Model Review, Peer Review & Independent Checking

What Is It?

Model review, peer review and independent checking are structured technical activities in which a qualified engineer other than the original analyst examines the analysis — its assumptions, inputs, methods, convergence, sensitivity, results and conclusions — to identify errors, unjustified assumptions and gaps that the original analyst may not see. Review is not a paperwork sign-off; it is a substantive technical challenge. The reviewer asks whether the model represents the right physics, whether the inputs are traceable, whether the convergence is adequate, whether the sensitivity has been explored, whether the results are physically sensible and whether the conclusion follows from the evidence. The purpose is to close the one route to error that no amount of self-checking can close: the blind spot of the analyst who built the model.

Why It Matters

The original analyst is the worst person to find their own errors. They built the model, chose the assumptions, selected the mesh and interpreted the results. They see what they intended to build, not necessarily what they actually built. A unit error, a wrong boundary condition, a missed load, an unjustified assumption — these are invisible to the analyst because they were part of the model construction. An independent reviewer, coming to the model fresh, can see these because they are not conditioned by the original intent. Studies of error detection in engineering analysis consistently show that independent review catches errors that self-checking misses. For any analysis that supports a significant decision, independent review is not optional; it is the single most effective error-detection activity available.

The original analyst is blind to their own errors — they see what they intended, not what is there. Independent review is the only check that can catch the assumptions and inputs that the analyst accepted without question during model construction.

Levels of Review

Review can be applied at different levels of independence and thoroughness, depending on the criticality of the analysis. An informal peer review is a colleague looking over the model and results — useful for catching obvious errors but limited in depth. A formal peer review is a structured review with a defined checklist, documented findings and a sign-off — more thorough and more auditable. An independent check is a separate engineer re-running key parts of the analysis independently — the highest level, used for safety-critical or certification work. The level should match the consequence of error: a screening analysis may need only an informal review, while a certification analysis may need an independent check by a reviewer from a different team or organisation.

Reviewer Independence

The value of review comes from the independence of the reviewer. A reviewer who reports to the same manager, who contributed to the model or who has a stake in the outcome is not fully independent — they share the original analyst's blind spots and incentives. True independence means the reviewer is free to challenge any aspect of the analysis without consequence, and has no involvement in the model construction. In practice, perfect independence is not always achievable, but the reviewer should be as independent as the organisation allows. For critical work, the reviewer should come from a different team, a different department or, ideally, a different organisation. The degree of independence should be stated in the review record, because a review by a non-independent reviewer carries less weight.

A reviewer who helped build the model or who reports to the same manager shares the analyst's blind spots. State the degree of independence — it determines how much weight the review carries.

What the Reviewer Should Examine

The reviewer should examine every aspect of the analysis, from the engineering question to the final conclusion. The review is not limited to checking the results — it starts with the requirements and works through to the conclusion. A results-only review misses errors in the inputs, the assumptions and the methodology that propagate silently into the results. The reviewer should ask: is the engineering question clear? Are the loads traceable and balanced? Does the geometry idealisation preserve the load path? Are the materials appropriate? Are the connections physically credible? Has convergence been demonstrated? Are the results physically sensible? What independent evidence exists? Does the conclusion follow from the evidence? Each question targets a different potential error source.

• Requirements — Is the engineering question clear?
• Loads — Are loads traceable and balanced?
• Geometry — Does the idealisation preserve the load path and stiffness?
• Materials — Are material definitions appropriate to condition, direction and temperature?
• Connections — Are joints and supports physically credible?
• Mesh / Numerics — Has convergence been demonstrated where required?
• Results — Are reactions, deformations and failure modes physically sensible?
• Sensitivity — Which assumptions most strongly affect the conclusion?
• Validation — What independent evidence exists?
• Conclusion — Does the evidence actually support the engineering decision?

Challenging the Assumptions

The most valuable activity in a review is challenging the assumptions. Every model is built on assumptions — linear elasticity, fixed boundary, isotropic material, small displacement, isothermal — and each assumption is a potential limit on the validity of the result. The original analyst made each assumption for a reason, but the reviewer should ask whether the reason still holds for the current analysis. Has the load increased beyond the linear range? Has the temperature moved into a regime where the material is no longer isotropic? Has the displacement grown large enough that geometric non-linearity matters? An assumption that was valid for a preliminary analysis may not be valid for the final analysis, and the reviewer is the person who catches this because they are not conditioned by the original decision.

• Is the linear elasticity assumption valid at the predicted stress level?
• Is the small-displacement assumption valid at the predicted deformation?
• Is the fixed-boundary assumption valid given the actual support compliance?
• Is the isothermal assumption valid given the actual thermal environment?
• Is the static assumption valid given the actual load dynamics?

Repeating Key Hand Calculations

A strong reviewer does not just read the report — they repeat key hand calculations independently. The reaction force balance, the order-of-magnitude deflection estimate, the natural frequency approximation: these can be checked in minutes and provide a completely independent reference point. If the reviewer's hand calculation agrees with the reported finite element result, the result is corroborated. If it disagrees, there is an error — in the model, in the report or in the hand calculation — that must be resolved. Repeating hand calculations is the most effective way for the reviewer to establish that the result is in the right order of magnitude, and it is a check that the original analyst cannot perform on their own model because they already know the answer.

Reviewer's independent hand check (illustrative):

  Reported FE result:  δ_FE  =  2.3 mm
  Reviewer hand calc:  δ_hand  =  P·L³ / (48·E·I)  =  2.1 mm
  Discrepancy:         |δ_FE − δ_hand| / δ_hand  =  9.5%

  Verdict: consistent within hand-calc approximation range.
  If discrepancy were 50% or 200% → investigate before sign-off.

Inspecting the Mesh and Convergence Evidence

The reviewer should inspect the mesh — not just the global view, but the mesh in the critical regions: near stress concentrations, at joints, at contact interfaces, in the regions where the result drives the conclusion. A mesh that looks adequate globally may be too coarse locally. The reviewer should also inspect the convergence evidence: the mesh convergence study, the time-step convergence study, the residual convergence. If the convergence study is absent, the reviewer should flag it. If the convergence study shows that the result is still changing significantly with refinement, the reviewer should flag that the result is not converged. A review that does not examine the mesh and convergence evidence is not a technical review — it is a reading exercise.

MISTAKE: A review that accepts the results without inspecting the mesh in the critical regions and examining the convergence evidence. A globally adequate mesh can be locally inadequate, and a result that is still changing with refinement is not converged.

Reviewing the Sensitivity and Validation Evidence

The reviewer should examine what sensitivity analysis was done and what validation evidence exists. Was the sensitivity study comprehensive — did it vary the controlling inputs over their plausible range? Or was it cursory — a single parameter varied by an arbitrary five percent? Was any test data used for validation, and if so, was the comparison pre-test or post-test? If no sensitivity and no validation exist, the reviewer should flag that the model rests on internal consistency alone and state the residual risk. The reviewer's job is not to redo the sensitivity and validation, but to assess whether what was done is adequate for the intended use and to flag what is missing.

Review scope:

  Requirements  →  Loads  →  Geometry  →  Materials  →  Connections
       ↓            ↓          ↓            ↓            ↓
     [check]      [check]    [check]      [check]      [check]
       ↓
  Mesh / Numerics  →  Results  →  Sensitivity  →  Validation  →  Conclusion
       ↓                ↓           ↓                ↓              ↓
     [check]         [check]     [check]          [check]        [check]

  A results-only review misses everything upstream.
  Review from requirements to conclusion.

Review Findings and Disposition

The review produces findings — issues identified by the reviewer that require response. Each finding should be documented with its location in the analysis, its nature (error, unjustified assumption, missing evidence, unclear presentation) and its severity (critical, major, minor). The original analyst responds to each finding: correcting an error, justifying an assumption, providing missing evidence or clarifying the presentation. The reviewer confirms that the response is adequate. Critical findings — those that could change the conclusion — must be resolved before the analysis is accepted. Minor findings — cosmetic or clarity issues — may be dispositioned as noted for future work. The review record, with findings and dispositions, becomes part of the analysis evidence chain.

VERIFICATION: Document every review finding with its location, nature and severity. Document the response and confirm it is adequate. Critical findings must be resolved before the analysis is accepted. The review record is part of the evidence chain.

The Flagship Review

For a flagship analysis — one that supports a critical decision, a certification milestone or a safety case — the review should be correspondingly rigorous. A flagship review involves an independent reviewer (or review board) with deep expertise in the relevant analysis type, a comprehensive checklist, documented findings and formal disposition. The reviewer should have the authority to require changes before the analysis is accepted. The flagship review is the final gate before the analysis enters the decision process, and its rigour should reflect the consequence of error. A perfunctory review of a critical analysis is a failure of the credibility process, no matter how well the analysis itself was executed.

A flagship analysis deserves a flagship review — independent, comprehensive, documented and authoritative. The rigour of the review should match the consequence of error in the decision it supports.

Key Takeaways

• Independent review is the single most effective error-detection activity — the original analyst is blind to their own errors
• Review from requirements to conclusion — a results-only review misses errors in inputs, assumptions and methodology
• Challenge every assumption — an assumption valid for a preliminary analysis may not hold for the final one
• Inspect the mesh in critical regions and examine the convergence evidence — not just the global mesh view
• Document findings, responses and dispositions — the review record is part of the evidence chain