Find the Maximum Number in a List

Finding the largest number is the easy half of the job

One pass down a list, keep whatever beats the current champion, and you have the answer. The arithmetic was never the hard part. What trips people up is everything wrapped around the arithmetic: a stray unit that turned a column into text, a minus sign the spreadsheet exported as a Unicode dash, a value of 1,024 read as two separate numbers, or a maximum sitting three rows below a typo nobody noticed. Those failures are quiet. You still get a number back, and the number looks fine.

So this page reports the work instead of only the result. It shows how many tokens turned into numbers, prints the ones it refused, marks the position the winner occupies, counts duplicates tied for first, and lines up the four values behind it. When the top result is wrong, the reason is usually visible somewhere in that output.

What the parser accepts

Numbers are pulled out token by token after the text is split on whitespace, commas, semicolons, pipes and line breaks. Brackets and quotation marks are stripped first, so a JSON array or a quoted CSV row drops in without editing. Each surviving token has to look like a number on its own, which keeps a header row or a stray label from becoming a silent zero.

The last three rows are the ones worth watching. A skipped token count above zero means part of your data never entered the comparison, and the maximum you are reading came from a smaller list than you think.

Position and ties change what the answer means

Two lists share a maximum of 98 and tell opposite stories. In the first, 98 appears once, at row 400 of 400, at the end of a rising trend. In the second, 98 appears eleven times scattered throughout, which is the shape of a sensor pinned against its ceiling rather than a real peak. A bare maximum hides both readings.

• Position is the index of the first occurrence, counting from 1 in the order you pasted. Sort your data first if you want a rank rather than a location.
• Occurrences counts every value tied for the top. A count above one on continuous measurements normally points at rounding, a capped instrument or duplicated rows.
• The five largest show the gap behind the winner. A top value far clear of second place is either a record or an error, and the distance is what tells you which question to ask.

Largest value against largest magnitude

Switching the ranking basis changes the answer whenever negatives are present. Ranked by value, the biggest number in -1240, 840, 92 is 840. Ranked by magnitude, the winner is -1240, because its distance from zero is greater. Neither reading is more correct than the other, they answer different questions.

Value ranking suits temperatures, prices, scores, counts and anything where the sign carries direction you care about. Magnitude ranking suits error terms, deviations from a target, latitude offsets, forces along an axis and profit or loss figures where the worst month deserves the same attention as the best. The result card names which basis produced the number so a copied report never loses that context.

Where the arithmetic stops being exact

Every value here becomes an IEEE 754 double, the same 64 bit type your browser uses everywhere else. Two limits follow from that, and both are worth knowing before you trust a result on unusual data.

• Integers above 9,007,199,254,740,991 lose precision. Long identifiers, snowflake IDs, credit card numbers and 20 digit keys get rounded to the nearest representable double, which merges distinct values and produces a maximum that never existed in your source. A warning appears above the results when the list crosses that line. Sort those as text instead.
• Decimal fractions are approximations. 0.1 and 0.2 have no exact binary form, so sums drift in the sixteenth digit. Comparisons stay reliable at the scale most data lives at, and the display rounds to twelve digits of precision so the noise stays out of sight, though a tie between two values differing at the eighteenth decimal will read as identical.

The maximum is the first place bad data shows up

Outliers and errors both live at the edges of a list, which puts them in exactly the spot this tool reports. Before treating a top value as real, three checks catch most problems.

1. Compare it against the fifth largest. A winner ten times the size of the value four places behind it is rarely a genuine reading. Missing data written as 9999, a decimal point in the wrong place and a units mismatch all produce that shape.
2. Look at the skipped list. Numbers dropped for carrying a unit or an unusual dash may include the real maximum, and their absence quietly lowers the answer.
3. Check the parsed count against the row count you expected. A column of 500 that reports 486 values means fourteen cells never made it into the comparison.

Where this tool stops

• One dimension only. Numbers are ranked in a single flat list. Finding the maximum per column, per group or per key needs a spreadsheet or a query, not this page.
• No date or time handling. A timestamp is text here, so it lands in the skipped list rather than becoming a comparable value.
• Fractions and mixed radix are not parsed. Entries such as 3/4, 0x1F and 1101b are skipped. Convert them first with the hex or binary tools.
• Around 100,000 values per run. Past that the input is truncated and a note says so. The chip grid stops drawing at 400 badges for the same reason, though the maximum and the statistics still cover everything read.
• Nothing is saved. Parsing runs in your browser and no data leaves the tab, which also means a refresh clears the list. Copy the report before closing the page.
• Locale decimal commas are unsupported. A value written as 3,14 reads as 314 with the thousands toggle on, or as 3 and 14 with it off. Replace decimal commas with points before pasting European exports.