CSV vs CSA is the question validation leads face when FDA’s software guidance lands on the desk: do we throw out our validation habits or keep them? Neither. Computer system validation (CSV) is the discipline of proving that a system does what you rely on it to do. Computer software assurance (CSA) is a method for deciding how much proof a given feature needs, and what the record of it should contain. This post sets the two side by side.
One date first. FDA issued its guidance, Computer Software Assurance for Production and Quality Management System Software, on 3 February 2026. It supersedes the 24 September 2025 version, which carried the slightly different title “Production and Quality System Software”. If your procedures cite the September 2025 file, replace the reference. The guidance is final and non-binding: it describes FDA’s current thinking, and “should” means recommended, not required.
CSV vs CSA: Who the Guidance Is Written For
The guidance covers computers and automated data processing systems used in medical device production or in a device quality management system. It is not drug guidance. It excludes device software functions, meaning software that is itself a device.
The February reissue arrived one day after the Quality Management System Regulation (QMSR) took effect on 2 February 2026. Under QMSR the device validation duty for this software sits in clauses 4.1.6, 7.5.6 and 7.6 of ISO 13485, which the guidance cites. The old 21 CFR 820.70(i) no longer exists, so do not cite it in current documents.
The guidance also draws a line around what is in scope. Software for general business operations, such as email or accounting, is outside it. So is infrastructure that is not specific to production or the quality system, such as networking, user authentication, or backup and restore. The exception is infrastructure that stores quality records. Cloud storage holding them is used directly in the quality system, and the assurance effort should focus on the functions that protect record integrity and the Part 11 requirements that apply.
CSV vs CSA at a Glance
The middle column is a common validation habit, not a regulatory text, and some teams already work the way the right column describes. Use the table to see where your own procedures sit.
| Topic | Common CSV habit | CSA guidance (3 Feb 2026) |
|---|---|---|
| Unit of analysis | The whole system gets one validation package | Each feature, function or operation gets its own intended-use call; simple single-purpose software may be assessed as a whole |
| Risk outcome | Multi-level scoring, often a matrix of severity, probability and detectability | Two outcomes: high process risk, or not high process risk |
| Testing | Fully scripted protocols for everything | Scripted testing where rigor is needed; unscripted methods (scenario, error guessing, exploratory) where the risk allows |
| Vendor work | Repeat the vendor’s tests on site | Start from vendor evaluation and validation records; add only what the risk calls for |
| Evidence form | Printed results and annotated screenshots | Digital records such as system logs and audit trails preferred over screenshots and copied results |
| Record | A fixed document set sized for the heaviest case | Seven record elements, with no more evidence than the risk needs |
In the CSV vs CSA comparison, nothing abolishes scripted testing, and nothing says to test less everywhere. The guidance asks you to put the rigor where a failure could foreseeably harm safety.
Intended Use Comes First
The guidance starts with the intended use of the software as used in production or the quality system. It is deliberately different from the intended use of the device itself. Software used directly includes tools that automate production, inspection or testing, or that maintain a quality record.
Break the software down by feature, function and operation, because one product can hold several intended uses. The guidance gives a spreadsheet example. Using its basic cells to document time and temperature readings poses low process risk, and a successful vendor assessment plus installation and configuration may be enough. Add a custom formula that calculates statistics used to monitor the curing process, and more validation by the manufacturer might be needed.
Record how you decided, feature by feature, whether it falls inside production or the quality system at all.
CSV vs CSA on Risk: High Process Risk or Not
This is the center of the method. A feature is high process risk when a failure of it could cause a quality problem, and that problem could foreseeably compromise safety. The guidance lists features that are generally high process risk, including those that:
- maintain process parameters that affect the physical properties of the product and are essential to device safety;
- measure, inspect or decide acceptability with little or no human review;
- adjust process parameters automatically, without human review;
- produce instructions for use or labeling that patients and users need for safe operation; or
- automate trending of data the manufacturer treats as essential to safety and quality.
Generally not high process risk, among FDA’s examples: managing or organizing data, automating an existing calculation, increasing process monitoring, recording data for monitoring, routing corrective and preventive actions, logging complaints, automating change control or procedure management, and alerting on exceptions in an established process.
The guidance’s own pair of examples is useful. An automated restocking feature where a qualified person still checks materials before use is not high process risk: a mix-up gets caught. The same feature when it also automates the check, with nobody verifying, is high process risk.
One practical point: the guidance presents risk as binary but lets you rate a risk “moderate” or “low” internally; anything short of high follows the not-high path. The analysis covers reasonably foreseeable failures, not only likely ones.
CSV vs CSA on Testing: Scripted vs Unscripted
Scripted testing means test cases are written down and can be run by hand or by a tool. It can be robust, with step-by-step cases, expected results, and independent review of the plan, or limited, with fewer cases and unscripted work identified alongside. Unscripted testing means the tester’s actions are not prescribed by written steps. The guidance names three kinds: scenario testing, error guessing based on known failure modes, and exploratory testing against high-level objectives with pass/fail criteria.
For high process risk features, the guidance suggests considering scripted testing or a hybrid. For the rest, it suggests unscripted methods suited to the risk. It then adds that these are not exclusive: unscripted testing may suit a high-risk function well, and automating scripted tests for a low-risk one can be the efficient choice. So the rule is “match the method to the risk”, not “unscripted is for low risk”.
Leveraging Vendor Evidence and Digital Records
Two shifts save real effort. First, vendor work counts. The guidance says a manufacturer can use the supplier’s development practices, validation work and electronic information as a starting point, then decide what more is needed. For some lower-risk features that may be all the assurance required. Evidence can include certifications such as service organization controls reports, the vendor’s development and security documentation, and its data integrity controls (audit trails, access and signature controls).
Second, evidence can be digital. The guidance recommends system logs, audit trails and other data the software already keeps, rather than paper, screenshots or copies of results the system already retains.
Other controls count too: if outputs are fully verified downstream, that can reduce the extra assurance needed. For the traceability side, see our post on the traceability matrix in computer system validation.
CSV vs CSA in the Record: Seven Elements
The guidance recommends the record for each assessed feature include:
- the intended use;
- the result of the risk-based analysis;
- a description of the testing performed;
- issues found, such as deviations, defects or failures;
- a conclusion that the software is acceptable for its intended use, with the resolution of any issues, or a risk justification for why they will not affect the intended use;
- who performed the testing or assessment, and when; and
- review and approval where appropriate, such as a signature and date from someone with signatory authority.
The record needs no more evidence than the risk requires, but enough detail to serve as a baseline for later changes. The guidance’s table scales the detail by method: a robust scripted test records a result for every case, an exploratory test a summary of objectives and what was done. The seven elements stay constant.
CSV vs CSA for a Drug Site: What You May Borrow
If you make medicines, the CSV vs CSA debate is only partly yours: the guidance is not addressed to you. FDA wrote it for device production and device quality systems. Your in-force documents are 21 CFR Part 11, EU GMP Annex 11 (January 2011) where it applies, and the GMP rules your predicate regulations set. The July 2025 draft revision of Annex 11 is a consultation draft, not adopted, and carries no official date.
You may still borrow the logic, with a documented justification. The draft Annex 11 expects, in its principle 2.3, that an alternative practice is acceptable if it is proven and documented to give the same or a higher level of control. That is a draft, so build the justification on your own quality risk management too. A sound justification covers:
- which parts of the method you adopt (for example unscripted testing and vendor evidence for lower-risk features);
- how you translate “high process risk”, which the guidance defines through device safety, into your own terms of patient safety, product quality and data integrity;
- what evidence shows the alternative gives equal control; and
- who approved it.
Do not assume the guidance relaxes Part 11 either. It notes that FDA’s enforcement discretion on Part 11 validation (11.10(a)) does not remove the device validation duties under ISO 13485, and it points to a risk-based approach for software that maintains Part 11 records. FDA’s 2003 Part 11 guidance narrows what FDA enforces but is not an exemption; the other Part 11 controls and every predicate rule still apply. See the 21 CFR Part 11 requirements for what stays enforceable.
If you are a device maker working to ISO 13485, our guide to ISO 13485 software validation covers the clause-level duties. You can read the primary text in FDA’s CSA guidance (February 2026).
If you want the records ready to fill, the 21 CFR Part 11 Toolkit includes a Computer Software Assurance Record, an Assurance Activity Selection Guide, a Functional Risk Assessment Workbook and an Alternative Practice Justification Record. It holds 105 templates (68 Word and 37 Excel) for $99. They are templates to adapt, and they do not make any system compliant by themselves.
CSV vs CSA FAQ
Is CSV vs CSA an either-or choice?
No. CSA does not replace CSV. Validation, meaning evidence that software is fit for its intended use, is still the duty. The guidance supplies a risk-based way to scale the effort and the record. It also supplements FDA’s older General Principles of Software Validation guidance and replaces that guidance’s Section 6.
Does CSA apply to a drug manufacturer?
Not directly. It is device guidance. A drug site can adopt its reasoning only through a documented justification that fits its own GMP framework, and the justification is yours to defend at inspection.
Is scripted testing still allowed?
Yes. The guidance suggests considering scripted or hybrid testing for high process risk features and says automating scripted tests can also suit lower-risk features. It changes the default, not the permission.
Do I still need a validation report?
The guidance asks for a record with the seven elements above. What you call it and how it ties to your validation plan and procedures is your decision. Your own quality system, and for drug sites the in-force rules, set the minimum.
Can I use vendor documentation instead of testing?
For some lower-risk features, vendor evaluation and installation records may be all the assurance needed, the guidance says. You still assess the vendor and keep your own record.


