All insights
    Compliance·5 min read

    What Annex 11 and 21 CFR Part 11 actually ask of a release decision

    Both frameworks are less about software features than about evidence: who decided, on what data, under which rule, and can you show it years later.

    The system is validated, the decision is signed

    EU GMP Annex 11 treats a computerised system as part of the process it supports. It expects a risk-based validation approach, a documented lifecycle, control over changes, and, importantly, that the system does not replace the responsibility of the qualified person.

    21 CFR Part 11 adds the record and signature layer: electronic records must be accurate, retrievable over their retention period, protected against unauthorised change, and any electronic signature must be unambiguously linked to its record and to a named person.

    What that means in practice

    For a temperature excursion, the compliance question is rarely 'was the software clever'. It is whether the chain of evidence holds.

    • Source data: the raw logger readings, with their calibration and time reference, preserved unchanged.
    • Rule: the stability limit applied, versioned, and traceable to the approved stability data for that product.
    • Audit trail: every change to data or rule attributable to a person and a timestamp, with a reason where required.
    • Signature: a named, authenticated approval by the person entitled to make the release decision.

    Why deterministic logic matters here

    A rule engine can be specified, tested and re-tested: the same inputs produce the same output, and a reviewer can read the rule. That maps cleanly onto validation.

    A statistical model does not offer that guarantee, which is why it belongs on the drafting and pattern-spotting side of the workflow, not on the accept-or-reject decision. Using language models to prepare a deviation write-up is a documentation aid; letting one decide disposition creates an evidence gap you will have to defend in an inspection.

    A short readiness check

    Before an integration project starts, most of the friction is visible in four questions.

    • Can you reconstruct any single batch's full time-temperature history from one place?
    • Is the stability limit used in practice the same one documented in the dossier?
    • Does the block in ERP happen before goods move, or after?
    • Can the audit trail be exported and read without the original vendor's tooling?

    Want this applied to your own lanes and systems?

    Talk to us