wxrks' Editor can run an AI-powered quality review — the Smells engine — alongside the standard, rule-based QA checks. It reads each source→target segment, flags likely quality problems, and turns them into a Confidence Score so you can see, at a glance, how trustworthy each translation is before you spend review time on it.
💡 Who is this for? This guide is for Linguists and Reviewers who need to prioritize their review effort using AI-detected quality issues (Smells) and Confidence Scores in the wxrks Editor.
Key Concepts
Term | Meaning |
Smell | A potential quality problem the AI review found in a segment (a mistranslation, awkward wording, a terminology miss, and so on). Each Smell has a category, a severity, and a short explanation. |
Severity | How serious a Smell is: Blocker, Critical, Major, Minor, or Info. |
Confidence Score | A 0–100% value showing how confident wxrks is in a segment's quality. A segment with no Smells scores 100%; a segment with Smells is scored by its single most severe Smell. |
Overall Confidence | The plain average of every checked segment's own Confidence Score, shown at the top of the QA panel for the whole work unit. |
Issues vs. Smells | Issues are the traditional, rule-based QA warnings (tags, numbers, length, spelling, and so on) — see How to use the wxrks QA Check Tool. Smells are the newer, AI-detected findings covered in this article. Both run side by side and appear together in the same QA panel. |
How severity maps to confidence
A segment's Confidence Score is driven by its single most severe Smell — not an average of that one segment's own Smells:
Severity | Segment confidence |
No Smells | 100% |
Info | 100% |
Minor | 80% |
Major | 50% |
Critical | 30% |
Blocker | 0% |
If a segment has both a Minor and a Critical Smell, its score is 30% — the worst finding always wins, so a single serious problem can't be masked by several minor ones.
Overall Confidence for the whole work unit is the plain average of every checked segment's score above — every segment with a translated target is checked (in batches for speed), not a sample of some of them, so the Overall Confidence reflects the entire work unit.
Confidence is color-coded everywhere it appears:
Green — High (80% and above)
Orange — Medium (50%–79%)
Red — Low (below 50%)
Turning On the AI Review
The AI review runs as part of QA, but only when you turn it on for the work unit you're checking. If you don't see Enable Smells at all in the steps below, this AI review may not be turned on for your account yet — check with your Account Admin or contact [email protected].
Open the QA panel in the Editor (the warning-triangle icon in the right-hand sidebar).
Click Filters to open QA Settings.
In the Execution section, turn on Enable Smells.
Click Apply — this both saves the setting and immediately runs QA with Smells turned on.
Running QA
Start (or re-run) QA with Run QA — or Run for Current / Run for All when several work units are joined together.
QA runs in two phases: Checks (the standard rule-based QA) and Smells (the AI review), with a progress bar counting how many segments have been analyzed.
When the run finishes, Confidence Scores and Smells appear automatically — you don't need to reload the Editor.
The date and time of the last run is shown as "The report was run on…" above the results.
Reading the Confidence Score
On each segment in the Editor
Each segment shows a small shield icon with its confidence percentage (for example, 92%). Hover over it to confirm it's the QA Confidence Score. It follows the same High / Medium / Low color scale above, and appears in both the translation view and the review view.
Showing or hiding the segment badge
Open the Editor's Settings panel (the gear icon).
Under your viewing preferences, use Show Confidence Score to turn the per-segment badge on or off.
This is saved to your own preferences and remembered the next time you open the Editor. Turning it off only hides the badge — it doesn't change any scores, Smells, or the QA report underneath.
Overall Confidence for the Work Unit
At the top of the QA panel, Overall Confidence summarizes the whole work unit using the same color scale — a red or orange score is a quick signal that the work unit needs attention before it moves on. See How severity maps to confidence above for exactly how this number is calculated.
Working With Smells in the QA Panel
Smells appear in the QA results next to the traditional Issues. Each Smell shows its category (for example, a terminology or accuracy label), a severity pill (Blocker, Critical, Major, Minor, or Info), and a short explanation of the problem.
Ignore a single Smell
Click the eye icon on a Smell to ignore it — ignored Smells no longer count against the segment's Confidence Score. Click the icon again (now an open eye) to bring it back.
Ignore an entire category
From a Smell's three-dots menu, choose Ignore all of this category, or use the eye-slash action on a category chip at the top of the results list. To bring a category back, use the eye action on the same chip.
Filter what you see
Category chips — click a chip to show only that category's Smells. Use the arrow at the end of the chip row to load more categories when there are many.
Segments with Issue / Segments without Issue — narrow the list to segments that do or don't have findings.
Show ignored issues — reveal items you previously ignored so you can review or restore them.
Clear — reset the active filters.
Filter Smells by confidence
Next to Overall Confidence, click Smells to open the Confidence filter. Drag Show confidence up to to show only Smells at or below a chosen confidence level (for example, ≤ 50% to focus on the most serious findings). A dot on the button shows the filter is active; use Clear filter to remove it.
Choosing How Issues and Smells Are Displayed
In the Editor's Settings panel, QA Issues View controls the panel layout — Grouped lists Issues and Smells together on each segment, while Split adds separate Issues and Smells tabs so you can review each type on its own. This is saved to your account.
Right above it, Compact Smells switches each segment's Smell list from a full table to a row of small badges — useful when a segment has several findings and you want to scan categories quickly before opening the detail.
Filtering Segments by Confidence
You can also filter the segment list itself (not just the QA panel) by confidence.
Open the Filter panel in the Editor.
Turn on Confidence Score to filter by score, then choose a comparison — Any score, Equal to, Greater than, Greater equal than, Less than, or Less equal than — and, for anything other than Any score, a value from 0 to 100.
Or turn on No Smells to show only segments with no Smells at all.
These filters help you jump straight to the segments that need work — for example, "confidence less than 50."
Downloading the QA Report
Use Download Report to export the QA results as a spreadsheet. When the AI review has run, the report includes the Smell findings alongside the standard QA Issues.
Important Behavior and Edge Cases
Smells only appear when enabled. If Enable Smells wasn't on before you ran QA, you'll see standard Issues but no Smells or Confidence Scores.
Confidence reflects the worst finding. A single Blocker drops a segment to 0%, even if everything else about it looks fine — read the explanation before acting on the number alone.
Ignoring is shared and reversible. Ignoring a Smell or category hides it for the work unit but never deletes it — turn on Show ignored issues to restore anything.
The badge is a display preference. Turning off Show Confidence Score only hides the per-segment badge; it doesn't change scores, Smells, or the QA report.
Editing a segment can change its score. Fix a segment and re-run QA, and its confidence and Smells refresh to reflect the new translation.
Troubleshooting
I ran QA but see no Confidence Scores or Smells. Confirm Enable Smells was on in QA Settings before the run, then run QA again. If it's still missing, this AI review may not be enabled for your account — check with your Account Admin.
The confidence badge is missing on segments. Check that Show Confidence Score is on in the Editor's Settings panel, and that the AI review has finished running.
A segment shows a very low score but the translation looks correct. Open the Smell's explanation. If you disagree, ignore that Smell (or its category) — the score updates to reflect the remaining findings.
A category is cluttering my results. Use Ignore all of this category from the Smell menu or the category chip. Restore it later with the chip's eye action.
I hid something by mistake. Turn on Show ignored issues, find the item, and click its eye icon to bring it back.
Still stuck? Search the wxrks Community or reach out to [email protected].
Related Articles
How to use the wxrks QA Check Tool — the rule-based tag, length, spelling, and formatting checks that run alongside Smells.
All about Quality & Review in wxrks — how reviewed changes get tracked, categorized, and reported once a segment leaves the Editor.
Quality Dashboard — account-wide charts of review changes by language pair and category.
wxrks Translation Portal — the same Smells engine also generates a Confidence Score for files translated through the Portal, outside the full Editor.