Passive Component Tolerance Notation: Three Ways to Write Plus or Minus, and a Filter That Returns 3,534 of 10,140

Walsin writes plus or minus three different ways. We counted every tolerance value in all 203 export files: 56 distinct strings across six of the eight families. A literal filter on the plus-or-minus symbol and 10 percent returns 3,534 real parts and silently hides 6,606 more.

Venatech Limited, an authorised Walsin distributor in Hong Kong, holds all 203 parametric export files published by Walsin Technology (Walsin). This post is written from the Tolerance column of the 114 files that hold parts, read as raw strings rather than parsed into numbers. We counted every distinct value and every row that carries it. Every part number below is real. The question is the one every parametric selector gets wrong at least once: if a buyer asks for 10 percent parts, what does the filter return?

How many ways does Walsin write plus or minus in its tolerance column?

Three, and all three are in the same catalogue. We read all 359,474 rows that carry a tolerance value and recorded the opening characters of each string. The Unicode plus-minus sign appears on 331,067 rows, the two-character ASCII sequence +- on 23,311, and the three-character ASCII sequence +/- on 1,262. They are different byte sequences, and a string comparison between any two fails silently.

We have not seen this counted anywhere. 56 distinct tolerance strings exist across the catalogue, and 8 of them restate a value written differently elsewhere in it. Three values — 5, 10 and 20 percent — are each written all three ways.

Which of the eight families publish a tolerance at all, and in which notation?

Six of eight, and each family uses one notation almost exclusively. The RF and antenna exports carry no Tolerance column at all, on 457 and 44 parts — 501 parts with no published tolerance of any kind. In the six families that do publish one, the column is filled on 100 percent of rows. The notation is a property of the family, not of the value.

The varistor row deserves a note. Walsin_MOV_ALL.xlsx publishes +/-10% on all 379 rows and nothing else, so the field looks like a selector and discriminates nothing. Walsin_DiscCap_ALL.xlsx is barely better with four values, and Walsin_Antenna_ALL.xlsx has no such column at all.

What does a filter on the plus-or-minus symbol actually return?

About a third of the 10 percent parts, and the shortfall differs at every value. We counted every row stating a 10 percent tolerance in any of the three notations. There are 10,140. A literal filter on the Unicode string returns 3,534 of them — 34.9 percent — and hides 5,918 MLCC and 688 varistor and disc capacitor parts.

Read the first and last rows together. A 10 percent filter loses two thirds of the real population, and almost all of what it loses is MLCC — the family a 10 percent request most often means. 213,918 of the 359,474 rows carrying a tolerance carry a value that is written more than one way somewhere in the catalogue, so this is not a corner case; it is most of the data.

Does any single Walsin file use two notations for the same value?

Yes, and one series does it at one case size. The MLCC export is the only family using two notations, 23,311 rows in +- and 38 in ±. All 38 sit in the Microwave RF and Ultra Microwave UF series, and at 0201 the Microwave RF series writes the same 0.5 picofarad tolerance both ways: +-0.5pF on 40 rows and ±0.5pF on 26.

RF03N0R3D500 and RF03N0R4D500 are adjacent capacitance steps in the same series, case size, voltage and dielectric, with the same tolerance — and 00_ALL_MLCC_parts_combined.xlsx states that tolerance two different ways. We checked all 151 rows carrying a 0.5 picofarad tolerance to be sure this was not an artefact of our read. It is not. For what the tolerance letter in the part number itself means, see tolerance, TCR and dielectric codes decoded.

What is in the tolerance column that is not a tolerance?

827 rows of something else, in four families. We read every value that does not open with one of the three notations. 3,834 rows carry something other than a symmetric plus-or-minus value, and 827 of those are not a tolerance in any form.

The 656 MLCC rows and the 22 attenuator rows are defects we established and published earlier. The 23 inductor rows holding a bare 0 in Walsin_Inductor_ALL.xlsx and the 126 chip resistor rows holding the word Jumper are the ones to watch if you sort this column numerically: the sort will succeed, and the answer will be wrong.

What should a selector filter on instead?

A normalised value, computed once, with the original string kept beside it. That is how our free cross-reference tool treats these columns, and it is why we state differences from your original rather than hiding them behind a tick. Three rules, each measured above:

1. Normalise all three notations before you compare. Otherwise a 10 percent request returns 34.9 percent of the catalogue’s 10 percent parts.
2. Never sort or cast this column numerically. 827 rows hold something that is not a tolerance, and 3,001 more hold an asymmetric band.
3. For RF and antenna parts, do not filter on tolerance at all. 501 parts publish no such column, so any filter excludes all of them.

We hold 28 Walsin datasheets, 451 pages, and the tolerance a datasheet states is the one to qualify against — ask for the sheet covering your part number and it will be in your inbox. For thin film, inductor, varistor, disc capacitor, RF and antenna we hold none, and specifications on those lines must be confirmed against Walsin’s own site. Which tolerance the word “tolerance” means on a cross result is covered in thickness and tolerance, and the structural reason none of these columns joins blind is in 10 lengths across 8 Walsin families.

What is in Hong Kong is on ready stock — 1,125 part numbers as at 23 September 2026, a dated snapshot and not a live feed. The capacitor range sits on the MLCC page.

The short version

Walsin writes plus or minus three ways and we counted every one: 331,067 rows use the Unicode symbol, 23,311 use ASCII +-, 1,262 use ASCII +/-. 56 distinct tolerance strings exist and 8 of them restate a value written differently elsewhere. A literal filter on the symbol and 10 percent returns 3,534 of 10,140 real parts, 34.9 percent; at 0.5 picofarad it returns 26 of 151. The MLCC export uses two notations itself: RF03N0R3D500 in ASCII and RF03N0R4D500 with the symbol are adjacent capacitance steps in one series at one case size. 827 rows hold something that is not a tolerance. RF and antenna publish no tolerance column on 501 parts. We will not tell you a part matches on tolerance. We will show you the string the file holds. Qualification stays with you.

Send us the tolerance line off your specification and the part numbers you filtered to, and we will tell you what that filter excluded. And a question back: when your selector returned nothing at 10 percent, did you conclude the part does not exist?

Frequently asked questions