PSLE Science Reality Lab Vol No.154 | “The Lot Was Accepted” — Were All 10,000 Items Actually Tested?

PSLE-SCI-REALITY-0154

Wait, What? Ten Thousand Items, but Only Eighty Were Opened

A factory packs 10,000 identical-looking water-test strips into one production lot. A quality team chooses 80 strips at random, checks them using a stated inspection rule, finds one defective strip and accepts the lot.

The shipment label later says:

LOT ACCEPTED AFTER QUALITY INSPECTION.

A reader turns that into a much stronger claim:

“All 10,000 strips were tested and every one passed.”

That does not follow.

In acceptance sampling, a selected sample is inspected and the result is used with a pre-set decision rule to accept or reject a larger lot. The purpose is a lot-level decision under a sampling plan. It is not a hidden way of individually testing every unit.

That distinction is the scientific job of this Reality Lab.

Reality Lab habit: A decision about a whole lot can be based on evidence from a sample. Keep the sample, the decision rule and the whole lot as three different objects.

Quick Answer

“Lot accepted” does not automatically mean every item was tested.
A sampling plan specifies how items are selected, how many are checked and what result leads to acceptance or rejection.
The inspected sample provides evidence about the lot, but an accepted lot can still contain some defective items.
A good plan makes the chance of accepting different-quality lots explicit rather than pretending the decision is certain.
Random or otherwise properly designed selection matters because an easy-to-reach sample may not represent the lot.
The defect definition matters: a lot can pass one inspected characteristic without proving every possible property is perfect.
Destructive tests, cost and time are common reasons not to inspect every unit.
Acceptance sampling is a decision tool, not proof that the lot contains zero defects.

The Exact Learner Job This Article Owns

This article owns one real-world evidence-transfer problem: how a Primary 5/6 learner should interpret a quality statement that a production lot was accepted after only a sample was inspected.

It does not become a statistics textbook, teach industrial quality engineering, calculate formal operating-characteristic curves or replace the existing PSLE Science owners for sampling, representativeness, fair comparison or evidence strength. Instead, it applies those skills to a real communication object: a lot-level acceptance decision.

Original Reality Lab Case: The Bluefin Filter Lot

The following case is fictional and uses invented data.

Bluefin Works manufactures 10,000 small filter cartridges in one lot. The buyer and supplier use a simple fictional acceptance rule for one visible sealing defect:

Part of the plan	Fictional rule
Lot size	10,000 cartridges
Sample size	80 cartridges selected across the lot
Characteristic inspected	Visible seal defect
Acceptance rule	Accept if 0 or 1 sampled cartridges have the defect; reject if 2 or more do
Observed sample result	1 defective cartridge out of 80
Decision	Lot accepted under the fictional plan

What was directly observed?

Eighty selected cartridges were inspected and one showed the defined defect.

What was decided?

The lot met the stated acceptance rule.

What was not directly observed?

The remaining 9,920 cartridges were not individually inspected by this sampling step. The acceptance decision therefore cannot be rewritten as “every cartridge was checked and found perfect”.

Observed, Decided, Inferred and Overclaimed

Layer	Bluefin example
Observed	1 defined defect among 80 selected cartridges
Decision under plan	Accept the 10,000-unit lot
Reasonable inference	The sample result was compatible with the lot being acceptable under this particular plan
Unsupported upgrade	All 10,000 units were individually tested
Unsupported perfection claim	No defective unit can exist anywhere in the lot

One of the most important scientific habits is refusing to let a decision word silently change the evidence that produced it.

What an Acceptance-Sampling Plan Actually Does

The NIST/SEMATECH Engineering Statistics Handbook describes lot acceptance sampling as selecting a sample from a lot and using the sample information to decide whether to accept or reject the lot. It also makes an important point: the main purpose is the lot decision, not a perfect estimate of the lot’s exact quality.

A simple single-sample plan can be represented as two numbers:

n: how many units are inspected;
c: the acceptance number, meaning how many defined defectives may be found before the decision changes.

The learner does not need to calculate an industrial plan. The important reasoning is simpler: the decision rule must be attached to the sample it uses.

Why Not Test Every Single Item?

“Why not inspect all 10,000?” sounds sensible until the test itself is considered.

Some tests destroy the item. Imagine testing a safety fuse by firing it, testing a food package by cutting it open, or measuring breaking strength by loading a part until it fails. If every item were destroyed in testing, none would remain for use.

Even when inspection is not destructive, testing every item can be expensive or slow. Sampling is therefore not automatically a shortcut caused by carelessness. It can be a deliberate evidence strategy.

But the trade-off must remain visible: when only a sample is inspected, uncertainty remains about uninspected items.

The Selection Check: Which 80 Items?

Suppose the quality worker chooses the first 80 cartridges from the top of one carton because they are easiest to reach. Even a large sample can mislead if it systematically misses other parts of the lot.

Questions to ask include:

Were units selected randomly or by a plan intended to cover the lot?
Did the sample include different cartons, positions or production times where relevant?
Could the chosen units have come from one unusually good or bad section?
Was selection decided before the inspector saw which items looked suspicious?

This is where Reality Lab hands the generic sampling skill back to its canonical owner. The new job here is understanding what that sampling evidence means for a lot-acceptance statement.

The Defect-Definition Check: What Was the Inspection Looking For?

“Passed inspection” sounds broad. Real inspections are usually much more specific.

Bluefin’s fictional plan checks a visible seal defect. It does not automatically test:

chemical composition;
long-term durability;
filter efficiency;
hidden internal cracks;
performance after a year of storage;
every possible way the product could fail.

A lot can therefore be accepted for one specified characteristic without becoming “scientifically proven perfect” in every respect.

The Decision Rule Is Not the Same as the True Number of Defects

Imagine two lots. Both happen to produce one defective item in the 80-unit sample. Lot A truly contains very few defects. Lot B contains more, but the sample happened to miss most of them.

Under the fictional rule, both samples lead to acceptance.

This is not a paradox. A sampling decision works with probability. Different true lot qualities can sometimes produce the same sample result.

A serious sampling plan therefore considers two directions of risk:

a good lot can sometimes be rejected because the sample looks worse than the lot really is;
a poor lot can sometimes be accepted because the sample looks better than the lot really is.

Industrial quality engineering gives these risks formal names and calculations. Primary learners only need the core scientific idea: a sample-based decision has uncertainty in both directions.

“Accepted” Is a Decision Word, Not a Measurement Word

Notice the grammar.

“One defect was found in 80 inspected units” describes an observation.

“The lot was accepted” describes a decision made by comparing that observation with a rule.

“The lot contains exactly 1 defective item in 10,000” would be a claim about the entire lot. The sample does not justify that exact statement.

Strong scientific reading asks what kind of sentence each one is before treating them as interchangeable.

Worked Case 1: Zero Defects in the Sample

A company inspects 100 randomly selected units from a lot of 20,000 and finds zero visible defects. Its sampling plan accepts the lot.

Can the company say, “There are definitely zero defective units in the entire lot”?

No. Zero defects were observed in the sample. That is encouraging evidence, but uninspected units remain. The conclusion must preserve the difference between “none found in the sample” and “none exist in the lot”.

Worked Case 2: A Biased Sample

A lot contains boxes produced across an eight-hour shift. The inspector samples only boxes from the final ten minutes because they are nearest the loading bay. Every sampled unit passes.

The problem is not that the sample is small by definition. The problem is that its selection may not represent the production periods the claim is meant to cover. A large convenient sample can still be weak evidence for the whole lot.

Worked Case 3: The Destructive Strength Test

A manufacturer tests the breaking strength of selected clips by loading each one until it snaps. The tested clips cannot be sold afterward.

Why might sampling be reasonable here?

Because 100% destructive testing would destroy the entire production lot. A planned sample can provide evidence while preserving most units. The scientific question then becomes whether the selection and decision rule are suitable for the claim.

Worked Case 4: One Characteristic Passes, Another Is Untested

Eighty bottles are checked for cap leakage. The lot passes. An advertisement says, “Quality inspection proves the bottles keep drinks cold for 24 hours.”

The evidence does not match the claim. A leakage inspection can support a statement about the inspected sealing characteristic. It does not become evidence for thermal performance unless that was separately tested.

Worked Case 5: Two Different Sampling Plans

Supplier A checks 20 units and rejects the lot if it sees one defined defect. Supplier B checks 200 units and uses a different acceptance number.

Can we compare the two labels “accepted” without seeing the plans?

Not safely. The same decision word can be produced by different sample sizes, selection rules, defect definitions and acceptance criteria. The plan is part of the evidence object.

Worked Case 6: The Rejected Lot

A sample contains more defects than the acceptance rule allows, so the lot is rejected.

Does rejection prove that every item is defective?

No. Acceptance and rejection are lot-level decisions. A rejected lot can contain many good units. The decision says the sample result did not satisfy the plan’s criterion for accepting the lot as submitted.

Worked Case 7: Reinspection Changes the Story

A first sample fails. A team then keeps drawing new samples until one finally passes and reports only the passing sample.

That procedure no longer follows the original decision rule. Selectively repeating until a desired result appears can change the evidence. If reinspection is allowed, its conditions need to be defined rather than invented after an inconvenient result.

What Evidence Strengthens a Lot-Acceptance Claim?

A clearly defined lot.
A sampling method chosen before results are known.
Selection that appropriately covers the lot.
A stated sample size.
A clear definition of what counts as a defect.
A stated acceptance and rejection rule.
A test method suitable for detecting the defined defect.
Transparent treatment of retests, invalid tests or damaged samples.
Records showing the actual sample result and decision.
A conclusion that stays at the level the plan supports.

What Weakens It?

Choosing only easy-to-reach units.
Changing the sample after seeing the first result.
Leaving the sample size unstated.
Calling a lot “perfect” because it was accepted.
Using one sampled characteristic to imply every product property was tested.
Presenting “zero defects found in the sample” as “zero defects in the lot”.
Hiding failed samples or unexplained reinspections.
Using the word “inspected” without saying whether the inspection was sampled or 100%.

Tempting Reasoning That Fails

“Accepted means every item passed.” Acceptance may be based on a sample.
“If only a sample was tested, the decision is meaningless.” Too cynical. A well-designed sampling plan can be a legitimate decision method.
“Zero defects in 80 means exactly zero in 10,000.” The untested units are still unknown individually.
“One sampled defect means the lot contains exactly 125 defects because 1/80 × 10,000 = 125.” A sample fraction is not an exact inventory count.
“A bigger sample automatically fixes biased selection.” A large unrepresentative sample can still miss important parts of a lot.
“Rejected lot means every item is bad.” Rejection is a decision about the lot under the plan, not a label on every unit.

How Far Can the Conclusion Travel?

A clean statement can travel this far:

A specified sample from the lot was inspected using a stated rule, and the observed result met the criterion for accepting the lot.

Without additional evidence, it should not automatically travel to:

every unit was inspected;
every unit is defect-free;
the exact number of defective units in the lot is known;
every possible product characteristic was tested;
the product can never fail in future use.

PSLE-Style Transfer Case

A company produces 5,000 clips. It randomly selects 50 clips and tests whether each clip closes fully. The company’s pre-set rule accepts the lot if at most one sampled clip fails. One sampled clip fails, so the lot is accepted.

A student writes: “All 5,000 clips were tested and 4,999 passed.”

Explain the error.

Answer: Only 50 clips were directly tested. The one observed failure belonged to that sample. The acceptance rule allows the company to make a decision about the lot, but it does not reveal the exact condition of every untested clip or prove that exactly one clip in the full lot is defective.

What additional information would help you judge the evidence?

Useful information includes how the 50 clips were selected, whether the test could reliably detect the defined failure, whether the sampling rule was decided beforehand, and what characteristic the inspection actually covered.

Explained Practice

Practice A: A lot of 50,000 batteries is accepted after 125 are sampled. Were all 50,000 tested by that sampling step? No.

Practice B: Zero sampled units fail. Can we report “zero failures were observed in the sample”? Yes. Can we automatically report “zero failures exist anywhere in the lot”? No.

Practice C: A strength test breaks every tested item. Why might sampling be necessary? Testing every unit would destroy the entire lot.

Practice D: The first sample fails, but the company keeps sampling until one passes without a pre-set reinspection rule. Does that strengthen confidence? No. It can selectively hide an inconvenient result.

Practice E: A lot passes a visual colour inspection. Does that prove its electrical resistance is correct? No. The inspected characteristic and claimed characteristic differ.

Delayed Independent Return: S-A-D

Later, when you see “lot accepted” or “batch passed”, ask three questions:

S — Sample: What part of the lot was actually inspected, and how was it selected?
A — Acceptance rule: What result causes accept or reject?
D — Decision boundary: What does that lot-level decision establish, and what does it leave unknown about individual units?

This is not a compulsory exam template. It is a compact return path to the evidence.

Parent and Tutor Teaching Guide

Use a bag containing 100 counters, with a small unknown number marked on one side. Ask the learner to draw ten without looking. If none are marked, ask: “What did we observe?” The correct answer is “none of the ten sampled counters were marked”, not “none of the hundred counters are marked”.

Then repeat with a pre-set rule such as “accept the bag if no more than one marked counter appears in the sample”. The important discovery is that the rule creates a decision from sample evidence. It does not magically reveal every hidden counter.

Finally change the selection method. Put all marked counters near the bottom and let one learner always pick from the top. Ask whether increasing the top-only sample from ten to twenty automatically solves the bias. This separates sample size from representativeness without turning the activity into formal statistics.

Authoritative Sources

SEAB’s 2026 objectives require learners to interpret and analyse information and evaluate observations, information and methods. MOE’s Primary Science syllabus advocates healthy scepticism: questioning observations, methods, processes and data. NIST’s acceptance-sampling guidance makes the lot decision, sampling plan and uncertainty explicit.

The Quiet Return

The label says accepted.

Do not weaken that useful decision by pretending it says every item was individually proven perfect.

Ask what was sampled, what rule was applied and how far the decision can travel.

Good science keeps the sample visible even after the lot has been accepted.

Small Group Tutorials