Building a Defensible Sample When the Contractor Offers a List
A third-party auditor must plan verification of a 14-story, 260,000 ft2 hotel tower. The approved QA plan identifies 1,850 discrete field joints: membrane laps at sheathing edges, sill and jamb terminations at 640 windows, and sealant terminations at 310 penetrations. The client budget allows one auditor 2.5 days on site at roughly 8 minutes per location, including access, photo, and log entry, plus a 15% reserve. The air barrier installer, told of the audit, submits a marked-up drawing highlighting 60 joints it recommends for review, noting they were chosen from areas the crew flagged as difficult. The specification states zero tolerance for unrepaired lap voids and allows no more than 2% nonconforming sealant terminations. The superintendent says the exterior will be concealed behind cladding in five working days.
1. Construct a sampling plan with defensible numbers, stratification, and selection method.
Read the worked answer
Start with the time budget: 2.5 days at 8 hours is 20 hours, or 1,200 minutes; at 8 minutes per location that is 150 possible stops. Apply the 15% reserve to cover access failures, photos, and re-checks, leaving about 125 usable inspections. A sample of roughly 120 is therefore affordable and should be used rather than an arbitrary round number like 20.
Stratify proportionally by joint class: about 58 laps (900 of 1,850), 42 window terminations, and 20 penetrations. Selection must be random and reproducible: number the joints from the as-built drawings, assign a recorded random seed, and draw 120 primary plus 30 reserve locations in the same random order for substitutions. Force coverage across elevations and installer initials so no single crew dominates. Pre-log the plan with the client before the site visit so the sample cannot be renegotiated after results are known.
2. At the end of the visit, 3 of 50 inspected locations are nonconforming (2 lap voids, 1 inadequate sealant termination). What is the correct disposition?
Read the worked answer
Treat the two classes separately, because the acceptance criteria differ. Lap voids carry zero tolerance, so even one confirmed void fails that population on its own; no statistical argument about an overall percentage can offset a critical defect. Two voids in 33 inspected laps also indicates the defect is not an isolated event, so the lap population cannot be accepted on sampling evidence alone.
For sealant terminations, 1 defect in about 17 inspected gives an observed rate near 6%, and the upper confidence bound at 90% is well above the 2% allowance, so that population also fails the acceptance test. The correct disposition is rejection with a written nonconformance report, expansion of the sample toward the full 120-location allowance, and a contractor corrective action plan requiring 100% review of accessible lap joints before concealment. Note also that the sample is smaller than planned; record why, since a truncated sample weakens every conclusion drawn from it.
3. The contractor argues its 60 pre-selected joints should have been used because they are the hardest areas. How should the auditor handle this?
Read the worked answer
Decline the substitution and document the reason. A list chosen by the party being audited is a judgmental, self-selected sample; it cannot support an inference about the 1,850-joint population, and it invites both under-selection of known-good areas and selection of areas where a repair narrative already exists. It also breaks the traceability chain the report depends on.
The auditor can still use the list without compromising the sample: inspect the contractor's 60 locations as a separate, clearly labeled supplementary review, reported apart from the random results and never pooled with them. If a location from the contractor list happens to fall inside the random draw, it counts once, in the random sample. The report should state the population, the selection method, the seed, the number inspected, and the number nonconforming, so a reviewer can reproduce and challenge the conclusion.