Sampling Instead of Certification: Pressure on Acceptance Criteria Mid-Build
Midway through a 1,200-link single-mode build in a leased colo, the general contractor finds an unlisted fire-rated demising wall. Three 48-fiber trunks must be re-routed, adding 40 m and two splice points per trunk. Under budget pressure and with a tenant move-in date looming, the owner's rep proposes replacing the specified 100% OLTS plus 100% end-face inspection with a 10% random sample, and waiving Tier 2 OTDR entirely, to recover four days. The installer's preliminary data shows 14 links between 0.15 and 0.30 dB above the loss limit; the installer says the meter 'reads conservative' and offers to re-test with a different unit. The manufacturer's 25-year warranty application requires 100% OLTS, 100% end-face inspection, and bidirectional OTDR on all links, submitted within 30 days of completion. The project holds 6 days of float on the fixed-price contract, which requires written approval before change work.
1. How do you respond to the proposal to replace full acceptance with a 10% sample?
Read the worked answer
Decline it, and put the reason in engineering terms rather than contractual ones. A 10% sample of 1,200 links detects gross, uniformly distributed defects; it does not detect systematic defects clustered in one crew, one splicer, or one batch of pigtails. The 14 marginal links are already a signal of a systematic problem - a sample of 120 links could easily miss the affected group entirely and certify a plant that fails in service.
Second, the sampling change would void the warranty pathway. The 25-year application requires 100% OLTS, full inspection records, and bidirectional OTDR within 30 days; substituting a sample means the owner inherits the entire failure risk on 1,200 links for the sake of four days. Quantify it that way: four days of schedule against the full replacement cost of the plant.
Offer instead a targeted deviation: keep 100% OLTS and inspection, run OTDR first on the 14 marginals and on any link sharing their crew or splice lot, and reduce OTDR on the remainder only if the warranty provider approves it in writing.
2. The installer claims the meter is conservative. How do you resolve the 14 marginal links?
Read the worked answer
Treat the claim as a test-method question, not an opinion. First verify the instrument: calibration date, reference-cord condition, launch condition, mandrel wrap on single-mode, and the correct test wavelength. A dirty reference connector or a missing wrap biases every result in one direction and would explain a cluster of 'marginal' links.
Second, inspect the end faces on those 14 links. Marginal loss on single-mode frequently traces to contamination, a mismatched polish (UPC mated to APC), or a damaged ferrule, all of which are visible and photographable.
Third, re-test bidirectionally with a second calibrated instrument and reconcile the two directions by the standard's method; single-direction readings can overstate or understate true loss depending on the launch. If the link still exceeds the limit after a clean, inspected, correctly referenced bidirectional test, it fails and gets re-terminated - 'conservative meter' is not a disposition. Record every step, because these 14 links will be the first thing a warranty auditor pulls.
3. Process the re-route: what is the technical and contractual path?
Read the worked answer
Technically: 40 m of additional single-mode adds roughly 0.012–0.015 dB at 0.3–0.35 dB/km, which is negligible; the two added splices at 0.1–0.3 dB each are the real budget item. Recalculate the link loss for those three paths with measured splice values and confirm margin against the optic, noting that the paths now carry two more potential failure points each. Re-test those trunks fully - re-routing does not invalidate results elsewhere, but any re-terminated or re-spliced fiber does.
Contractually: issue a written change order before work starts, per the clause. Include the added cable and splice hardware, the re-test of affected trunks, and the updated completion date. Then spend float deliberately: the re-route consumes part of the 6 days, so publish a revised float position and identify which downstream activities now have zero float. Keep testing off the critical path by re-sequencing the unaffected rows ahead of the three re-routed trunks rather than pausing the test crew to wait for them.