Guide 04 / Evidence and responsibility

Challenge the output

A candidate layout is the start of a review. Establish which claims the model supports, which require physical evidence and which belong to the approval process.

Separate three different questions

Verification asks whether the model or software implements its stated specification correctly. Validation asks whether it represents the relevant real-world behaviour well enough for its intended use. Aircraft approval addresses applicable regulatory and design obligations. These activities can inform one another, but none is a substitute for the others.

A model can correctly apply an incomplete rule set. It can also reproduce a test case while failing outside the conditions represented by that case. State the intended use narrowly: early screening of arrangements is different from substantiating an installed design. The evidence burden should follow the decision and the consequences of being wrong.

Set up independent challenge before the team becomes attached to a concept. The reviewer should be able to inspect the baseline, constraint register, assumptions and rejected options. A presentation showing only the winning arrangement makes meaningful review harder and encourages confirmation bias.

Check the computation and its limits

Use known cases to check units, coordinate transformations, clearances and rule behaviour. Include edge cases and deliberately invalid configurations. Confirm that an error is visible to the user and that incomplete data cannot quietly become an unconstrained input. Preserve test results with the software and rule-set versions they cover.

For learned models, ask how training and evaluation data relate to the aircraft, seats and passenger populations in scope. Performance on one dataset may not transfer to a different configuration. Check error distributions and consequential failure modes, not just a favourable average metric. A prediction outside the supported domain should be flagged rather than presented with routine confidence.

Review the interfaces between tools. Geometry exported from one application and analysed in another can lose metadata, change units or simplify features. Verify the exchanged representation, not only the original drawing. The same principle applies to manually transcribed values in a spreadsheet or report.

Assumption note: confidence is claim-specific

A highly reliable geometric check does not make an unvalidated comfort estimate reliable. Report confidence for each material claim and keep separate evidence chains for different analyses.

Connect physical evidence to decisions

Choose validation methods according to the question. A digital reach study may identify candidates for a mock-up. A representative mock-up can support task evaluation. Installation substantiation may require additional analyses, inspections or tests under the applicable certification programme. Avoid using one favourable result as a blanket endorsement of the whole cabin.

Applicable obligations can involve occupant protection, structural loads, flammability, emergency provisions, access, systems interfaces and other aircraft-specific considerations. EASA CS-25, FAA 14 CFR Part 25, and the relevant certification basis should be interpreted by qualified specialists. SAE AS8049 and RTCA DO-160 may be relevant to particular equipment or programmes, but citing a standard does not establish that a proposed installation meets it.

Keep evacuation claims especially disciplined. A circulation animation or boarding simulation does not demonstrate compliance with emergency evacuation requirements. Assumptions about behaviour, conditions and available exits are consequential. Determine the required means of compliance through the responsible engineering and approval process rather than inferring approval from a software visualisation.

Close-up of a cabin seat mounting interface and adjacent floor finish
Seat-to-floor interface: a visible fit does not establish structural substantiation or installation approval.

Hand over a reviewable package

The handover should let another qualified team understand what was analysed without reconstructing the project from emails. Include controlled input references, candidate identifiers, model versions, search settings, assumptions, metric definitions, test evidence and known limitations. Mark obsolete outputs so that an early concept is not mistaken for the current configuration.

Use a claim-to-evidence register. For each important statement, identify the evidence, applicability, uncertainty, reviewer and next action. “No conflict found in the stated digital envelope” is a narrower and more useful claim than “installation ready”. Wording should reflect the actual coverage of the check.

Allocate unresolved items explicitly. An open action needs an owner and a closure condition, not just a colour in a dashboard. Commercial selection may proceed while some technical work remains, but the decision record must state what is conditional and which changes could reopen the selection.

Finally, manage later changes. A revised seat, monument, software version or operating assumption may invalidate part of the previous evidence. Define triggers for re-analysis and preserve the configuration relationship between the layout and its supporting checks. Validation is not a ceremonial final step; it is a controlled link between the decision and the evidence available at that time.

ClaimEvidence to retainBoundary
Geometry passes modelled checksInput revision, rules and resultsOnly represented features and conditions
Comfort target appears achievableProtocol, population and findingsTested tasks and exposure
Concept ready for engineering reviewTraceable package and open actionsNot approval to install

Decision gate: no claim travels further than its evidence. Approval responsibilities remain with the appropriate design organisations and authorities.