagues. CSV here means computer system validation, not comma-separated values.
What the exercise produces
A session run this way leaves a written scope statement, a self-reported profile, and a review plan that lists the questions nobody could answer and the actions the group accepted, each with an owner and a date.
The profile describes practices as the group reports them. Nobody inspects your records, and the model cannot tell a supported answer from a hopeful one. That is why most of this method happens before anyone opens the questionnaire.
How the workshop maps to the assessment journey
The assessment runs in five stages, always in this order.
| Stage | What happens | Prepared by |
|---|---|---|
| 1. Scope | You choose one team, one site or a system portfolio. An optional sector field adds context and does not affect scoring. | Steps 1 and 2 |
| 2. Seven practice areas | You answer 18 questions, one area at a time. Each question offers four practice descriptions and "I do not know yet". | Steps 3 to 5 |
| 3. Review | Every answer is listed by area. You check them against the scope and confirm the review. | Step 6 |
| 4. Contact form | You enter first name, last name, work email and company. | Deciding in advance who submits |
| 5. Profile | The profile appears on screen with a PDF download. | Steps 6 and 7 |
The seven steps
Step 1: Choose one boundary
The assessment offers three scope types: one team, one site, or a system portfolio, meaning a defined group of systems such as one platform or one process area. Pick the smallest boundary for which the people in the room can answer from records.
The tool records the scope type and an optional sector, nothing more. It has no field for a description and asks you not to enter system names, so the detail has to live in your own notes. Write it before the session.
If you cannot list the systems inside the boundary, that is already a result, and the first practice area will show it. The companion how-to on building a system inventory you can maintain starts from that position.
Record the regulatory context in the statement, and leave the determination to the people responsible for it. The questions describe practices. They do not decide which requirements apply. The boundary matters here because different scopes read different primary sources:
- FDA's final guidance on computer software assurance [S02], issued in February 2026, gives nonbinding recommendations for software used as part of medical device production or the quality management system.
- Annex 11, on computerized systems, belongs to the EU GMP guidelines for medicinal products. As of November 12, 2026, the EudraLex Volume 4 index [S04] lists it as the January 2011 revision.
- Whether 21 CFR Part 11 applies depends on which records a predicate rule requires and whether they are kept in electronic form. FDA's 2003 guidance on the scope and application of Part 11 [S06] sets out how the agency interprets that scope. The assessment is not a Part 11 applicability determination.
In our view, a scope that mixes these contexts makes the answers harder to interpret, which is one more reason to keep the boundary narrow.
Step 2: Identify respondents
Invite the people who hold the evidence. Seniority is not the criterion: a manager who has never opened a change record cannot answer for change management. The table lists, for each practice area, the role the assessment's own playbooks suggest and the records to ask for. Your organization may assign these roles differently.
| Practice area | Questions | Who usually holds the evidence | Records to sample |
|---|---|---|---|
| System inventory | Q01, Q02 | System owner and quality lead | Current inventory with named owners, a relevance rationale, a recent reconciliation record |
| Intended use | Q03, Q04 | Process owner with users and quality | Current intended-use statement and its approver, a requirements sample with acceptance expectations |
| Risk-based assurance | Q05 to Q07 | Quality and relevant technical owners | A function-level risk assessment, the assurance rationale it led to, an issue record with its disposition |
| Traceability | Q08 to Q10 | Validation or assurance lead | One trace from need to evidence, the approved evidence version, a changed item with updated links |
| Change management | Q11 to Q13 | Change owner, system owner and quality | A recent change with impact rationale and closure, a supplier release assessment, a completed lifecycle review |
| Approvals and record controls | Q14, Q15 | Record owner and quality | A record showing who approved it and why, plus its access, retention and retrieval controls |
| Evidence and review readiness | Q16 to Q18 | Quality, process owner and training owner | A sample evidence package, a responsibility matrix with competence evidence, an action record with its closure rationale |
One person enters the answers. Each area needs at least one respondent who can name a real record. Note roles on the worksheet as well as names, so that a later reassessment can tell whether the respondents changed.
Step 3: Collect a small evidence sample
Ask each owner to bring one or two recent records of the kind listed above. A sample this small cannot show that a practice is applied everywhere. Its job is narrower: it stops the group from answering from memory, or from the procedure alone. A procedure shows what should happen. A completed record shows what did.
Keep the evidence inside your organization. The assessment has no upload function. Write the internal reference of each record on the worksheet so that someone else can find it again.
Step 4: Choose supported practice descriptions
Each question offers four statements, scored 0 to 3. Read all four before choosing, then select the highest one the sampled records support across the whole scope. When the group is split between two adjacent statements, take the lower and write down which record would justify the higher. The assessment is not an examination, and nobody gains credibility by choosing the top option.
In most questions the fourth statement adds checking: reconciliation, sampling, or a review of whether the practice works. A documented practice that is applied but never checked fits the third statement at most.
Mixed practice needs a rule. When two teams, or two groups of systems inside the scope, work differently, do not answer for the stronger one and do not average, because no statement describes half of each. Either narrow the scope and assess each part separately, or keep the scope, select the statement that holds for all of it, and record the difference on the worksheet. "I do not know yet" is for missing information. It is the wrong answer for a difference you already know about. The two-team example further down shows what this does to a profile.
Step 5: Mark unknowns
Select "I do not know yet" whenever the group cannot point to a record. An unknown is not scored as zero. The profile then reports numeric coverage, which is the share of the 18 questions with a scored answer, and a possible range in place of an overall score and band. The range is arithmetic: each unknown is replaced by 0 for the lower end and by 3 for the upper end. It is not a forecast or a confidence interval.
Synthetic sample: with every answer at 2 except three unknowns in risk-based assurance, 15 of 18 answers are numeric, numeric coverage is 83.3%, and the possible range runs from 57.1 to 71.5. No overall score or band is shown. The assessment page explains how unknown answers are scored.
An unknown is also different from a skipped question. You can finish with unknowns. You cannot finish with a blank.
Three questions carry a priority review prompt: relevance decisions (Q02), function-level risk rationale (Q05) and record protection (Q15). An answer of 0 on any of them produces a priority review prompt, and an unknown produces a verification prompt. Neither changes the arithmetic.
Write each unknown on the worksheet as a question with an owner: which record would settle it, who will look, and by when.
Step 6: Review the seven areas
The review step lists every answer by area before the profile is shown. Read them against the scope statement, one area at a time. Answers drift when a respondent pictures their best-run system instead of the scope.
After the contact form, read the profile in the same order: areas first, overall number last. The seven areas carry equal weight, whether they hold two questions or three, so the average can sit well above one weak area. In a second synthetic sample, every answer is 3 except record protection (Q15) at 0. The overall score is 92.9 and the band is Measured, yet the priority review prompt for Q15 still appears. "Measured" is a label in this rubric. It is not proof that controls are effective.
The tutorial that reproduces the 54.8 worked example goes through each area result and the arithmetic, and the published methodology gives the formulas and band boundaries.
Step 7: Assign actions
The profile proposes up to three primary actions, selected by fixed rules: priority review prompts first, then unknown answers to verify, then answers of 0 or 1. Treat them as a starting agenda, not a remediation plan. For each action the group accepts, write down an owner, the record that will show the action is done, and the date the group will look again.
Set dates through your own quality and change processes. If sampling a real record turns up something urgent, it follows your established process and does not wait for the next session.
When you reassess, reuse the scope statement and check the model version printed on the PDF report. A different score can come from different respondents, a different boundary or a new model version as easily as from a change in practice.
With the scope statement written and the records on the table, start the assessment for that scope.
Mixed practices: an illustrative example of two teams
Illustrative example. Site A has two teams that own computerized systems. Team A looks after manufacturing systems and Team B looks after laboratory systems. The site quality lead plans one assessment for the whole site. During evidence sampling, four answers turn out to differ. For the arithmetic, assume both teams would select the third description (score 2) for the other 14 questions.
| Question | Team A can support | Team B can support |
|---|---|---|
| Q01, identifying systems in scope | 3: inventory reconciled against changes and checked for gaps | 1: individual lists, coverage or ownership incomplete |
| Q02, relevance decisions | 2: decisions and rationale documented | 0: no consistent relevance decision |
| Q11, change impact assessment | 2: pre-implementation impact review with owner and approval | 1: impact reviews inconsistent or after implementation |
| Q12, supplier and external changes | 2: an owner reviews release information before reliance | 1: release information collected, ownership inconsistent |
Run through the model, those answers give the following profiles (synthetic).
| Profile | System inventory | Change management | Overall | Priority review prompt |
|---|---|---|---|---|
| Team A assessed alone | 83.3 | 66.7 | 69.0, Defined | None |
| Team B assessed alone | 16.7 | 44.4 | 56.3, Defined | Q02, relevance decisions |
| Site answered with Team A's practices | 83.3 | 66.7 | 69.0, Defined | None |
Letting Team A stand for the site goes wrong in three ways.
First, the site profile is simply Team A's profile. It reports 69.0 for a scope that includes a team working quite differently.
Second, the priority review prompt on relevance decisions never appears. Team B has no consistent relevance decision for its systems, which is exactly the kind of answer the prompt exists to surface.
Third, the band hides the difference. Both teams are "Defined", 12.7 points apart, with different first actions. For Team A the profile suggests sampling a recent record to confirm that the described practices are applied. For Team B it starts with relevance decisions and the inventory, then change impact assessment.
There are two honest ways out.
Narrow the scope. Run the assessment twice, once per team, each with its own scope statement. Each owner gets a profile they recognize and actions they can take. In our view this is the better choice when the teams have different owners, because actions need owners.
Keep the site scope and document the difference. For each of the four questions, select the statement that holds for the whole site, which here is Team B's, and note on the worksheet that Team A's practice is stronger and which record shows it. The site profile then reads 56.3 with the Q02 prompt: a fair description of the site as a whole, with the difference written where the next reviewer will find it.
The one-page workshop worksheet
Copy the two tables into whatever your team uses for controlled notes. The first is the header. The second holds one line per question. All entries shown are synthetic. The worksheet is a facilitation aid: it is not an exhaustive regulatory checklist, and completing it shows nothing about compliance.
| Worksheet field | What to record | Synthetic entry |
|---|---|---|
| Scope statement | The boundary in one or two sentences, including what is excluded | Laboratory systems team, Site A, as practiced on 2026-11-12. Covers the systems on inventory list INV-LAB. Excludes manufacturing and enterprise systems. |
| Model version | The model version printed on the PDF report | csa-maturity-2.0.0 |
| Respondent roles | The role of each person who answered, and for which areas | System owner (inventory, change); laboratory process owner (intended use); validation lead (risk, traceability); quality lead (approvals, readiness) |
| Evidence references, kept internally | Internal ID and location of each record sampled. Nothing is uploaded to the assessment | INV-LAB rev. 6; change record CR-0412; approval sample AP-0077 |
| Differences within scope | Any question where practice differs by team or system, and which statement was selected | None recorded |
| Unresolved questions | Each unknown: question ID, the record that would settle it, who will check | Q05: is there a function-level risk rationale for the result-calculation functions? Validation lead to check |
| Action owners | Each accepted action, its owner and the record that will show it is done | Record relevance decisions for the systems that have none. Owner: system owner. Evidence: approved rationale for each system |
| Review dates | When unknowns are checked, when actions are reviewed, when to reassess | Unknowns: 2026-11-26. Actions: 2027-01-14. Reassessment: after the action review, same scope |
| Question | Selected answer | Evidence reference | Open point | Owner and date |
|---|---|---|---|---|
| Q01 | 1 | INV-LAB rev. 6 | Two instruments are missing from the list | System owner, 2026-11-26 |
| Q02 | 0 | None found | No consistent relevance decision for the systems in scope | Quality lead, 2026-12-10 |
| Q05 | Unknown | To be located | Does a function-level rationale exist? | Validation lead, 2026-11-26 |
What this method does not establish
- It does not verify evidence. The profile reflects the statements a group selected. A sample of one or two records per area reduces guessing. It is not an audit sample.
- It does not determine compliance. The assessment is not an audit, a certification, an inspection prediction or a decision about which requirements apply to your systems.
- It does not compare you with anyone. No benchmark population exists for this model, so a score is not a percentile or an industry position.
- Bands are rubric labels. "Measured" does not mean that controls are effective or that evidence has been inspected.
- Equal weighting is a design choice. It keeps each practice area visible. It does not claim that all areas carry equal operational or regulatory risk.
Related reading
- GxP computerized system inventory template with three synthetic rows: read this first if step 1 showed that you cannot list the systems in scope.
- Tutorial: interpret a maturity profile area by area: use it before presenting a profile to colleagues who will ask where the number comes from.
- Why unknown answers and high scores need careful interpretation: a longer treatment of coverage, ranges and why a score is not a benchmark.
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.
- 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.
- S06Part 11, Electronic Records; Electronic Signatures - Scope and Application. U.S. Food and Drug Administration.
Final guidance (nonbinding) from 2003; the cover is dated August 2003 and FDA's web listing shows September 2003.
