The Excel mistakes that produce no error
We wrote 103 lessons, and beside each one the mistake people actually make. Only 12 of those mistakes make Excel show an error. The other 91 return a number that looks like an answer.
The count
Each of the 103 lessons carries a
concept.mistake field. It is one short paragraph, written by
hand, naming the specific way people get that feature wrong. Counting how many of them
mention an Excel error value is a search, not a judgement. It is the one figure here that
needs no interpretation.
103
lessons, each with a hand-written common mistake
12
name an Excel error value
91
name none
When an error is named it is nearly always the same one.
- #N/A 7
- #DIV/0! 2
- #VALUE! 2
- #REF! 1
What that sounds like, in the authors' words
These are quoted from the lesson files, not written for this page. Each is the sentence in which the author said the mistake goes unflagged.
-
It looks right today.
-
The total is wrong, and there is no error message.
-
SUM skips text and gives no warning.
-
The results are clearly wrong, but there is no error.
-
So the column fills with numbers that are too small, and it looks normal.
-
So a MIN meant for the minutes column can return a distance, with no error.
-
Written loosest first, every score above the lowest band gets the lowest label, and no error says so.
-
Swapping the ranges and getting a number that looks right.
Which mistakes a machine can prove
We wrote the engine that marks these lessons. So we know which wrong formulas can be diagnosed from the text, and which cannot. The split is sharper than it looks.
Decidable by reading the formula
- a bracket is opened and never closed
- this function was given the wrong number of arguments
- the column number is past the end of the range
- that function name is not one this engine knows
Undecidable from the text alone
For these, only one method is honest. Apply the repair, and check whether the result is the answer the step is asking for. That turns a guess into a fact. A false positive becomes impossible by construction, rather than merely unlikely.
-
“that semicolon should be a colon”
A semicolon between two cells and a comma between the same two cells are the same shape, and the comma version is a real answer somewhere in this library. No amount of reading the characters separates a mistyped range from a deliberate two-cell sum. Swapping the semicolon for a colon and evaluating does.
-
“those two ranges are the wrong way round”
An earlier version guessed from the data — numbers in the range being tested, text in the range being added. Any lesson filtering on a numeric column defeats that completely, so the heuristic was deleted rather than tuned.
-
“that should be a range”
Same shape as a reference to a defined name. The repair decides it.
One we got wrong ourselves
On 12 September 2026 our own published VLOOKUP article stated that
=VLOOKUP("Brake*",B4:D7,2,FALSE) returns 12. Our own engine
returned #N/A. Excel expands wildcards in an exact VLOOKUP;
ours did not.
Digging into it found three more. The lookup family had borrowed the criteria language that
SUMIF uses. A lookup target was being read as an expression rather than a value. So
MATCH("2041",…,0) found the number 2041 instead of the text.
MATCH(">300",…,0) compared instead of matching. And
MATCH("Brake~*",…,0) never un-escaped the tilde. All four are
fixed and held by twelve tests.
It is on this page because it is the same finding as everything above it, pointed at ourselves. The formula ran, returned a value, and nothing anywhere said it was wrong.
Method, and what this is not
The count is every concept.mistake field in
content/lessons/*.json, searched for an Excel error value with
this pattern:
/#(?:N\/A|REF!|VALUE!|NAME\?|DIV\/0!|NUM!|SPILL!|NULL!|CALC!)/
The quoted sentences come from a second search. It looks for an author saying, in their own words, that nothing is flagged. That one is a filter for quotation, not a classifier. A lesson it does not match has not been judged to be anything. We tried to classify all 103 by keyword and it did not work. Some mistakes return a wrong number and some are about technique. A keyword rule cannot tell those apart. A number produced by a rule that does not work is not worth publishing. So the page reports what a mechanical search can decide, and quotes the rest.
The limits, plainly. This is one curriculum, not Excel in the wild. It describes the mistakes we chose to write lessons about, and that choice is ours. The lessons were written by one person, so the judgement behind every mistake field is one person's. And the count measures what an author wrote down, not what a user did.