Stockertim Dictionary · Technology · Artificial intelligence
Silent error
Also called silent failure, silent bug, silent data error
In short
A mistake that lets code or a formula run to the end without any warning yet produce a wrong result, such as numbers stored as text being joined instead of added.
- Beginner
- Concept
- Checked
In detail
A silent error is the opposite of a crash. The program finishes, the spreadsheet calculates and a tidy number appears, but the number is wrong. Because nothing signals a problem, the result can pass through review and into a decision. Errors that stop a program are inconvenient; errors that do not stop it are dangerous.
Many come from data types and missing values. In Python, '10' + '5' gives '105', and a column of numbers read with commas or currency symbols arrives as text. Most sums and averages skip blanks, so a total can quietly cover fewer rows than intended. An inner join drops unmatched rows, and a join on a duplicated key inflates a total, both without a message.
Others come from logic. A filter uses > where >= was meant, a date is read as day-month in one place and month-day in another, an accumulator is reset inside a loop, or a formula refers to a fixed range that new rows fall outside. AI-generated code is not immune; it is fluent and runs cleanly, which can make a silent error more convincing rather than less.
Defences rely on checks that do not depend on the code itself. Row counts before and after each step, control totals compared with the source, a known-answer test on a small data set and a sanity check of the final figure against what is plausible catch most of them. Writing those checks into the code, so that it stops loudly when a count or total is wrong, turns a silent error into a visible one.
Worked example
A charity shop in Bristol imports 200 sales from its till, and the analysis script reports takings of £1,600.
- A control total from the till report shows £2,000 for the same period.
- Twenty prices were stored as text, such as
'£20', and the script skipped them without an error: 20 × £20 = £400. - After converting the text prices, the script gives £1,600 + £400 = £2,000.
The control total revealed a £400 shortfall that the script had produced without any warning.
In practice
Analysts build controls into their work: reconciling totals to source systems, asserting expected row counts, and checking types after every import. Reviewers of AI-assisted code treat a clean run as the start of checking, not the end.
Watch out
- No error message is not evidence of a correct result.
- Check data types straight after loading; numbers read as text are the commonest cause.
Learn it properly
Official sources
- Quality assurance of code for analysis and research · Government Analysis Function
- The AQuA Book: guidance on producing quality analysis for government · HM Treasury
Facts last checked on 9 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.