The Small Mistakes a QA Run Catches Before They Become Big Ones

 

 Introduction

A Tamil product page goes live on a Friday. The English price is 1,499 rupees. The Tamil price reads 1,949. Two digits have swapped. Nobody notices until Monday, and by then customers have seen it.

This is an illustrative scenario, not a real client case. A QA run would have flagged the mismatch in seconds. That is the plain answer to why QA run is important in translation: it finds the slips a tired eye skips.

For a founder or a content head, the stakes are easy to see. A wrong price costs revenue. A broken tag breaks a page. A misspelled headline costs trust. Each slip is small at the source and large after publishing.

The goal is catching errors before delivery, not after it. This post shows what a QA run checks, where it falls short, and a translation QA checklist you can run before any file leaves your desk.

What a QA Run Checks

Automated Checks in CAT Tools

A QA run is a set of automated checks inside your CAT tool. It compares source and target text, segment by segment. Phrase’s QA documentation lists what such checks cover.

• Missing and extra numbers

• Repeated words and extra spaces

• Trailing punctuation that differs from the source

• Spelling, forbidden terms, and missing terms

• Broken tags and formatting

• Empty targets and text copied from the source

Crowdin and XTM publish similar lists. This is CAT tools quality assurance at its simplest. The tool does the dull comparison. You do the thinking.

Untranslated text is on that list too. Which terms should stay in English is a separate decision. I covered it in which English words should never be translated.

What It Cannot Catch

A QA run check form. It cannot judge meaning. A fluent sentence with a wrong meaning passes every automated check.

Crowdin’s guide draws the same line. It treats automated checks as QA and human review as quality control. You need both.

Tools also raise false alarms. A brand name can trip the spell check, so Crowdin suggests an ignore list. Read each warning. Do not click through them.

One more limit matters for Tamil. Phrase notes that its number check works on Latin-based numbers. It does not recognize script-based numerals, such as Bengali digits. If your file uses native-script numerals, check them by eye.

Three Small Mistakes That Grow Big

Wrong Numbers

Numbers are the quietest errors. Phrase’s own example is a source reading “1 and 2” with a target reading “1 and 3.” A reader would not blink. The tool does.

Severity scales fast. Smartling’s MQM guide rates an omitted negation in a terms and conditions text as a critical error, with a weight of 25. A typo that does not affect understanding is "minor," with a weight of 1.

Misspellings and Formatting Errors

Misspellings and formatting errors look harmless in a draft. They look careless on a live page. Double spaces, stray punctuation, and repeated words all signal a file nobody checked.

Tags carry the bigger risk. The phrase warns that tag issues are a common cause of export errors, especially in Excel-based and XML files. A file can pass a read-through and still fail on export.

Fuzzy Match Reviews

A fuzzy match is a translation memory suggestion that is close to your source, not identical. Acclaro notes that five differing words, a changed tag, a new word order, or different punctuation can each trigger one. It adds that the algorithms differ between tools.

Fuzzy match reviews need slow eyes. Take this example. The stored sentence reads “The offer ends on 12 October.” The new one reads “The offer ends on 21 October.” A tool may score that as a high match. Accept it unedited, and the date is wrong.

The phrase has a QA check for this. It flags a fuzzy match confirmed without any edit. Treat that flag as a prompt to look again.

Case Study: Reviewing at Volume

I reviewed and proofread 4,974 rows of education content over three weeks. The subjects were English, chemistry, social studies, and general knowledge.

At that size, reading alone stops working. Attention drifts. A slip in row 3,800 looks like any clean row.

Three habits help at volume. Work in batches. Run checks after each batch. Keep a short list of the slips you find.

The same logic applies to a client file. Small, frequent checks cost less than one large repair. Crowdin’s guide says a typo found during translation costs a few cents. Found after release, it can cost a hundred times more.

A Translation QA Checklist Before Delivery

How Leaders Can Make QA Pay Off

Make QA a Gate

The phrase lets project managers require a QA run before a linguist can mark a job complete. They can also stop linguists from ignoring chosen warnings. That turns QA from a habit into a rule.

Read the Trend, Not Just the File

Smartling’s MQM page notes that a rise in terminology errors may mean the glossary needs updating. Repeated flags show where your process leaks.

A clean QA report needs care too. It feels reassuring. It is not proof. I wrote about that gap in why we trust confident people more than correct ones. A green tick says the checks passed. It does not say the translation is right.

Questions Leaders Ask About QA Runs

What is a QA run in translation?

It is a set of automated checks that compares source and target text for wrong numbers, extra spaces, missing terms, and broken tags.

Does a QA run replace human review?

No. It checks form, not meaning. Crowdin’s guide separates automated QA from human quality control. Use both.

Can we ignore QA warnings?

Some are false alarms, such as brand names flagged as misspellings. Read each one. Project managers can block ignoring certain checks.

How often should a small team run QA?

Before every delivery. On large files, run it after each batch as well.

Your Next Step

Open the last file you delivered. Run a full QA on it now. Count the warnings. Sort them into real errors, false alarms, and items to check by eye.

Then pick three checks from the table above and make them a rule before your next delivery.

If you ran a full QA on your last delivered file today, what do you think it would find?

Reference Links:

Link	Used for
https://support.phrase.com/hc/en-us/articles/5709703799324-Quality-Assurance-QA-TMS	External: QA checklist, manual vs instant QA, tag export warning, number check limit
https://crowdin.com/blog/translation-quality-assurance	External: QA vs quality control, ignore list, cost-of-a-fix claim
https://www.acclaro.com/resources/fuzzy-matches	External: fuzzy match triggers and tool differences
https://help.smartling.com/hc/en-us/articles/19601042280731-LQA-MQM-Schema-Templates	External: severity weights and terminology note
https://writenestmusings.blogspot.com/2026/10/which-english-words-should-never-be-translated.html	Internal: Which English words should never be translated?
https://writenestmusings.blogspot.com/2026/09/why-we-trust-confident-people-more-than-correct-ones.html	Internal: why we trust confident people more than correct ones

About the Author

P. Maria Sundara Raj is a Translator, Proofreader and Content Writer with 3+ years in content and 8+ years in translation. He works between English and Tamil, and writes WriteNest Musings on language, communication and the small choices that shape how people understand each other. Find his translation and localization work at WriteNest9.com, connect on LinkedIn, or write to raj26.musings@gmail.com.