Smells is wxrks's automatic quality evaluation (AQE): it reviews each translated segment and reports what it finds. A Smell is wxrks's AI-powered warning that a translation might have a problem worth a second look — inconsistent terminology, awkward phrasing, or a mistranslated idiom, for example. It runs right in the Editor, on one segment at a time, either automatically as you work or on demand whenever you want a second opinion.
Smells now runs against the same context your translation was produced with: the project's Glossary, its Translation Memory (TM), the Sous Chef Instructions in force, and — when you haven't edited the target since — the reasoning behind the Context Sensitive translation itself. That means fewer findings that argue with decisions the project already made on purpose, and one new finding type that catches a divergence from an approved TM translation.
💡 Who is this for? This guide is for anyone translating or reviewing in the Editor — Vendors working on assigned tasks, and Project Managers reviewing their projects' segments — who need to read in-Editor Smells warnings and judge which ones to act on. No special permission is required beyond access to the task: if you can open the segment in the Editor, you can use Smells.
What is a Smell?
A Smell flags a translation that reads as potentially wrong or awkward — not necessarily one that definitely is. wxrks checks for smells automatically as you work (if enabled in your Editor preferences), or on demand via Check Smells in the segment's Augmented Actions menu.
A Smell is a signal, not a verdict: the final call is always yours. A smell can flag a real issue, and its absence doesn't guarantee the translation is error-free.
Smells are one of several Augmented Actions — AI-powered tools that assist your translation work without replacing it. Turning them on for your account, Organizational Unit (the group your projects belong to, which carries its own settings), and Editor preferences is covered in wxrks Augmented Actions; this article focuses specifically on Smells.
Key Concepts
Term | Meaning |
Smell | A potential quality problem the AI review found in one segment's translation. Each Smell has a type (category) and a short explanation. |
Check Smells | The button in a segment's Augmented Actions menu that runs a Smells check on demand. |
Fix Smell | Asks the AI to rewrite the segment's translation to address one specific Smell. |
Ignore Smell | Hides one Smell finding you've reviewed and dismissed. Reversible. |
Context chips | The small badges under the Smells header showing which context the check used: Glossary (n), TM (n), Sous Chef, CS rationale. Hover any chip to see exactly what was used. |
Context Sensitive (CS) | wxrks's AI translation mode, which combines machine translation, translation memories and glossaries. Smells reuses the context it resolved. See Context Sensitive Translate. |
Sous Chef Instructions | The style controls a translation is produced under — domain, audience, formality, tone, creativity, and free-text instructions. |
TM Match | A finding type raised when your target diverges from a Translation Memory match the project's own settings say should have won. |
Running a Smells Check
Automatically as you work
With Check Smells turned on in the Editor's Settings panel, wxrks checks a segment for smells whenever you edit it and move on — it skips segments you've already confirmed or locked, so it only runs while you're actively working on a translation, not on finished work.
On demand
Select any segment, click its Augmented Actions button, and choose Check Smells to run the check yourself — useful when you'd rather trigger it on your own schedule than have it run automatically.
Turning Check Smells off in your Editor Settings only affects you — it removes the automatic check and the button from your own Augmented Actions menu, without changing anything for other users on the account.
What Smells Checks Against
Before the check runs, wxrks gathers the same context the Context Sensitive translation resolved for that exact segment. When the check has just followed a Context Sensitive translation, this costs nothing extra; otherwise wxrks looks the pieces up itself, using the project's own prioritization.
Four pieces can apply:
Context | What Smells does with it |
Glossary | Mandatory terms matched in this segment are treated as correct by definition; forbidden terms must not appear in the target. A Terminology finding is raised only when a mandatory term is missing or replaced, or a forbidden term is used. |
Translation Memory (TM) | Approved translations for this segment are supplied as references, so Smells prefers their terminology and style. A divergence from a fuzzy match (anything below 100%, where the stored source only partly matches yours) is never reported as an error. |
Sous Chef Instructions | The style controls the translation was produced under. Smells will not report something that merely reflects them — an informal register when formality is set to informal, expressive wording when creativity is high. |
CS rationale | The Context Sensitive translation's own explanation of its choices, treated as claims to verify rather than facts. It still reports a decision that contradicts the source meaning or a glossary entry. |
ℹ️ Note: Only pieces that actually apply to the segment are used. A segment with no glossary match and no TM hit is checked exactly as before.
Reading the context chips
When context was used, a row of chips appears directly under the Smells header, showing what the check had in hand:
Glossary (n) — n glossary terms matched in this segment. The tooltip lists each pair as
source → target, marking any entry that is (forbidden).TM (n) — n Translation Memory references at or above the reference threshold. The tooltip lists each as
[match%] target.Sous Chef — Sous Chef Instructions were in force. The tooltip lists the controls used — Domain, Audience, Formality, Tone, Creativity and the free-text Instructions.
CS rationale — the Context Sensitive reasoning was reused. The tooltip shows the reasoning text.
Hover any chip to see exactly what went into the check — the tooltips mirror the check's own input, so what you read is what Smells read.
A worked example
An English–Brazilian Portuguese project has a glossary entry making Entrega Rápida the mandatory translation of Fast Delivery, a Translation Memory holding the approved sentence, and Sous Chef Instructions asking for an informal, conversational voice. The target for Our Fast Delivery service brings your order in up to two business days. reads Nosso serviço cesta traz seu pedido em até dois dias úteis.
Before, Smells had no idea a glossary or a TM existed. It could only judge the sentence on how it read — so a wrong product name either passed unnoticed or came back as a vague fluency complaint.
Now, the chips show Glossary (1), TM (1) and Sous Chef, and two findings come back: a Terminology issue — The mandatory product term "Fast Delivery" was replaced with "cesta." Use "Entrega Rápida" instead. — and a TM Match issue for the divergence from the stored 101% translation.
The same mechanism works in reverse. The informal register those Sous Chef Instructions ask for is not flagged as a problem — the Sous Chef chip shows the check knew it was deliberate.
💡 Tip: If a finding surprises you, read the chips before acting on it. They usually explain why the check reached that conclusion — and they equally tell you when a piece of context was not available.
Reading a Smell Finding
A finding shows the Smell's type (a category badge, such as Grammar, Semantic, Wording, Fluency, Terminology, or TM Match) and a plain-language explanation of the problem — right under the segment itself, so you don't have to leave the translation to see what was flagged.
Two actions are available on most findings (not on the UI Term type, which has no rewrite to offer):
Click the hammer icon (Fix Smell) to have the AI rewrite the translation to address that specific finding. The rewrite is given the same context as the check, so it respects the glossary and the Sous Chef Instructions.
Click the eye icon (Ignore Smell) to dismiss a finding you've reviewed and don't agree with. The finding leaves the list and stays one click away — see Ignoring and restoring a Smell below.
If a segment has several findings, the Compact view option under Smells in the Editor's Settings panel (AI assistance section) controls how they're shown: as a full list, or as a row of type badges you open one at a time.
If nothing was flagged, wxrks shows No Smells found. — a reminder that a clean result means no likely issue was detected, not a guarantee the translation is perfect. The context chips still appear on a clean result, so you can confirm what the check considered.
Ignoring and restoring a Smell
Ignoring a Smell tells wxrks you reviewed that finding and decided to keep your translation. The finding stops competing with the ones you still have to act on, but it is never lost: you can see every finding you ignored on the segment and bring any of them back.
Click the eye icon (Ignore Smell) on a finding. The row leaves the list, and the header updates to count only what's still open — for example, Smells · 1 check · 1 ignored. Fix all no longer counts the ignored finding, so it won't rewrite the translation for something you already dismissed (and it disappears when fewer than two findings are left to fix).
Changed your mind straight away? Click Undo in the line under the list (for example, Grammar ignored on this segment · Undo). The finding returns to its place right away, without running a new check. Undo only covers your last ignore and disappears as soon as you take another action on the panel, such as Fix Smell or a new check.
To review everything you've ignored, click the N ignored toggle under the list (for example, 1 ignored). The ignored findings are listed struck through, each with its type and explanation.
Click the undo arrow (Restore Smell) next to a finding to bring it back into the open list, where you can fix it or ignore it again.
ℹ️ Note: Ignoring is saved on the segment, so it survives a re-check. When you run Check Smells again, findings you already ignored come back straight into the N ignored list instead of reopening, and you can still restore them from there.
A few states look different from the regular list:
Everything is ignored. The panel stays open and reads No open Smells. N ignored on this segment., so you can still reach the ignored list.
Compact view is on. A type whose findings are all ignored drops out of the row of type badges. Restoring one of its findings brings the badge back and opens that type.
Ignored for a whole category. If someone used Ignore all of this category for a Smell type in the QA panel, the findings of that type appear in the ignored list with Ignored for all type (for example, Ignored for all Grammar) instead of the Restore Smell button. Restoring a single finding can't undo a category-wide rule, so undo it in the QA panel — see QA Confidence Score and Smells.
If an ignore or a restore can't be saved, the finding goes back where it was and the line under the list explains what happened: Could not ignore this Smell. Try again. or Could not restore this Smell. Run the check again. The second message usually means the segment's check results are more than about two hours old — run Check Smells again from the segment's Augmented actions menu, then restore the finding from the refreshed N ignored list.
The TM Match finding
TM Match is the one finding type wxrks decides without the AI. It appears when your target diverges from a Translation Memory match that the project's own settings say should have won, and its explanation quotes the match and its stored translation:
This project prioritizes translation memory matches over CS translation. Target diverges from the 101% translation memory match: "Nosso serviço Entrega Rápida traz seu pedido em até dois dias úteis."
Its severity depends on your project's translation settings:
Situation | Severity |
The project has Prioritize TM matches over Context Sensitive translation set, and the match meets its Minimum TM match threshold | Minor |
No threshold is configured, but the match is 101% — an exact match whose surrounding segments also match, so the stored translation is known to fit this context | Info |
Anything else | No finding is raised |
Because it is deterministic, a TM Match finding is a statement of fact about a divergence, not an opinion about quality. Keep your wording when you diverged on purpose — use Ignore Smell to dismiss it.
Important Behavior and Edge Cases
Smells need Augmented Translation Actions enabled. If your account, Organizational Unit, or personal Editor preferences have Augmented Actions turned off, Check Smells won't be available — see wxrks Augmented Actions for the three levels this is controlled at.
No chips means no context applied. An absent chip row is not an error: the segment simply had no glossary match, no TM reference above the threshold, and no Sous Chef Instructions in force.
The CS rationale only survives an unedited target. As soon as you change the target text, the Context Sensitive reasoning no longer describes what's on screen, so wxrks drops it and the CS rationale chip disappears. Glossary, TM and Sous Chef context still apply.
Context follows the project's own prioritization. If Translation Memory or Sous Chef Instructions are disabled as providers in your project's Context Sensitive Priority list, Smells won't use them either — it respects the same configuration the translation did.
A clean result isn't a guarantee. No smells found means nothing was flagged, not that the translation is certainly error-free.
Confirmed and locked segments are skipped automatically. The automatic check only runs while you're actively editing a segment.
This is a different surface from the QA panel's Smells. This article covers the per-segment check in the Suggestions area of the Editor. wxrks also runs the same AI review in batch, across a whole work unit, from the QA panel — which additionally scores each segment's severity into a Confidence Score. The batch run does not yet use this context, so a finding you see there may differ from the in-Editor result. See QA Confidence Score and Smells for that flow.
Glossary-aware checking does not replace the QA Check Tool. Smells now understands the glossary entries that match the segment in front of you; the rule-based QA Check Tool still runs the exhaustive, whole-file consistency checks.
Troubleshooting
I don't see Check Smells anywhere. Confirm Augmented Translation Actions is turned on for your account and Organizational Unit, then check your own Editor Settings panel — see wxrks Augmented Actions.
I don't see any context chips. The chips only appear when context applied to that specific segment. Check whether the segment actually matches a glossary term or a TM entry, and whether Translation Memory, Glossary and Sous Chef Instructions are enabled as providers in the project's Context Sensitive Priority settings. Your Project Manager or Account Admin manages that list.
Smells flagged a term my glossary requires. Check the Glossary (n) chip. If the term isn't listed there, it didn't match this segment — confirm the entry's source form matches the segment's wording. If it is listed and was still flagged, treat the finding as a false positive and use Ignore Smell.
Smells flagged the informal tone my project asked for. Check for the Sous Chef chip. If it's missing, the instructions weren't in force for this segment — the check had no way to know the register was intentional.
A TM Match finding won't go away. It's deterministic: it reappears while your target still differs from that match. Use Ignore Smell to dismiss it if the divergence is deliberate.
A finding is in the wrong language. Set your preferred Smells Language (Source, Target, or English) in the Editor's Settings panel.
I disagree with a finding. Click its eye icon to ignore it — this doesn't change the translation, just hides that one finding for you.
Still stuck? Search the wxrks Community or reach out to [email protected].
Related Articles
wxrks Augmented Actions — the full set of AI-powered Editor actions, and how to turn them on for your account.
Context Sensitive Translate — the AI translation mode whose context Smells now reuses.
Translation Reasoning — the reasoning behind a Context Sensitive translation, shown as the CS rationale chip here.
QA Confidence Score and Smells — the same AI review run in batch across a whole work unit from the QA panel, with severity-based Confidence Scores.
How to use the wxrks QA Check Tool — the rule-based, whole-file checks that complement Smells.
Auto-Confirming Segments by Confidence Score — the quality gate that uses the batch Confidence Scores.
Connect your own AI engine for automatic post-editing (Context Sensitive) and quality evaluation (Smells) — which engine runs Smells and QA Smells.