What is completeness, and how does it change your reference-mode score?

Completeness is a score that only exists in reference (read-aloud) mode, and it answers a question accuracy alone can't: how much of the passage did you actually attempt? It's the fraction of the expected words you produced, as opposed to skipped, stopped before, or never got to — and it's reported as its own number precisely so a short or partial read isn't scored as if it were a finished one, in either direction.

30-second version: Because reference mode knows the full target text in advance, it can measure completeness directly: words attempted divided by words expected. Your overall score then blends accuracy and fluency for the words you did read, scaled down by how much of the passage you covered — so stopping halfway costs you something honest, rather than either vanishing into a flattering accuracy-only number or getting silently double-penalized.

How is completeness actually calculated?

It's the share of the passage's expected words that you produced some recognizable attempt at, out of the total the passage contains. If you read 18 of 30 words in a passage before stopping, completeness sits around 60%. Words you never got to are marked with their own verdict, unread, and are excluded from your accuracy score entirely — they don't count as wrong, because you never attempted them, but they do count against completeness, because the passage wasn't finished.

Why not just score the words I actually read and ignore the rest?

Because that alone can be misleading in the opposite direction: reading eleven clean words out of a two-hundred-word passage and stopping could still average out to a strong accuracy score, which would tell you nothing about how much of the exercise you actually completed. Completeness exists specifically so accuracy — a measure of how well you said what you said — is never mistaken for a measure of how much you said. The minimum-sample rule goes a step further for very short reads, withholding a headline score entirely rather than publishing one built on a handful of words.

How does completeness affect my overall score?

Your overall score blends accuracy (weighted more heavily) and fluency for the portion you actually read, then that blended number is scaled by completeness — reading half a passage well scores lower overall than reading all of it equally well, because finishing what you set out to read is part of the task. The scaling is gentle rather than punishing: a half-finished read isn't cut in half, but it is honestly reduced, in a way you can see reflected in the completeness number itself rather than hidden inside a single opaque score.

What if I go off-script partway through?

Speech that drifts away from the practiced text — a stray remark, a restart, a correction you say out loud — is reported separately as extra content, not charged against the passage's own words. So going off-script doesn't tank your completeness the way skipping ahead would; the passage's own word-by-word record stays about the passage.

Reference-mode scores at a glance

Does completeness apply outside read-aloud passages?

No — it's specific to reference mode, because it needs a known target length to measure against. Freeform speech, like an open IELTS answer, has no fixed passage to be "complete" relative to, so that mode reports fluency and accuracy without a completeness figure.

Try it

Start reading a longer passage in sayit and stop partway through on purpose — the results will show your completeness percentage alongside accuracy, so you can see the two numbers doing separate jobs rather than one number trying to do both.