Why ‘Read Everything’ Beats ‘Sample Some’ in High-Stakes Review

Every senior technical lead in AEC knows the feeling: a five-hundred-page design package lands, the programme allows four days for review, and so you triage. You sample. You sign off what you can. This article makes the case that sampling is not a principled choice but a capacity constraint, and that the constraint is no longer fixed.

The economics that force sampling.

Expert review is bottlenecked by one thing: the hours of your most experienced people. Senior reviewers are scarce, expensive, and already overcommitted. As submission packages have grown, driven by consultant proliferation, BIM deliverables, and post-Grenfell evidencing requirements, the review burden has outpaced capacity at roughly the same rate as headcount has stayed flat.

The result is a structural compromise built into every team’s process. You can’t hire your way out of it. A mid-tier developer cannot staff a team of senior reviewers large enough to read every document in every submission, on every revision cycle. So the team samples, and calls it review.

What sampling misses on complex packages.

On a simple project, sampling is a reasonable risk bet. On a high-rise residential scheme, a life sciences lab, or a data centre with dense operational standards, the bet changes character entirely.

Complex packages carry correlated risk: an error in the structural engineer’s loading assumptions may not announce itself in the structural drawings at all. It surfaces when the facade consultant’s cladding specification contradicts a fire strategy assumption two hundred pages away. Sampling each document in isolation will not catch it, because the problem doesn’t live in any single document. It lives in the gap between them.

Post-Grenfell, this gap has a name and a legal consequence. Dame Judith Hackitt’s 2018 Independent Review of Building Regulations and Fire Safety concluded that the system had suffered from fragmented responsibility and insufficient oversight, precisely the conditions that sampling perpetuates. The Building Safety Act 2022 and its dutyholder regime now place personal competence liability on the individuals signing off higher-risk buildings. The signature on a sampled review is a personal liability on an incomplete read.

Cross-referencing and citation at scale.

The hardest part of full review isn’t volume. It’s coherence. A package is not a collection of independent documents. It is a web of cross-references: a performance specification cited in a design responsibility matrix, a fire strategy assumption embedded in a consultant’s scope, a planning condition threaded through five drawing revisions.

Manual cross-referencing across a large package is combinatorially expensive. A reviewer who finishes reading document A may not hold the relevant clause from document B in working memory by the time a conflict becomes visible. This is not a competence failure. It is a cognitive one. Even the best senior reviewers cannot reliably trace every cross-reference across five hundred pages.

Citation is the paired problem. When a finding is raised, an inconsistency, a gap, a missing requirement, it needs to be pinned to the source clause and the conflicting passage. Without that, it is an opinion. With it, it is evidence. Sampling makes citation harder still, because you can only cite what you’ve actually read.

Re-reading on every revision.

Revision cycles make the economics of sampling even worse. A package at RIBA Stage 3 that a team has sampled and partially reviewed does not arrive at Stage 4 as a clean slate. The consultant has revised some drawings, held others, and introduced new coordination, often without flagging which documents changed and which stayed the same.

A sampling-based team now faces a choice: re-sample the revision set (missing any carry-forward errors that happened to land outside the sample), or re-review the whole package (which is the thing the economics prevented in the first place). In practice, most teams track changes manually, comparing revision clouds where consultants have remembered to add them, and trusting cover notes where they haven’t.

Confidence that reflects actual coverage.

Here is the honest version of a sampled review sign-off: “I am confident about the thirty percent I read, and I am making a professional judgement about the rest.” That is not a criticism. It is a description of the only option available under manual review economics. But it is not the same as confidence that reflects actual coverage.

The distinction matters most at the moment of signature. Under the dutyholder regime introduced by the Building Safety Act, the person signing a Gateway 2 submission for a higher-risk building is not signing off a best-efforts assessment. They are asserting that the design has been reviewed against the requirements, fully, traceably, and with evidence. Sampled review cannot produce that assertion honestly.

An honest “not yet checked” is safer than a false all-clear. Both need to be visible. Sampling makes “not yet checked” invisible by design.

Assurance Engineering, the discipline of treating quality as a volume you raise and calibrate rather than a verdict you stamp, requires confidence that is graded and evidence-backed.

When full review becomes feasible.

Full review has always been the right answer. It has been economically impossible, until the constraint changes. The constraint is reviewer hours. Change that constraint, and the entire logic of sampling collapses. If every document in a package can be read, cross-referenced, and cited against a firm’s own requirements, on every revision, in the time it currently takes to sample, there is no residual argument for sampling on a high-stakes project.

This is precisely what Tektome is built to do. Rather than flagging a pass or fail on a subset of documents, Tektome reads the full package, traces cross-references, pins findings to source clauses, and re-runs the review on every revision cycle.

The economics that made sampling necessary are not a fixed law of the industry. They are a capacity constraint. Assurance Engineering means building the discipline to raise that confidence, and the technology to make full review the default, not the exception.