The short answer: FAR 46.401(a) requires a quality assurance surveillance plan to specify all work requiring surveillance and the method of surveillance. Four methods are in common use: 100 percent inspection, random sampling, periodic inspection, and customer complaint. Choosing between them is a judgment about volume, consequence, and how much surveillance your own staff can sustain for the life of the contract, including option years. The wrong choice is not usually caught at award. It is caught when nobody has the hours to do what the plan promised.
Every instance of the work is checked.
Fits: low volume, high consequence. A monthly deliverable. A quarterly report. Calibration of a small number of instruments. Anything where a single failure matters enough to justify checking everything.
Fails: the moment volume makes it impossible. A plan promising 100 percent inspection of nightly cleaning across a 400,000 square foot building is a promise the Government breaks first, and a contractor who is under-surveilled for a year has a strong argument that the standard was never really enforced.
Test before you choose it: multiply the instances per month by the minutes each check takes. If the answer exceeds what your COR can actually spend, choose something else.
A subset is drawn and inspected, and the result is treated as representative of the whole.
Fits: high volume, repetitive, reasonably uniform work. Restroom servicing across dozens of restrooms. Ticket handling across thousands of tickets.
Requires an acceptable quality level. This is the part most often missed. A sample is only defensible against a stated tolerance: you sample in order to say whether performance is inside or outside the AQL. A plan that says random sampling while the performance requirements summary leaves the quality level blank has no defensible sample size, because there is nothing for the sample to be defensible against.
State how the sample is drawn. How many, how often, and by what method of selection. A sample the contractor can predict is not a sample.
Scheduled or unannounced checks at a stated interval.
Fits: continuous services where performance is a condition rather than a countable event. Grounds appearance. Building cleanliness. Anything where you are judging a state rather than counting transactions.
State the interval and whether it is announced. Weekly, monthly, unannounced: all three change what you are measuring. An announced inspection measures whether the contractor can prepare for an inspection.
Watch for: a plan that says periodic without saying how often. That is not a method, it is an intention, and it will not settle a dispute.
Occupants or users report failures, and those reports are the surveillance record.
Fits: as a supplement to another method, and as a way of catching what scheduled inspection misses.
Weak as a primary method, because it measures what people bothered to report rather than what happened. Complaint volume tracks how easy the complaint process is, how much occupants expect, and how recently somebody reminded them the process exists. Two identical buildings can produce very different complaint counts.
If you use it, define validity. A complaint-based acceptable quality level counts valid complaints, so somebody has to decide what makes one valid, and the plan should say who and against what.
The surveillance method is one of three things that must agree with each other for a row to be enforceable. The performance standard says what good looks like. The acceptable quality level says how much shortfall is tolerated. The surveillance method says how the Government will observe it. If any one of them does not fit the other two, the row cannot be enforced.
Common mismatches worth checking before a solicitation goes out:
A method without a frequency is incomplete. Periodic inspection weekly and periodic inspection annually are different contracts. State the frequency for each objective, and state it in terms somebody can hold you to: weekly, monthly, per shift, or on each occurrence.
ArcPerform drafts the performance work statement, the performance requirements summary and the quality assurance surveillance plan together, choosing a surveillance method for each objective and stating why. It flags random sampling with no acceptable quality level, and an AI review reports where the three documents contradict each other.
100 percent inspection, random sampling, periodic inspection, and customer complaint. FAR 46.401(a) requires a quality assurance surveillance plan to specify all work requiring surveillance and the method of surveillance, and these four are the methods in common use across federal service contracts.
For low-volume, high-consequence work where a single failure matters enough to justify checking everything: a monthly deliverable, a quarterly report, calibration of a few instruments. It fails at volume, and a plan promising it at a volume nobody can cover is a promise the Government breaks first.
For high-volume, repetitive, reasonably uniform work. It requires a stated acceptable quality level, because a sample is only defensible against a tolerance. Random sampling with the quality level left blank has no defensible sample size. The plan should also state how many, how often, and how the sample is selected.
Scheduled or unannounced checks at a stated interval, suited to continuous services where you are judging a condition rather than counting events, such as building cleanliness or grounds appearance. The interval and whether it is announced both have to be stated, since an announced inspection measures whether the contractor can prepare for one.
As a supplement, yes. As a primary method it is weak, because it measures what people bothered to report rather than what happened, and complaint volume tracks how easy the complaint process is as much as it tracks performance. If you use it, the plan should define what makes a complaint valid and who decides.
Ask how many instances there are per period, what one failure costs, whether you are judging an event or a condition, and whether your own staff can sustain the surveillance through the option years. The last question eliminates most of the wrong answers and is the one most often skipped.
Yes. The standard, the quality level and the method have to fit each other for a row to be enforceable. A complaint-based quality level surveilled by periodic inspection with no complaint log, or random sampling with no tolerance to sample against, are both unenforceable and both common.