READING THE RESULTS

The traffic light

How EasyQLab decides that a language — and the whole project — is Approved, Needs review or Rejected, how to read the reason it gives, and what you can do to change it.

Verdict	Colour	What it tells you
Approved	Green	No major findings. These files require no attention.
Needs review	Amber	There are major findings worth checking before delivery.
Rejected	Red	There are critical findings: this cannot be delivered as is.

Three things decide the verdict: the severity of each finding, its confidence, and the thresholds. Then the project takes the worst of its languages.

Severities

The language cards with their severity bars coloured from red (5) to green (1) — The five small bars on each card are the five severities, 5 on the left. A coloured bar means the language has at least one finding of that severity.

Every finding has a severity from 1 to 5. Each check has a default severity, listed on What it checks.

Severity	Name	What it means	Colour
1	acceptable	Optional suggestion	Green
2	low	Minor formatting detail	Blue
3	medium	Review recommended	Yellow
4	high	Severe inconsistency	Orange
5	critical	Blocks delivery · serious error	Red

Severities 4 and 5 are the “major” findings: the card says “N major”, and the header counts them as “of high severity”.

You can give any check another severity in Settings → Checks. A new severity applies to new analyses. While reviewing, you can also change the severity of one finding with edit; see Your own comments.

Confidence

The findings column with the line saying how many doubtful findings are not counted — Hidden findings are never secret: the line above the list says how many there are, and how many of them are severe.

Every finding also carries a confidence: how sure EasyQLab is that it is a real problem. Only the findings at or above the cut-off count.

Confidence	What happens to the finding
60 % or more (50 % or more for linguistic findings)	It is shown and counts towards the card, the traffic light, the annotated PDF and the CSV.
Between 40 % and the cut-off	It is marked “doubtful” and hidden. It doesn’t count and doesn’t reach the PDF or the CSV. The list says how many there are, and you can show them.
Below 40 %	It is not kept at all. This floor is fixed: below it a finding is not worth looking at.

Linguistic findings — the translation review, spelling and forbidden terms — use their own cut-off, 50 % by default, because a badly translated sentence with medium confidence is still worth a look. Everything else uses 60 %.

Seeing the doubtful findings

Open a language. If it has doubtful findings, a line above the list says “N doubtful findings · not counted”, and “· N of them severe” when some are severity 4 or 5.
Click show ignored on that line.
Look at the end of the list. They appear under a divider, “Low confidence · N doubtful · Not counted until someone confirms them”, each one marked “doubtful”.
Click Real error on any that is real. It starts counting at once, even though EasyQLab was unsure.

What you’ll see: the card, the header and the traffic light update as soon as you confirm. A confirmed finding also goes into the annotated PDF and the CSV.

The cut-off can be changed per review profile on some plans, under Sensitivity. A saved analysis keeps the sensitivity it ran with: the Summary says “Analysed at 60% sensitivity…”. An approved project never changes verdict because someone moved the setting later.

How a language gets its verdict

Settings, Traffic light: the four thresholds and the Back to the standard values button — Settings → Traffic light. These four numbers are the whole rule. They belong to the review profile, so each client can have its own.

Each language is judged on its own findings, after leaving out what you ignored, what you marked Not an error and what fell below the cut-off. EasyQLab goes down this list and stops at the first line that matches:

Order	If the language has…	Verdict and reason
1	3 or more critical findings (severity 5)	Rejected: “Rejected: N critical findings (threshold is 3)”
2	12 or more high findings (severity 4)	Rejected: “Rejected: N high findings (threshold is 12)”
3	At least one critical finding	Needs review: “Needs review: N critical findings”
4	At least one high finding	Needs review: “Needs review: N high findings”
5	8 or more medium findings (severity 3)	Needs review: “Needs review: N medium findings (threshold is 8)”
6	Anything else	Approved: “Approved: nothing critical or high”

With these defaults, findings of severity 1 and 2 never change the verdict. They are still listed, and still go into the report.

Worked examples

A project with three languages, default thresholds:

Language	Sev 5	Sev 4	Sev 3	Sev 2 and 1	Verdict
Korean	3	0	2	4	Rejected: 3 critical findings (threshold is 3)
Spanish	0	1	3	6	Needs review: 1 high finding
French	0	0	5	10	Approved: nothing critical or high

French has 15 findings and is still approved: none is major, and it has fewer than 8 medium ones. Spanish has fewer findings than French but one of them is high, so it needs review.

The project is Rejected: “Rejected: one language rejected (Korean)”.

Now you review Korean. One of the three critical findings is a false alarm, and you mark it Not an error. Korean drops to 2 critical findings and becomes “Needs review: 2 critical findings”. Its card moves to the right, next to Spanish, and the project becomes “Needs review: 2 languages need review (Spanish, Korean)”.

Changing the thresholds

Click Settings. Pick the review profile you want to change.
Open Traffic light. The section is titled “When it is rejected and when it needs review”.
Change the four numbers. Rejected at this many critical (5), Rejected at this many high (4), Needs review at this many high (4) and Needs review at this many medium (3).
Click Save changes. To go back to the defaults, click Back to the standard values.

What you’ll see: the new thresholds apply to the next analyses with that profile. Saved analyses keep the thresholds they ran with.

A stricter client? Lower Needs review at this many medium (3), for example to 4. A client who only cares about blockers? Raise it. Change one number at a time and run a project you know well to see the effect.

What your review changes

The Summary tab with each language's verdict and the Is an error, Not an error, Ignored and Pending columns — The Summary tab. The verdict column moves as you review; the Review columns show what you have decided so far.

The verdict is live: it follows your decisions while you review.

You…	Effect on the count and the verdict
Click `ignore`	The finding stops counting. The verdict may improve.
Click `Not an error`	The finding is dismissed and ignored: it stops counting.
Click `Real error` on a counted finding	Nothing changes in the count: it already counted. It is marked “confirmed”.
Click `Real error` on a doubtful finding	It starts counting. The verdict may get worse.
Change a finding’s severity with `edit`	It counts with the new severity.
Add a `QA comment`	Your comment counts with the severity you gave it.
Turn off `Show translation review (N)`	The translation review findings stop counting in every language.

The cards re-sort themselves when a verdict changes, so the worst language stays first.

How the project gets its verdict

The results header with the project verdict and the reason naming the languages that decide it — The project verdict in the header. The reason names the languages that decide it.

The project takes the worst verdict of its languages:

One language rejected → the project is Rejected, even if the others are approved.
No language rejected, at least one in review → the project Needs review.
Every language approved → the project is Approved.

The reason names the languages that decide it, for example “Needs review: 2 languages need review (French, Korean)” or “Approved: all 8 languages approved”. With a single language, the project shows that language’s own verdict and reason.

The thresholds are per language. Adding up all languages would make a big project look worse than any of its parts: eight approved languages with five medium findings each are still eight approved languages.

Incomplete projects

If a language could not be analysed, the project can’t be approved: nobody looked at that language. It shows Needs review with the reason “Incomplete: one language was not analysed” (or “N languages were not analysed”). The Summary lists each one under “Could not be analysed:” with its reason.

Run the missing languages again to complete the project. See Troubleshooting.

The severities of the AI review

The two AI checks give a fixed severity for each kind of error, so a typo never weighs the same as an omission.

AI review of the translation

Kind of error	Severity
Omission, changed figure or data, changed meaning	5
Added text, untranslated text, wrong label	4
Typo	3
Terminology, punctuation	2

AI visual review

Kind of error	Severity
Text running out of its place	5
A changed structure, such as a table; a rule, box or logo missing or added	4
Formatting applied differently; other layout issues	3
Spacing; cut or broken text	2

The severity you set for these checks in Settings works as a ceiling. Set the AI review of the translation to 3, and an omission comes out at 3 instead of 5; a punctuation finding stays at 2.