When wxrks pre-translates your files with AI, it can also judge how good each of those translations is — and confirm the strong ones for you instead of sending them to a linguist. This article shows you how to turn that on, how to decide what happens to the segments it approves, and how the approved words are counted and priced on the quote.

ℹ️ Note: Only the segments the current job wrote are scored, and only ones that aren't already confirmed. A second automation pass re-scores just what it rewrote, so you don't pay for the same AI review twice.

The gate lives on the Translation Settings tab, which exists at two levels. Pick the one that matches how widely you want it applied:

ℹ️ Note: The panel opens by itself only at the moment you switch the feature on. When you come back to a unit or project where it's already enabled, the rules are collapsed — click Show the auto-approve rules to see them.

Drag the Minimum confidence slider. It runs from 50% to 100% in 5-point steps and defaults to 85%.

The summary line under the slider restates your choice in plain terms, for example: Segments scoring 85% or higher are confirmed automatically. Anything below stays open for the linguist.

💡 Tip: Start high — 90% or above — on your first project and review what the engine approved before you lower it. A threshold that's too low quietly ships unreviewed translations; one that's too high simply saves you less. The first mistake is much more expensive than the second.

⚠️ Warning: Confirm and Lock are independent, but at least one must stay on. wxrks blocks you from switching off the last active one rather than letting the feature go silently dead — a threshold with no action attached would score segments and then do nothing with the result.

ℹ️ Note: If Auto confirm pre-translated segments is already on, it wins. It confirms every segment regardless of score, which makes the confidence gate beneath it pointless. Rather than disable one of the two behind your back, wxrks shows Auto confirm is on, so every segment is confirmed no matter its score. with a Turn auto confirm off link. Follow that link if you want the score to decide.

Turn on Include in the cost analysis to count approved segments in the Confirmed by Confidence Score band of the logfile.

Leave it off and the gate still confirms and locks segments — it just doesn't change how anything is billed. This split is deliberate: you can adopt the time savings first and the cost change later, once you've configured the price line in Step 3.

ℹ️ Note: With this on, an automated project translates before it is costed, so the very first quote your client sees is already the discounted one.

A segment can qualify for two buckets at once: it may be both a repetition of an earlier segment and confidence-approved. Confidence Score Match Over Repetition decides which one wins.

ℹ️ Note: This mirrors the two toggles beside it that do the same job for translation memory and non-translatable matches. Off is the default deliberately — repetitions are usually your cheapest words, so letting another bucket override them unconditionally can move a large block of words to a different price without anyone intending it.

You can set it in two places, and the more specific one wins:

Organizations > Organizational Units > [Unit] > Project — the default for new projects in that unit. It sits directly beneath Count 100% TM Match Over Repetition and Ignore Subsequent Non-Translatable(s) as Repetitions, with its explanation printed underneath.

Projects > [Project] > Logfiles — this project only. The three precedence toggles sit in a row above the logfile list.

A project created without an explicit value inherits the default in force when it was created, so changing the Organizational Unit later does not retroactively alter existing projects.

On the Logfiles tab, click New Logfile. The Create Logfile modal shows the same setting as a checkbox, pre-filled from the project's current configuration, beside Count 100% TM Match Over Repetition and Ignore Subsequent Non-Translatable(s) as Repetitions. Change it there to analyze one logfile differently without altering the project.

💡 Tip: The modal always opens with the project's current values, so you can flip a toggle on the Logfiles tab and open New Logfile immediately — no page reload needed.

Approved words are only discounted if you tell wxrks what they're worth. Two pieces need to line up.

Your price list includes a price type named Words Translation Confirmed by Confidence Score, added automatically. This is the line the new bucket bills against.

⚠️ Warning: If this price line is missing from the price list you're using, the bucket falls back to the full word rate — silently, as far as the quote is concerned. Approved words would then be billed as brand-new translation, and a quote you expected to fall could rise instead. Confirm the line exists before you rely on the discount.

Go to Settings > Translation Settings > Edit Coefficients and open the coefficient set you use. In its coefficient table, Confirmed by Confidence Score is the top row, directly above Repetitions, and like it has a single factor rather than a percentage range.

The factor is the share of the full word rate you charge for these words. It defaults to 0 — approved words treated as free, on the reasoning that the AI did the work and no linguist touched them. Raise it if you want to bill something for the review the engine performed.

ℹ️ Note: The band has only the TM column. The MT and NT columns are blank for it, exactly as they are for Repetitions.

On the Logfiles tab, click the eye icon in a logfile's Details column to open its summary, then expand the eye icon again on a language row. The breakdown gains a Confirmed by Confidence Score row directly under Total, above Non-translatable and Repetition, with its own Net Rate Factor.

ℹ️ Note: Logfiles generated before you enabled the feature keep their original layout exactly. Exported CSV and XLSX files from those older analyses are unchanged, so historical comparisons stay valid.

Open Projects > [Project] > Costs. Under Cost Results, a green savings line sits between Subtotal and Subtotal with Savings, with its share of the subtotal in brackets.

How that line is labelled depends on where the savings came from:

ℹ️ Note: Both percentages are calculated against the subtotal, so the expanded rows always add up to the combined figure above them.

A project with 10,000 words billed at $0.20 per word, the gate on at 85%, Include in the cost analysis enabled, Repetitions at a factor of 0, and Confirmed by Confidence Score at 0.05.

Of the 10,000 words, 3,000 are confidence-approved — and 1,200 of those are also repetitions.

With Confidence Score Match Over Repetition ON, approved words win the tie:

With it OFF, the 1,200 overlapping words stay repetitions instead:

⚠️ Warning: The same 10,000 words, the same approvals, a $12.00 difference — purely from which bucket wins the overlap. Here repetitions are the cheaper bucket, so turning the toggle on costs you more. Whichever of the two buckets has the lower factor is the one you want winning the overlap. This is the misconfiguration to watch for: two buckets at different factors, and a precedence toggle nobody deliberately set. Check both factors in Edit Coefficients before you change it.

The savings attribution shifts too. In the first case the Costs tab attributes more of the saving to Confidence Score Savings; in the second, more of it sits under the translation memory Savings line — useful to know when you're explaining a quote to a client.

Editing a segment revokes its approval. If a linguist unconfirms a confidence-approved segment, or edits its target text, the approval is cleared. The next logfile counts that segment normally instead of discounting work the translator actually redid. The same applies when a segment is unlocked in bulk, or when it's unconfirmed because its target exceeded the maximum character limit.

A split segment counts as approved only if every piece is. If one part of a split segment falls below the threshold, the whole segment stays out of the approved bucket.

Already-confirmed or locked segments are never touched. Neither are segments with an empty target, or ones carrying a warning-level confirmation.

⚠️ Warning: Importing a translation kit can rewrite target text without clearing the approval, if both confirm only changed segments and confirm all not empty are switched off in that import. Vendor rework arriving that way could keep its discount on a recalculated quote. This affects recalculations only — the initial quote is produced before anyone touches the file.

Client billing only. Vendor payables follow the resulting word counts automatically, but no vendor-payment behavior was changed by this feature.

The confidence switch is greyed out. Context Sensitive is off. Turn on AI translation under Auto Translate From first.

I can't switch off Confirm (or Lock). It's the last active action. Turn the other one on first, or switch off Confirm segments by confidence score entirely to disable the feature.

Everything is being confirmed, regardless of score. Auto confirm pre-translated segments is on and takes priority. Use the Turn auto confirm off link in the warning inside the rules panel.

My quote went up after enabling the cost analysis. The Words Translation Confirmed by Confidence Score price line is almost certainly missing from your price list, so approved words fell back to the full word rate. See the warning in Step 3.

The logfile has no Confirmed by Confidence Score row. Either Include in the cost analysis was off when the analysis ran, or the analysis predates the feature. Generate a new logfile.