Smells Scoring decides how much each Translation Smell lowers a segment's confidence score, which severity each kind of smell gets, and whether pre-translated segments are confirmed automatically when their score is good enough. This article explains how the score is calculated, how to set each part at the account, Organizational Unit and project level, and how to avoid the settings that most often produce unexpected results.

Who can do this:

Linguists don't change these settings. They see the result: the confidence score of each segment in the Editor, and a tooltip on each smell's severity that says what it costs.

Every smell has a severity. Each severity has a weight: the confidence a segment keeps when that smell is the worst one still open on it.

A segment's confidence is the weight of its worst open smell, and 100% when there is none. Smells are not added up: a segment with one Major and three Minor smells scores 50%, the same as a segment with only the Major. Smells a linguist ignores no longer count, so ignoring the worst smell raises the score to the weight of the next one. The same calculation is used everywhere the score appears: the QA batch, the Editor's QA panel, and the new score after a smell is ignored.

Worked example. A recent batch has 26 segments: 23 with no smell that counts, 1 whose worst smell is Minor, and 2 whose worst smell is Major. With the default weights, the average confidence is (23 × 100% + 1 × 80% + 2 × 50%) ÷ 26 = 95.4%. With the confirmation rule set to Up to Minor, 24 of the 26 segments are confirmed and the 2 Major segments stay for review.

The page has four parts: Smells sensitivity, How much each severity weighs, Which severity each smell type gets, and When segments are confirmed. Organizational Units and projects follow the level above for each part separately until someone changes that part there (see Customize an Organizational Unit or project below).

Sensitivity controls how much the Smells check reports in the Editor:

Every level reports meaning errors, missing or invented content, untranslated text, numbers, placeholders and project rules. Turn on Translators can adjust it in the editor to let linguists change the level for themselves.

ℹ️ Note: Sensitivity only changes what linguists see while they work. Quality scores and confirmation by confidence always use Standard, so a translator who switches to Low doesn't raise the scores.

Under How much each severity weighs, use − and + to set each weight, in steps of 5%. Or open Presets and pick one:

A lighter severity can never weigh less than a heavier one. If it does, the page shows an error such as Minor can't lower the score more than Major. and the impact preview asks you to fix the weights first.

Why change the weights? The weights decide where segments land relative to your confirmation rule and how the average confidence moves. Lower weights make the same smells cost more. Weights that are equal make severities indistinguishable for the rule (see Common misconfigurations below).

Which severity each smell type gets lists every kind of smell the detector has reported for your account, with how often it was seen in the last 90 days (Seen, 90 days) and the severity the model usually gives it (Model usually says). For each type, choose a Severity:

Use the filters (All, New, Set by you, Changed below) to find types, and Bulk actions to select several rows, choose Set to, and click Apply. The Set by column shows where each severity comes from.

The list starts empty and fills as the detector reports smells, so a new account shows No smell types yet.

When pre-translation finishes, the Smells check scores the segments it wrote. When segments are confirmed decides what happens to them.

Each option shows how many of your recent segments it would confirm, for example confirms 24 of 26, and Segments with these severities stay for review lists what will not be confirmed. 3. Choose what happens to the segments the rule accepts: * Confirm: marks them as confirmed. They stay editable. * Lock: locks them so they can no longer be edited. * Include in the cost analysis: counts their words in the Confirmed by Confidence Score band of the analysis, so they can be priced at their own rate.

⚠️ Warning: With Lock on, confirmed segments can't be edited by linguists. Only lock when a segment at your chosen severity can safely be delivered without a human look.

Before you save, the Impact preview shows what your draft would have done to the QA and pre-translation runs of the last 90 days. It is computed from smells already stored, so it makes no AI calls and costs nothing.

When the numbers look right, click Save changes. The confirmation dialog lists each change, how many Organizational Units and projects follow these settings, and which ones keep their own. Click Save to apply, or Discard on the page to drop the draft.

ℹ️ Note: The new settings apply to QA runs from now on. Scores already computed stay as they are.

At an Organizational Unit or a project, each part shows Follows account or Follows followed by the Organizational Unit's name, and a Customize button.

The other parts keep following. For smell types, a lower level changes only the rows it needs: set a row back to what the level above says and it follows again. If most types end up different, change the account's table instead, as the page suggests.

On a project, Translation & Editor Settings > Confirm & lock still shows a read-only Confirm segments by confidence score line with the current state and an Edit in Smells Scoring button that opens this tab.

Smells Scoring replaces the old Confirm segments by confidence score switch and its minimum confidence slider. Your previous settings were converted automatically: a minimum above 80% became No smell that lowers the score, above 50% became Up to Minor, and above 30% became Up to Major. Confirm, Lock and the cost setting were kept.