PSLE Science Reality Lab Vol No.120 | “It Was Still Working When the 1,000-Hour Test Ended” — Is Its Lifetime Exactly 1,000 Hours?

PSLE-SCI-REALITY-0120

Wait, What? “Survived 1,000 Hours” Does Not Mean “Died at 1,000 Hours”

Six fictional lamps begin a durability test at the same time. Three fail during the test. Three are still glowing when the laboratory switches off the test rig at exactly 1,000 hours.

A summary table records:

Lamp	What happened during the test	Recorded time
A	Failed	230 h
B	Failed	610 h
C	Failed	880 h
D	Still working when test ended	1,000+ h
E	Still working when test ended	1,000+ h
F	Still working when test ended	1,000+ h

Someone then calculates the “average lifetime” by treating D, E and F as if they failed at exactly 1,000 hours.

That calculation throws away information.

We do not know the exact lifetimes of D, E and F. We know something different and still useful: each survived beyond the end of observation. Its lifetime is greater than 1,000 hours under the stated test conditions.

This kind of partly observed result is called right-censored in reliability and survival analysis. Primary learners do not need the mathematics of survival curves. They need the evidence habit: do not turn “we stopped watching” into “the event happened”.

Quick Answer

Separate items that actually failed from items that were still working when observation stopped.
For a failed item, the recorded failure time can be an observed lifetime under the test conditions.
For an item still working at the end, the test provides a lower bound: it lasted at least that long.
Do not replace “1,000+ hours” with “1,000 hours” merely because the test ended there.
Ask whether the test duration was long enough to support the product claim being made.
Check whether the test conditions resemble the conditions to which the claim is being transferred.

The Exact Learner Job This Page Owns

This Reality Lab article owns one applied communication problem: evaluating durability, reliability or lifetime claims when observation ends before every tested item has failed.

It does not become a survival-analysis or reliability-mathematics textbook. The wider mathematical owner remains Real-World Mathematics: Reliability, Failure Rates, Redundancy and Availability. A broader explanatory owner of partly observed events also exists at How Lossy Works | Censoring When the Event Is Only Partly Observed.

This page keeps the Primary 5/6 job concrete: a product test stops, some items are still working, and a public claim must preserve what is known and what remains unknown.

Original Reality Lab Case: Six Mini Fans on a Test Rack

This is an original composite case with fictional devices and constructed data.

Six identical mini fans run continuously under stated laboratory conditions. The test has funding and equipment time for 1,000 hours.

Fan	Observed outcome	What we know	What we do not know
A	Stopped at 310 h	Failure observed at 310 h	—
B	Stopped at 740 h	Failure observed at 740 h	—
C	Stopped at 960 h	Failure observed at 960 h	—
D	Running at 1,000 h	Lifetime >1,000 h in this test	Exact eventual failure time
E	Running at 1,000 h	Lifetime >1,000 h in this test	Exact eventual failure time
F	Running at 1,000 h	Lifetime >1,000 h in this test	Exact eventual failure time

A promotional graphic says, “Average fan lifetime: 835 hours.” It got that number by pretending D, E and F all failed at 1,000 hours.

The arithmetic can be performed, but the evidence assignment is wrong. Those three fans did not fail at 1,000 hours. The test simply ended.

Observed, Claimed and Inferred

Layer	Statement
Observed	Three fans failed before 1,000 hours.
Observed	Three fans were still operating at 1,000 hours.
Supported statement	Those three surviving fans lasted at least 1,000 hours under the test conditions.
Unsupported shortcut	Those three fans each had a lifetime of exactly 1,000 hours.
Unknown	The eventual failure times of the surviving fans.

A Test End Is Not a Failure Event

Scientific records should distinguish what happened to the object from what happened to the observation period.

If the device stops working, the event of interest occurred. If the laboratory stops the test while the device still works, the observation ended but the event did not occur during the observed window.

The two endings can look similar in a spreadsheet because both produce a final time entry. Their meanings are different.

What Does “Right-Censored” Mean Without the Jargon?

NIST’s reliability handbook describes right censoring as a situation in which an item has not failed by the time observation ends, so its exact failure time is unknown but is known to be later than the last observed time.

A useful learner translation is:

We know it made it this far. We do not yet know how much farther it would have gone.

That is not “missing data” in the ordinary sense. We still have information. The information is a boundary rather than an exact failure time.

The Lower-Bound Check: “At Least” Is Scientific Information

Suppose a coating has not cracked after 500 cycles when the test stops. The honest statement is not “the coating lasts 500 cycles”. It is “the tested specimen survived at least 500 cycles under these conditions”.

The words at least preserve the direction of the unknown. They tell us that the actual lifetime lies beyond the observed boundary, not somewhere below it.

This is a useful scientific habit far beyond reliability testing. When evidence gives a bound, keep the bound instead of inventing an exact number.

The Representation Check: How Charts Can Accidentally Turn Survivors Into Failures

Imagine a bar chart with six bars. Every surviving unit stops at the same height: 1,000 hours. If the chart uses the same symbol for failure and test end, a reader can easily assume all three survivors failed exactly where the bars stop.

A clearer display distinguishes:

a failure event;
a unit still operating when observation ended;
a unit removed for another reason;
the planned test horizon.

Good representation preserves event type, not just event time.

The Comparison Check: Did Two Products Have the Same Opportunity to Fail?

Suppose Product X was tested for 2,000 hours and Product Y for only 500 hours. X had several failures after 700 hours. Y had none during its shorter test.

Can we say Y is more durable because it had zero observed failures? Not from that comparison alone. Y was not observed long enough to experience the same later time window.

Exposure time is part of the comparison. “No failure observed” means little without knowing how long and under what conditions the item was actually observed.

The Endpoint Check: What Counted as Failure?

Durability claims also depend on the definition of failure. Did a lamp count as failed only when it went completely dark? Did reduced brightness count? Did a fan fail only when it stopped spinning, or when airflow dropped below a requirement? Did a coating fail at the first visible crack or only when a large area peeled?

Two tests can both report “lifetime” while using different endpoints. Reality Lab Vol No.085 owns that comparison problem. This article adds a second question: was the endpoint actually reached, or did observation stop first?

The Test-Horizon Check: Was the Study Long Enough for the Claim?

If a product claims “designed to last 10,000 hours” but the direct durability test ended at 1,000 hours with most units still operating, the test may provide useful early evidence but cannot directly observe 10,000-hour survival.

Other evidence may exist: accelerated testing, physical models, field history or longer studies. Each has its own assumptions. A 1,000-hour survival record should not silently become a 10,000-hour direct observation.

Accelerated Tests Are a Separate Evidence Job

Some durability studies increase temperature, load, cycling frequency or another stress to make failures occur sooner. That can be scientifically useful when a justified model connects the accelerated condition to normal use.

But “survived 1,000 accelerated hours” does not automatically mean “survives the same number of normal-use years”. Vol No.071 owns that transfer problem. In this article, the narrower question is what to do when units are still alive at the end of whichever observation period was used.

Alternative Explanation 1: The Surviving Units Would Have Failed Soon After the Test

Possible. If the test had continued for another hour, one could have failed at 1,001 hours. But the present evidence does not tell us that it would.

Alternative Explanation 2: The Surviving Units Would Have Lasted Much Longer

Also possible. They may have continued for thousands more hours. Again, the current test does not reveal their exact eventual failure times.

Alternative Explanation 3: Different Units Have Different Lifetimes

Manufactured products, biological specimens and materials can vary. One unit can fail early while another survives much longer. A single “lifetime” number can therefore hide a distribution of outcomes.

Alternative Explanation 4: The Test Conditions Do Not Match Ordinary Use

Continuous laboratory operation may differ from stop-start household use. Temperature, humidity, load, vibration, maintenance or user behaviour can change the stress on a product. Even perfectly interpreted lifetime data must stay connected to the conditions under which they were produced.

What Evidence Would Strengthen a Lifetime Claim?

Failure events are distinguished from items still running when observation ends.
The planned test duration is stated.
The failure criterion is defined before testing.
All tested units are accounted for.
Still-running units are marked as “greater than” the observed duration rather than assigned a false exact lifetime.
Test conditions are documented.
Longer or repeated tests show whether the early pattern persists.
Accelerated-to-normal-use claims include a justified transfer model rather than a simple clock conversion.
The public summary preserves uncertainty instead of replacing it with one neat unsupported average.

What Would Weaken It?

Surviving units are counted as failures at the test-end time.
Units removed early are silently omitted.
Products are compared over different observation lengths without adjustment or explanation.
The report never states what counted as failure.
A short test is advertised as direct proof of a much longer lifetime.
Only surviving units are shown while early failures disappear from the graphic.
Normal-use claims are made from very different laboratory conditions with no transfer evidence.

Worked Case 1: The Lamp Still On at Midnight

A lamp begins at noon and is still on when a six-hour classroom test ends. Its observed lifetime is not exactly six hours. The test establishes only that it remained operational for at least six hours under those conditions.

Worked Case 2: Two Batteries, Unequal Test Windows

Battery A is observed for ten hours and fails at hour eight. Battery B is observed for five hours and is still running when the test stops. Can we say B lasts longer than A? Not yet. B’s lifetime is known to exceed five hours, but it might fail before eight. The shorter observation window prevents that comparison.

Worked Case 3: “No Failures” in a Very Short Demonstration

A product video runs five devices for ten minutes and proudly reports zero failures. That observation can be true while giving almost no evidence about a claim of years of durability. The event was given too little opportunity to occur.

Worked Case 4: A Stronger Report

A report states: “Ten units were tested for 2,000 hours. Two failed at 1,260 and 1,740 hours. Eight remained operational when testing ended, so their exact lifetimes are greater than 2,000 hours under the stated conditions.” That wording preserves both the observed failures and the surviving evidence without inventing exact lifetimes.

Tempting Reasoning That Fails

“The bar stops at 1,000, so the product failed at 1,000.” The test may have stopped while the product continued working.
“Still working means infinite lifetime.” No. It means failure was not observed within the available window.
“No failures prove the product never fails.” The observation length and sample size still matter.
“Censored means useless or missing.” No. A survivor gives a lower-bound statement: its lifetime exceeds the observed time.
“Just use the test-end time in the average.” That treats an observation boundary as an event that did not occur.

Model and Measurement Limits

Reliability analysis has formal methods for combining observed failures with censored observations. Those methods belong beyond this Primary article. The important boundary is that simply replacing every still-running unit with the test-end time is not a neutral simplification; it changes the meaning of the data.

A small sample also limits how confidently results can generalise to a large population of products. The more ambitious the lifetime claim, the more evidence is usually required about variability, conditions and repeat production.

How Far Can the Conclusion Travel?

If a device is still working at 1,000 hours, a defensible conclusion is: this tested device survived at least 1,000 hours under these test conditions.

That does not by itself tell us its exact eventual lifetime, the average lifetime of all products, the probability that every unit will survive 1,000 hours, or how long the device will last in every real-world environment.

PSLE-Style Transfer Case

Four identical model motors are run continuously for 300 minutes. Motor A stops at 120 minutes. Motor B stops at 250 minutes. Motors C and D are still running when the investigation ends at 300 minutes.

Question: Why is it incorrect to record the lifetime of Motors C and D as exactly 300 minutes?

Reasoned answer: The investigation stopped at 300 minutes while Motors C and D were still running. Therefore their exact failure times were not observed. The evidence only shows that each lasted more than 300 minutes under the test conditions.

Explained Practice

Practice A: A coating has not cracked after 400 cycles when testing stops. What can you say? It survived at least 400 cycles under the tested conditions.

Practice B: Product X has no failures in 100 hours. Product Y has one failure at 400 hours during a 1,000-hour test. Is X automatically more reliable? No. X had a much shorter opportunity for late failures to appear.

Practice C: All ten units survive a 500-hour test. Can a report say “none failed during 500 hours”? Yes. Can it say “their exact lifetime is 500 hours”? No.

Delayed Independent Return: Four Questions for Any Lifetime Graphic

What counted as failure?
How long was each item actually observed?
Which items failed and which were still working when observation stopped?
Does the public claim preserve lower bounds, or turn them into false exact lifetimes?

Parent and Tutor Teaching Guide

Use an unfinished story. Tell the learner: “A toy motor has been running for twenty minutes. We leave the room while it is still running.” Then ask, “What is its lifetime?” The correct answer is not twenty minutes. The evidence says only that its lifetime exceeds twenty minutes.

Next, compare two statements: “failed at 20 minutes” and “still working at 20 minutes when observation ended”. Ask why the same number carries opposite information about the event. This makes the evidence boundary intuitive before introducing the technical term censoring.

For stronger transfer, draw six horizontal timelines. Put crosses where actual failures happened and arrows at the test boundary for items still operating. Ask the learner to explain which times are exact events and which are only lower bounds.

Authoritative Sources

NIST explains that right-censored reliability observations retain information that an item survived beyond a particular time even though its exact failure time was not observed. For a PSLE Science learner, this becomes a broad scientific habit: preserve the difference between an event and the end of observation, then communicate only what the evidence actually establishes.

The Quiet Return

Sometimes science ends before the object does.

That does not leave us with nothing. It leaves us with a boundary.

If the test stops while the product is still working, record how far it definitely got—then leave the rest honestly unknown.

Small Group Tutorials