Methodology
CSV/CSA Maturity Assessment Methodology
How 18 self-reported answers become seven area scores, one overall score and a set of review prompts. Every rule is stated so that a result can be reproduced and challenged.
Model csa-maturity-2.0.0
Question bank 2.0.0
On this page24 sections
The CSV/CSA Maturity Assessment turns 18 self-reported answers into a profile of seven practice areas. Each answer has a value from 0 to 3. An area score is 100 times the sum of its answer values divided by the highest sum that area could reach, and the overall score is the mean of the seven unrounded area scores, shown to one decimal place. An overall score exists only when all 18 answers are numeric, because "I do not know yet" is never counted as zero.
This page documents model version csa-maturity-2.0.0 (question bank 2.0.0), first published on November 12, 2026. The model covers computer system validation (CSV, which here does not mean comma-separated values) and computer software assurance (CSA) practices in life sciences, for one team, site or system portfolio at a time. The result is a self-reported profile for planning. It is not an audit, a certification, a compliance percentage, an inspection prediction or a benchmark. The rubric for all 18 questions, the band table and worked profiles are published further down this page.
Model identity and maintenance
| Item | Value |
|---|---|
| Name | CSV/CSA Maturity Assessment |
| Model version | csa-maturity-2.0.0 |
| Question bank version | 2.0.0 |
| Structure | 18 questions, 72 scored answer statements, seven practice areas |
| Maintained by | CSV/CSA Maturity Editorial Team |
| First published | November 12, 2026 |
The model is written and maintained by the CSV/CSA Maturity editorial team. The team is not a regulator, and no regulator has reviewed, approved or endorsed the model. A subject-matter reviewer is named on this page only after a review has taken place. Until then, read the question wording and the source statements below as the editorial team's own.
Every profile carries the model version and the question bank version it was computed under. The scoring code runs only the versions it supports and rejects any other version instead of converting it.
Purpose and scope
The assessment gives the people responsible for computerized systems a structured way to describe current practice and to decide what to verify first. A respondent picks one scope (one team, one site or a defined system portfolio), answers 18 questions for that scope only, and receives area scores, numeric coverage, review prompts and up to three suggested actions.
The assessment runs in a fixed order: scope, the seven areas, a review of the answers, a short contact form (first name, last name, work email and company), then the profile on screen with a PDF download. This methodology is public and needs no form.
Every result is bounded in the same ways:
- It is self-reported. The model receives the statements a respondent selects. It never receives documents, system names or records, and the assessment asks respondents not to enter them.
- It describes one scope. A profile says nothing about other teams, sites or systems. Mixing scopes within one run makes the answers unreliable.
- It decides nothing about regulation. The model does not determine which regulations or guidance apply to a system, and a score does not measure conformity with any of them. Regulatory expectations depend on your activities, systems and jurisdiction. Applicability belongs with your quality and regulatory specialists.
Scope type and an optional sector context are recorded with the profile. Neither is an input to the scores, the prompts or the suggested actions.
The seven practice areas and why they weigh the same
| Practice area | Questions | What the questions ask |
|---|---|---|
| System inventory | Q01, Q02 | Whether the systems in scope can be identified, have accountable owners and carry a documented decision about their relevance to regulated work |
| Intended use | Q03, Q04 | Whether real operational use is described clearly, and whether important needs and acceptance expectations are agreed |
| Risk-based assurance | Q05, Q06, Q07 | Whether assurance effort follows documented function-level risk reasoning, whether the chosen activities are justified, and how unexpected results are handled before release |
| Traceability | Q08, Q09, Q10 | Whether needs and risks connect to evidence, whether the exact evidence version is identifiable, and whether those links survive change |
| Change management | Q11, Q12, Q13 | Whether the impact of internal changes and supplier releases is assessed before reliance, and whether continued suitability is reviewed |
| Approvals and record controls | Q14, Q15 | Whether approval responsibility is controlled, and whether assurance records are protected and retrievable on paper, hybrid and electronic media |
| Evidence and review readiness | Q16, Q17, Q18 | Whether an evidence package can be retrieved, whether responsibilities and competence are maintained, and whether findings are followed to closure |
Three areas have two questions and four have three. If all 18 answers were pooled, a three-question area would carry half as much weight again as a two-question area, for no better reason than the number of questions written for it. The model therefore gives each area one seventh of the overall score and each question an equal share of its area.
One consequence is worth knowing before you read a profile. A question in a two-question area carries 1/14 of the overall score and a question in a three-question area carries 1/21. Moving one answer up by one level changes the overall score by about 2.4 points in a two-question area (100 / 42) and by about 1.6 points in a three-question area (100 / 63).
Equal weighting is a design decision made for transparency. It is not a finding that the seven areas matter equally to product quality, patient safety or inspection outcomes, and the model has not been validated as a predictor of any of those.
What the four answer levels and "I do not know yet" mean
Each question offers four practice statements written for that question, 72 in all, plus "I do not know yet". The statements are scored 0, 1, 2 and 3. The wording differs from question to question, but the levels follow a common pattern. The pattern below is our summary of the rubric. The statements themselves, published in full further down, are what a respondent chooses between.
| Value | General pattern | Example: Q11, change impact assessment |
|---|---|---|
| 0 | The practice is absent or cannot be relied on | "Changes can proceed without a dependable impact assessment." |
| 1 | The practice exists but is informal, inconsistent or incomplete | "Impact reviews occur inconsistently or after implementation." |
| 2 | The practice is defined, documented and applied in the scope | "Changes have an appropriate pre-implementation impact review, responsible owner and approval." |
| 3 | The practice is also checked, reviewed and improved | "The process checks whether actual outcomes match the expected impact and improves where needed." |
Each question also names the evidence to check and carries a short interpretation note. Two rules apply to every answer:
- Choose the closest statement you could support with evidence. The assessment is not an examination, and the highest statement earns no credibility if no record stands behind it.
- Technology earns nothing by itself. No statement awards points for buying software, for a product category or for automation. A spreadsheet inventory, a paper approval or a manually maintained trace can sit at the highest level when its controls are effective, and an electronic system with weak controls cannot.
"I do not know yet" is a response in its own right, offered on every question. Use it when you cannot support any of the four statements. The model records it as an explicit unknown and never converts it to a value.
The formulas
D_d = 100 * sum(x_i) / (3 * n_d) for each area d
Overall = sum(D_d for the seven areas) / 7 (equal-area mean, not a pooled mean)
Coverage = 100 * numeric_answer_count / 18
Here x_i is the value (0 to 3) of a numeric answer in area d, and n_d is the number of questions in that area (2 or 3). The arithmetic is exact, carried as fractions, and rounded only for display. The overall score is always computed from the unrounded area values, never from the rounded ones shown on screen.
In the published worked profile, a synthetic sample, the seven area scores are 66.7, 50.0, 55.6, 55.6, 44.4, 66.7 and 44.4. Their unrounded mean is 54.76190476, displayed as 54.8. Pooling the 18 answers (29 points out of a possible 54) would give 53.7, which is not this model. The worked example tutorial shows each step.
Complete, partial and incomplete runs
A single answer is in one of three states, and the model keeps them apart.
| Answer state | Meaning | Treatment |
|---|---|---|
| Numeric 0 | The lowest practice statement was selected | Scored as zero and counted in numeric coverage |
| "I do not know yet" | No statement could be supported | Counted as addressed, left out of numeric coverage, never scored |
| Missing | No response was given | The run is not finished |
Those answer states give a profile one of three states.
| Profile state | Condition | What the profile contains |
|---|---|---|
| Complete | All 18 answers are numeric | Seven area scores with bands, the overall score with its band, numeric coverage of 100.0% |
| Partial | Every question is addressed and at least one answer is unknown | Numeric coverage, a possible range, and scores only for areas without an unknown answer. No overall score and no overall band |
| Incomplete | At least one question has no response | No overall score or band. The assessment asks for a practice statement or an explicit "I do not know yet" before it presents a profile |
The model never rescales the answered questions to stand in for the open ones, and it never fills in a value on the respondent's behalf.
Numeric coverage and the possible range
Numeric coverage is the share of the 18 questions that have a numeric answer: 15 numeric answers give 83.3%. Coverage says how much of the profile is scored. It is not statistical confidence, and it is not the share of evidence anyone has verified. A profile with 100.0% coverage can rest entirely on unchecked answers.
For a partial profile the model reports the interval in which the overall score must lie. The lower bound puts 0 in place of every open answer and the upper bound puts 3, and each is aggregated with the same equal-area formula. The lower bound is rounded down and the upper bound is rounded up to one decimal place, so the displayed interval never understates the real one.
Synthetic sample: every answer is 2 except Q05 to Q07, which are unknown. Six areas score 66.7 and the risk area can lie anywhere from 0 to 100. The bounds are 400 / 7 = 57.142857, shown as 57.1, and 500 / 7 = 71.428571, shown as 71.5. No band is shown, although both ends sit inside the Defined band, because the model does not classify a result it cannot compute.
When all 18 answers are unknown, coverage is 0.0% and the possible range is 0.0 to 100.0.
Rounding and the four bands
Scores are rounded half up to one decimal place, so 54.76190476 becomes 54.8. A band is assigned from the displayed value, and each threshold belongs to the higher band: a displayed 25.0 is Developing, never Ad hoc.
- Ad hoc, 0.0 to below 25.0. The answers describe practices that are not yet consistently established in the scope.
- Developing, 25.0 to below 50.0. Some practices are established, and consistency and review need attention.
- Defined, 50.0 to below 75.0. The answers describe a documented baseline across several areas.
- Measured, 75.0 to 100.0. The answers describe practices that are reviewed and maintained.
Bands are shown for each fully answered area and for the overall score of a complete profile.
"Measured" is a name in this self-report rubric. It does not mean that anyone has inspected the evidence, that controls are effective or that an organization is compliant.
The band names belong to this rubric. They are not regulatory classes and they have no regulatory status. Regulatory texts use their own terms for other purposes. FDA's final guidance on computer software assurance [S03], issued on February 3, 2026, gives nonbinding recommendations for software used as part of medical device production or the quality management system, and it presents the risk of a software function as "high process risk" or "not high process risk", while noting that risk sits on a spectrum. Our reading is that those terms describe a manufacturer's judgment about one function. A band on this page describes self-reported practice across a scope and says nothing about the risk of any function.
Priority review prompts on Q02, Q05 and Q15
Three questions are marked as priority questions.
| Question | Practice | Prompt heading |
|---|---|---|
| Q02 | Regulated-work relevance decisions | Check how regulated-work relevance is decided |
| Q05 | Risk-linked assurance effort | Check the function-level risk rationale |
| Q15 | Record protection and trust | Check how assurance records are protected |
The rules are short:
- An answer of 0 raises a priority review prompt: "Review this practice independently of the overall score."
- An answer of "I do not know yet" raises a verification prompt: "Verify the evidence for this practice before relying on the profile."
- An answer of 1, 2 or 3 raises no prompt. A 1 still appears among the suggested improvements.
- A prompt never changes the arithmetic. There is no cap, no deduction and no hidden adjustment.
A high total and a prompt can therefore appear together. In the synthetic profile where every answer is 3 except Q15 at 0, the overall score is 92.9, Measured, and the Q15 prompt is still shown.
We marked these three questions because the other answers lean on them. A system never judged relevant to regulated work may be missing from everything that follows (Q02). Assurance effort without a function-level risk rationale cannot be shown to be proportionate (Q05). Records that cannot be trusted weaken every decision that cites them (Q15). This is an editorial design choice, not a regulatory ranking of practices.
Why a prompt is not a finding
A prompt is raised by a selected statement. Nobody has examined a record, so a prompt cannot establish that a control is missing or failing, and the respondent may simply not be the person who knows. A finding is the conclusion of someone who examined evidence against a requirement. A prompt names one question and the evidence to look at, and asks an owner to check.
How suggested actions are chosen
Suggested actions come from fixed rules, with no generated advice. The model builds one ordered list:
- Priority questions (Q02, Q05, Q15) answered 0: a priority review.
- Priority questions answered "I do not know yet": a verification request.
- Other unknown answers, in area order and then question order: an evidence-gathering step.
- Numeric answers of 0 or 1: an improvement step. The lower answer value comes first. Among equal values, the question in the lower-scoring area comes first. Remaining ties follow area order and then question ID.
- If nothing is low or unknown: a single suggestion to sample recent evidence and confirm that the described practices are applied. The model does not invent a gap.
The model then walks down the list and takes the first three entries that fall in different practice areas. An entry in an area that already has a primary action is folded into that action, because both would send the same owner to the same kind of record, and the action lists the questions it also covers. The full ordered list is kept alongside the three primary actions. In rule 4, an area that has no score because of an unknown answer is ranked by the lower bound of its possible range.
Each action states the practice, a first step, the evidence to look at, a suggested owner role and a related guide. For the 54.8 worked profile the rules select "Assess change impact before implementation" (Q11, also covering Q12), "Run a small evidence-retrieval exercise" (Q16, also covering Q18) and "Agree needs and acceptance expectations" (Q04).
Nothing about the respondent enters these rules. Company name, company size, email domain, sector and interest in any product or service cannot change a score, a prompt or the order of actions.
The order is a facilitation sequence, not a risk ranking, and a reviewer who knows the systems may change it. Completing an action does not raise an answer automatically. To change a score, answer the question again against its four statements.
What the model deliberately does not do
- It has no "not applicable" option in this version. The practices in the rubric can be met by manual, hybrid or electronic means, so the absence of a tool does not make a question inapplicable. A real "not applicable" answer would need an applicability rule, a denominator policy, a minimum coverage policy and tests, published as a new model version. Without them it would become a way to remove inconvenient questions. The scoring code rejects a "not applicable" value outright instead of guessing how to count it. If a question still does not describe work your scope performs after you have checked the scope, the least misleading response is "I do not know yet", with the reason noted in your own records. The profile is then partial, which is accurate: that practice was not assessed.
- It publishes no benchmark. A score is computed from one respondent's answers and is never compared with anyone else's. There is no industry average, percentile or peer ranking, because no research design or sample stands behind the model that could support one.
- It gives no points for technology. Controls are scored, and the medium is not. This applies to software purchases, automation, artificial intelligence and any vendor's product.
- It does not adjust totals. No score is capped, raised or hidden, and no area is weighted by risk.
- It does not read evidence. Nothing is uploaded, so nothing is verified.
- It does not decide applicability. Q02 asks how your organization establishes relevance to regulated work. It does not decide which laws apply.
Limitations
- Self-report. A profile records which statements were selected, often by one person. Two honest respondents can describe the same scope differently.
- Conventions in the arithmetic. Equal area weights are a design decision. So is the equal spacing of the answer values: the model treats the step from 0 to 1 as the same size as the step from 2 to 3, which is a convention and not a measured property of practice.
- A sample of practice. Eighteen questions cannot cover everything a quality system does. A high score in an area means the selected statements were high, and other practices in that area were not asked about.
- No outcome validation. The model has not been tested against inspection results, deviations, product quality or any other outcome, and it predicts none of them.
- Limited comparability. Two profiles can be compared only when the scope, the model version and the question bank version match. The article on unknown answers and benchmarks sets out what each kind of profile lets you say.
- No clause mapping. The practice areas are the editorial team's synthesis of validation and assurance practice. They are not a clause-by-clause mapping of any regulation or guidance, and an area score is not a statement of conformity with one.
Change log
| Version | Date | Change |
|---|---|---|
2.0.0 (model csa-maturity-2.0.0) |
2026-11-12 | First public version of the 18-question, seven-area, equal-area model. |
Version 2.0.0 is the first version published in full, with its rubric, formulas and worked profiles. It does not preserve scores from any earlier form of the assessment, and earlier results should not be compared with profiles from this model.
A change to a question, an answer statement, an area label, a weight, the treatment of unknown answers or the band logic requires a new version and a new entry in this table. When two profiles come from different versions, describe the difference between them as a change of methodology, not as a change in practice.
Source-review policy
The model is the editorial team's own rubric. Regulatory sources shape how questions, notes and articles are worded and scoped. They are not scoring inputs, and no question claims to implement a clause. The sources below were opened and checked on November 12, 2026.
- FDA, computer software assurance. The final guidance issued on February 3, 2026 is nonbinding and addresses software used as part of medical device production or the quality management system. It recommends examining the intended use of individual features, functions and operations [S03].
- EU GMP Annex 11. As of November 12, 2026, the EudraLex Volume 4 index [S04] lists Annex 11 as the January 2011 revision. The annex is part of the EU GMP guidelines for medicinal products. It says, for example, that an up-to-date listing of relevant systems and their GMP functionality [S36] should be available (clause 4.3).
- Draft revisions. The European Commission consulted on a revised Annex 11, a revised Chapter 4 and a new Annex 22 [S05] from July 7 to October 7, 2025. As of November 12, 2026 these remain consultation drafts, and no question or note treats draft wording as an expectation.
- Part 11. Q15 asks about record protection in general and is not a Part 11 or Annex 11 assessment. FDA's 2003 guidance on the scope and application of Part 11 [S06] says the agency interprets the scope of Part 11 narrowly. Whether Part 11 applies depends on which records a predicate rule requires and whether they are kept in electronic form. The assessment does not make that determination.
Our reading is that each practice area touches topics these documents also discuss, in two different regulatory contexts: medical device production and quality management system software for the FDA guidance, and GMP for medicinal products for Annex 11. That overlap does not turn a score into a measure of conformity with either document.
The policy has three parts:
- Sources are rechecked when a listed text is revised, replaced or withdrawn. The date of the last check is stated on this page.
- A source change triggers a copy review of scope notes, interpretation notes, articles and FAQs. It does not change the scoring automatically.
- A scoring change is a separate decision. It needs a new model version and an entry in the change log.
Related reading
- Interpret a CSV/CSA maturity profile: a worked example reproduces the 54.8, partial and 92.9 profiles with every arithmetic step.
- Why unknown answers and high scores need careful interpretation explains what a profile lets you say and what it does not.
- How to assess CSV/CSA maturity covers scope, respondents and evidence before you answer the first question.
Model and versions
This reference is generated from the model as it runs in the CSV/CSA Maturity Assessment. The tables, formulas and worked profiles below are produced by the same question bank and scoring function, so they always describe the version in use.
| Scoring model | csa-maturity-2.0.0 |
|---|---|
| Question bank | 2.0.0 |
| Structure | 7 practice areas, 18 questions (Q01 to Q18), four scored statements per question plus “I do not know yet” |
A profile is tied to the model and question bank it was produced with. A change to a question, an answer statement, an area or a weight creates a new version; results from different versions are a change of method, not a change in practice.
The seven practice areas
Each area has equal weight in the overall result, and each question has equal weight within its area. An area with three questions does not count more than an area with two.
| No. | Practice area | Questions | What it covers |
|---|---|---|---|
| 01 | System inventory | 2 (Q01 to Q02) | Whether the computerized systems in your chosen scope can be identified, have accountable owners, and carry a documented decision about their relevance to regulated work.Evidence to have at hand: A current system inventory, named owners, a relevance rationale and a recent reconciliation record. |
| 02 | Intended use | 2 (Q03 to Q04) | Whether each system's real operational use is described clearly, and whether important needs and acceptance expectations are agreed by the people who use and oversee it.Evidence to have at hand: A current intended-use statement, its approving owner and a requirements sample with clear acceptance expectations. |
| 03 | Risk-based assurance | 3 (Q05 to Q07) | Whether assurance effort follows documented function-level risk reasoning, whether the chosen activities are justified, and how unexpected results are handled before release.Evidence to have at hand: A function-level risk assessment, the assurance rationale it led to, and an issue record with its disposition. |
| 04 | Traceability | 3 (Q08 to Q10) | Whether a reviewer can connect important needs and risks to the assurance evidence, identify the exact evidence version relied on, and keep those links intact through change.Evidence to have at hand: One trace from a need through risk, activity and evidence, with the approved evidence version and a changed item's updated links. |
| 05 | Change management | 3 (Q11 to Q13) | Whether the impact of internal changes and supplier releases is assessed before reliance, and whether systems are reviewed for continued suitability over time.Evidence to have at hand: A recent change with its impact rationale and closure, a supplier release assessment and a completed lifecycle review. |
| 06 | Approvals and record controls | 2 (Q14 to Q15) | Whether review and approval responsibility is controlled, and whether the records used for assurance decisions are protected and retrievable, across paper, hybrid and electronic media.Evidence to have at hand: A record showing who approved it and why they were authorized, plus the access, retention and retrieval controls that apply to it. |
| 07 | Evidence and review readiness | 3 (Q16 to Q18) | Whether a coherent evidence package can be retrieved, whether responsibilities and competence are maintained, and whether findings are followed through to effective closure.Evidence to have at hand: A sample evidence package with a retrieval exercise, a responsibility matrix with competence evidence, and an action record with its closure rationale. |
Answer rubric for all 18 questions
Every question offers four practice descriptions, scored 0 to 3, and the option “I do not know yet”. Choose the closest description you can support with evidence for the scope you selected. No statement awards points for buying software, using a particular vendor or automating a process: a manual or hybrid practice can meet the highest description when its controls are effective.
System inventory2 questions
Q01Identifying systems in scope
How reliably can your team identify the computerized systems in the scope you selected?
| Score | Practice description |
|---|---|
| 0 | We cannot provide a dependable list of the systems in scope. |
| 1 | Individual teams maintain lists, but coverage or ownership is incomplete. |
| 2 | A maintained inventory identifies the systems in scope and their accountable owners. |
| 3 | The inventory is reconciled against changes and periodically checked for omissions and ownership gaps. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A current system inventory, named owners and a recent reconciliation record.
- Interpretation
- Assess the same site, team or system portfolio throughout. A spreadsheet can qualify if its controls are effective.
Q02Regulated-work relevance decisionsPriority question
How is the relevance of each system to regulated work established and kept current?
| Score | Practice description |
|---|---|
| 0 | We have not established a consistent relevance decision for the systems in scope. |
| 1 | Some decisions exist, but their rationale or review is informal. |
| 2 | Relevance decisions and their rationale are documented for the systems in scope. |
| 3 | Changes in use or configuration trigger a review of the documented relevance decision. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- Approved applicability rationale and one change that prompted reconsideration.
- Interpretation
- Do not infer relevance from the vendor name alone. This question does not decide which laws apply.
Intended use2 questions
Q03Intended-use descriptions
How clearly is the intended use of each system described?
| Score | Practice description |
|---|---|
| 0 | Its intended use has not been recorded in a dependable form. |
| 1 | Descriptions exist, but key users, activities or boundaries are unclear. |
| 2 | Intended use identifies the users, activities, boundaries and outputs relevant to the selected scope. |
| 3 | Owners review those descriptions when workflows change and check them against actual use. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A current intended-use statement and the owner who approved it.
- Interpretation
- Answer for real operational use, not only a generic supplier description.
Q04Needs and acceptance expectations
How are important functional needs and acceptance expectations agreed?
| Score | Practice description |
|---|---|
| 0 | The team has no shared record of the needs or acceptance expectations. |
| 1 | Needs are listed, but important acceptance expectations remain ambiguous. |
| 2 | Responsible users and quality stakeholders agree specific needs and acceptance expectations. |
| 3 | Changes, review feedback and observed use are used to maintain those needs and expectations. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- An approved requirements sample with clear acceptance expectations and a change history.
- Interpretation
- A specific, usable record matters more than document length or the software used to author it.
Risk-based assurance3 questions
Q05Risk-linked assurance effortPriority question
How is assurance effort connected to the risk of the functions being used?
| Score | Practice description |
|---|---|
| 0 | We do not use a documented function-level rationale to choose assurance effort. |
| 1 | A general risk label exists, but its connection to specific functions is inconsistent. |
| 2 | Function-level risk reasoning is documented and used to determine assurance effort. |
| 3 | The rationale is reviewed when use changes and sampled to check that effort remains proportionate. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A function-level risk assessment and its connection to an assurance decision.
- Interpretation
- This is a practice question, not a regulatory risk classification or a determination that a function is safe.
Q06Justified assurance activities
How does your team justify its choice of testing and other assurance activities?
| Score | Practice description |
|---|---|
| 0 | The same activity is applied without a recorded rationale for the chosen approach. |
| 1 | Some tailoring occurs, but reviewers cannot consistently reconstruct why. |
| 2 | The approach is justified for the function, risk and available evidence, with appropriate records. |
| 3 | The team reviews whether its activity choices find meaningful issues and improves the approach. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- One assurance rationale, its execution evidence and review outcome.
- Interpretation
- Neither more scripts nor fewer scripts automatically means greater maturity. Unscripted work is not undocumented work.
Q07Unexpected results and issues
How are unexpected results and unresolved issues evaluated before a release decision?
| Score | Practice description |
|---|---|
| 0 | Unexpected results are not consistently recorded or considered in the decision. |
| 1 | Issues are recorded, but impact assessment or ownership is inconsistent. |
| 2 | Issues have an owner, an assessed impact and a documented resolution or justified disposition. |
| 3 | Repeated issues are reviewed for patterns, and follow-up confirms that agreed actions were effective. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- An issue record, its disposition and evidence of follow-up where relevant.
- Interpretation
- Count a justified, approved disposition as a controlled outcome; do not equate maturity with having no reported issues.
Traceability3 questions
Q08Needs-to-evidence connections
Can a reviewer connect important needs and risks to the assurance evidence?
| Score | Practice description |
|---|---|
| 0 | The connections cannot be reliably reconstructed. |
| 1 | Some links exist, but important needs or risks are missing. |
| 2 | Important needs and risks are connected to the relevant assurance activities and evidence. |
| 3 | The connections are reviewed for omissions and kept current when the scope changes. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- One traceability sample that follows a need through risk, activity and evidence.
- Interpretation
- A separate matrix is not the only acceptable representation. Evaluate whether the relationship is usable and controlled.
Q09Evidence version identification
How reliably can the team identify the evidence version used for an assurance decision?
| Score | Practice description |
|---|---|
| 0 | The exact evidence version is often uncertain. |
| 1 | Naming conventions help, but conflicting or uncontrolled copies remain. |
| 2 | The approved evidence version and its relationship to the decision are identifiable. |
| 3 | Checks detect broken links, superseded copies or incomplete evidence packages before reliance. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A decision record and the exact approved evidence it references.
- Interpretation
- Evaluate control and retrievability, not whether the evidence lives in a particular platform.
Q10Evidence links after change
What happens to evidence relationships when a requirement, test or system changes?
| Score | Practice description |
|---|---|
| 0 | Links are not routinely reassessed after a change. |
| 1 | People update links manually when they remember or a reviewer asks. |
| 2 | The change process identifies and updates the evidence relationships that are affected. |
| 3 | Sampling or periodic checks confirm that the updated relationships remain complete and usable. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A changed item with its impact review and updated links.
- Interpretation
- Manual updating can be mature when ownership, review and evidence are reliable.
Change management3 questions
Q11Change impact assessment
How is the impact of a system or configuration change assessed before implementation?
| Score | Practice description |
|---|---|
| 0 | Changes can proceed without a dependable impact assessment. |
| 1 | Impact reviews occur inconsistently or after implementation. |
| 2 | Changes have an appropriate pre-implementation impact review, responsible owner and approval. |
| 3 | The process checks whether actual outcomes match the expected impact and improves where needed. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A recent change with impact rationale, approval and closure record.
- Interpretation
- Emergency handling may differ, but its authorized process and follow-up should still be explainable.
Q12Supplier and external changes
How are supplier releases and externally initiated changes managed?
| Score | Practice description |
|---|---|
| 0 | Supplier changes are discovered mainly after they affect users. |
| 1 | Release information is collected, but ownership and assessment are inconsistent. |
| 2 | An owner reviews relevant release information and determines required actions before reliance where practicable. |
| 3 | The team checks the effectiveness of release handling and closes missed notifications or follow-up gaps. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A supplier release notice with an internal assessment and disposition.
- Interpretation
- Supplier evidence can be useful; it does not remove the need to understand your own intended use.
Q13Ongoing suitability review
How does the team review whether systems remain suitable for their intended use over time?
| Score | Practice description |
|---|---|
| 0 | No planned review or equivalent ongoing control is in place. |
| 1 | Reviews happen mainly in response to problems or inspection preparation. |
| 2 | A defined review approach examines changes, issues and continued suitability at appropriate intervals. |
| 3 | Review findings lead to owned actions, and their effectiveness is checked at subsequent reviews. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A completed lifecycle review, open actions and evidence of follow-up.
- Interpretation
- The tool does not prescribe a universal review frequency. Set the approach through your own quality system.
Approvals and record controls2 questions
Q14Review and approval responsibility
How is responsibility for review and approval controlled for the records in scope?
| Score | Practice description |
|---|---|
| 0 | Review or approval responsibility is unclear or cannot be reliably demonstrated. |
| 1 | Named reviewers exist, but authority or completion records are inconsistent. |
| 2 | Authorized reviewers, decision meaning and approval completion are identifiable for relevant records. |
| 3 | Checks detect inappropriate approvals, missing decisions or outdated authorization and prompt correction. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A record showing who reviewed it, what the decision meant and why that person was authorized.
- Interpretation
- Apply this to paper, electronic or hybrid records. Buying an e-signature product does not itself earn points.
Q15Record protection and trustPriority question
How are the records used for assurance decisions protected and kept trustworthy?
| Score | Practice description |
|---|---|
| 0 | The team cannot explain dependable controls against loss, alteration or inappropriate access. |
| 1 | Some controls exist, but ownership, retention or recovery is uncertain. |
| 2 | Controls appropriate to the record medium address access, changes, retention and dependable retrieval. |
| 3 | The team checks control effectiveness, investigates exceptions and tests recovery where applicable. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- Applicable access and record controls, retention responsibility and a retrieval or recovery check.
- Interpretation
- This broad control question is not an Annex 11, Part 11, cybersecurity or disaster-recovery audit.
Evidence and review readiness3 questions
Q16Evidence package retrieval
How easily can the team retrieve a coherent evidence package for a system in scope?
| Score | Practice description |
|---|---|
| 0 | The package cannot be assembled without substantial uncertainty. |
| 1 | Key people can find much of it, but gaps or dependencies remain. |
| 2 | A responsible reviewer can retrieve the current evidence and explain its structure. |
| 3 | Periodic retrieval exercises check completeness, accessibility and independence from one individual. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A sample evidence package and a documented retrieval exercise.
- Interpretation
- Do not upload confidential evidence into this public assessment. Use internal records to inform your answer.
Q17Responsibilities and competence
How are validation responsibilities and the competence to perform them maintained?
| Score | Practice description |
|---|---|
| 0 | Responsibilities or required competence are not clearly established. |
| 1 | Roles and training exist, but coverage or effectiveness is inconsistent. |
| 2 | Responsibilities are assigned and relevant competence or training is documented. |
| 3 | The team checks effectiveness, covers role changes and addresses capability gaps. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- A responsibility matrix and appropriate competence or training evidence.
- Interpretation
- Attendance at a training session alone is not proof of competence for every activity.
Q18Findings follow-through
How are review findings and improvement commitments followed through?
| Score | Practice description |
|---|---|
| 0 | Findings and commitments are not managed in a dependable way. |
| 1 | Action lists exist, but ownership, due dates or closure evidence is incomplete. |
| 2 | Findings have owners, agreed actions and documented closure or an approved disposition. |
| 3 | The team reviews recurrence and checks whether completed actions improved the underlying practice. |
| Not scored | I do not know yet. Recorded as unknown: never counted as zero and never filled in. |
- Evidence to check
- An action record, closure rationale and effectiveness review where relevant.
- Interpretation
- A high score is not proof that an inspection will have no findings; this question concerns your follow-through practice.
Formulas
For an area d with nd questions, each numeric answer xi is 0, 1, 2 or 3.
Dd = 100 × (x1 + x2 + … + xn) ÷ (3 × nd)
Overall = (D1 + D2 + … + D7) ÷ 7
Coverage = 100 × numeric answers ÷ 18
The overall score is the mean of the seven area scores, not a pooled average of all 18 answers. It is calculated from unrounded area values; rounding happens only for display. Coverage counts numeric answers only. It is neither statistical confidence nor a measure of how much evidence has been verified.
Nothing outside the answers enters the calculation. Company size, sector, email domain and interest in any product do not change a score.
Complete, partial and incomplete profiles
An answer is in one of three states, and they mean different things.
| Answer state | Meaning | Effect on the calculation |
|---|---|---|
| Numeric 0 | The lowest practice description was actively selected. | Counted in the denominator and scored as zero. |
| Explicit unknown | “I do not know yet” was selected. | Counts as addressed, not as a numeric answer. Never scored as zero. Raises a verification action. |
| Missing | No response exists for the question. | The assessment is not finished. A description or “I do not know yet” has to be chosen. |
Those states decide which kind of profile is shown.
| Profile | When | What is shown |
|---|---|---|
| Complete | All 18 questions have a numeric answer. | Area scores and bands, the overall score and band, coverage of 100.0%. |
| Partial | Every question is addressed and at least one answer is unknown. | Scores for fully answered areas, numeric coverage and the possible range. No overall score and no overall band, even when both ends of the range fall inside one band. |
| Incomplete | At least one question has no response. | No profile. The missing questions are requested first. |
The possible range of a partial profile is a mathematical interval, not a prediction and not a confidence interval. For each unknown answer the lower bound substitutes 0 and the upper bound substitutes 3, and the same equal-area mean is applied. An area score is shown only when every question in that area has a numeric answer.
Rounding rules
- All arithmetic is exact. Values are rounded only when they are displayed.
- Area scores, the overall score and coverage are rounded half up to one decimal place.
- The band is assigned from the displayed value, so a displayed 25.0 is “Developing” and never carries the “Ad hoc” label.
- The bounds of a possible range are rounded outward to one decimal place: the lower bound down, the upper bound up. The displayed range therefore always contains the exact one.
- Scores, bounds and coverage are always shown with exactly one decimal place.
Band boundaries
Bands are names in this self-report rubric. They describe the practices you report, not verified control.
| Band | Displayed score | How to read it |
|---|---|---|
| Ad hoc | 0.0 to below 25.0 | Your answers describe practices that are not yet consistently established in the selected scope. Start with ownership, intended use and evidence you can verify. |
| Developing | 25.0 to below 50.0 | Some practices are established, but consistency and review need attention. Choose specific gaps rather than trying to improve every score at once. |
| Defined | 50.0 to below 75.0 | Your answers describe a documented baseline across several areas. Check whether it is applied consistently and how changes are handled. |
| Measured | 75.0 to 100.0 | Your answers describe practices that are reviewed and maintained. Verify the supporting evidence and examine any weaker area separately; the label does not establish control effectiveness. |
Priority questions
Three questions raise a separate prompt, because an average can hide a weakness in one of these practices. The prompts never change the arithmetic: a high score and a priority prompt can appear together.
- Answered 0: a priority review prompt. Review this practice independently of the overall score.
- Answered “I do not know yet”: a verification prompt. Verify the evidence for this practice before relying on the profile. Uncertainty is not reported as a failed control.
| Question | Practice | Evidence to check |
|---|---|---|
| Q02 | How is the relevance of each system to regulated work established and kept current? | Approved applicability rationale and one change that prompted reconsideration. |
| Q05 | How is assurance effort connected to the risk of the functions being used? | A function-level risk assessment and its connection to an assurance decision. |
| Q15 | How are the records used for assurance decisions protected and kept trustworthy? | Applicable access and record controls, retention responsibility and a retrieval or recovery check. |
Worked profiles
Illustrative profile. These answers were created to explain the model. They do not describe a customer, an industry average or a recommended target.
Each example below is computed on this page with the live scoring function. The unrounded column shows the area value before rounding, to four decimal places, so that every step can be repeated by hand.
Profile A: a complete profile
The public worked example. The equal-area mean is 54.76190476, displayed as 54.8. A pooled 29 / 54 would give 53.7, which is not this model.
| Practice area | Answers | Sum ÷ maximum | Unrounded | Displayed | Band |
|---|---|---|---|---|---|
| System inventory | 2, 2 | 4 ÷ 6 | 66.6667 | 66.7 | Defined |
| Intended use | 2, 1 | 3 ÷ 6 | 50.0000 | 50.0 | Defined |
| Risk-based assurance | 1, 2, 2 | 5 ÷ 9 | 55.5556 | 55.6 | Defined |
| Traceability | 2, 2, 1 | 5 ÷ 9 | 55.5556 | 55.6 | Defined |
| Change management | 1, 1, 2 | 4 ÷ 9 | 44.4444 | 44.4 | Developing |
| Approvals and record controls | 2, 2 | 4 ÷ 6 | 66.6667 | 66.7 | Defined |
| Evidence and review readiness | 1, 2, 1 | 4 ÷ 9 | 44.4444 | 44.4 | Developing |
Overall = (66.6667 + 50.0000 + 55.5556 + 55.5556 + 44.4444 + 66.6667 + 44.4444) ÷ 7 = 54.76190476
Displayed: 54.8, Defined
Coverage = 100 × 18 ÷ 18 = 100.0%
Pooled average, not used by this model: 100 × 29 ÷ 54 = 53.7
No review prompts: none of the priority questions is answered 0 or unknown.
Profile B: a partial profile with one unknown area
All answers 2, but Q05 to Q07 are explicitly unknown.
| Practice area | Answers | Sum ÷ maximum | Unrounded | Displayed | Band |
|---|---|---|---|---|---|
| System inventory | 2, 2 | 4 ÷ 6 | 66.6667 | 66.7 | Defined |
| Intended use | 2, 2 | 4 ÷ 6 | 66.6667 | 66.7 | Defined |
| Risk-based assurance | unknown, unknown, unknown | 0 ÷ 9 | 0.0000 to 100.0000 | 0.0 to 100.0 | Not available |
| Traceability | 2, 2, 2 | 6 ÷ 9 | 66.6667 | 66.7 | Defined |
| Change management | 2, 2, 2 | 6 ÷ 9 | 66.6667 | 66.7 | Defined |
| Approvals and record controls | 2, 2 | 4 ÷ 6 | 66.6667 | 66.7 | Defined |
| Evidence and review readiness | 2, 2, 2 | 6 ÷ 9 | 66.6667 | 66.7 | Defined |
Lower bound = (66.6667 + 66.6667 + 0.0000 + 66.6667 + 66.6667 + 66.6667 + 66.6667) ÷ 7 = 57.1429, rounded down to 57.1
Upper bound = (66.6667 + 66.6667 + 100.0000 + 66.6667 + 66.6667 + 66.6667 + 66.6667) ÷ 7 = 71.4286, rounded up to 71.5
Overall score and band: not available
Coverage = 100 × 15 ÷ 18 = 83.3%
This profile is partial: 15 of 18 questions have a numeric answer. The range is the possible mathematical interval. The lower bound counts each unknown answer as 0 and the upper bound counts it as 3; it is not a prediction, and the profile has no overall score or band.
- Q05, verification prompt: Check the function-level risk rationale. Verify the evidence for this practice before relying on the profile. Evidence to check: A function-level risk assessment and its connection to an assurance decision.
Profile C: a high average with a priority prompt
A high average that still carries a priority review prompt.
| Practice area | Answers | Sum ÷ maximum | Unrounded | Displayed | Band |
|---|---|---|---|---|---|
| System inventory | 3, 3 | 6 ÷ 6 | 100.0000 | 100.0 | Measured |
| Intended use | 3, 3 | 6 ÷ 6 | 100.0000 | 100.0 | Measured |
| Risk-based assurance | 3, 3, 3 | 9 ÷ 9 | 100.0000 | 100.0 | Measured |
| Traceability | 3, 3, 3 | 9 ÷ 9 | 100.0000 | 100.0 | Measured |
| Change management | 3, 3, 3 | 9 ÷ 9 | 100.0000 | 100.0 | Measured |
| Approvals and record controls | 3, 0 | 3 ÷ 6 | 50.0000 | 50.0 | Defined |
| Evidence and review readiness | 3, 3, 3 | 9 ÷ 9 | 100.0000 | 100.0 | Measured |
Overall = (100.0000 + 100.0000 + 100.0000 + 100.0000 + 100.0000 + 50.0000 + 100.0000) ÷ 7 = 92.85714286
Displayed: 92.9, Measured
Coverage = 100 × 18 ÷ 18 = 100.0%
- Q15, priority review prompt: Check how assurance records are protected. Review this practice independently of the overall score. Evidence to check: Applicable access and record controls, retention responsibility and a retrieval or recovery check.