Based on last 12 sessions. The band reflects how much your recent scores vary — not a predicted NAATI result.FR-RS-001
The patient has been experiencing persistent lower back pain for the past three weeks. It appears to be muscular in origin, though we need to rule out any spinal involvement.
Should we schedule an MRI, or would X-rays be sufficient at this stage?
Let's start with an X-ray. If the results are inconclusive, we'll escalate to an MRI. Please ensure the patient is aware of the preparation requirements.
Scoring: Accuracy · Fluency · Omissions · Additions · Delivery
Speak clearly and interpret the entire dialogue in sequence.
The real test plays the dialogue in segments of 35 words or less, with a chime after each, and the candidate interprets segment by segment — not the whole dialogue in one 3-minute take. See Conflicts → C1.
"The immigration officer asked the applicant to present their identification documents and proof of current residential address."
Tip: speak clearly, include all key details — especially proper nouns.
Above your 72% average. Accuracy and Delivery were your strongest areas this session.
The doctor says your blood pressure is quite high, so she'd like you to come back next Tuesday for another check before she prescribes anything.
Replays are unlimited in practice. The mock test restricts them.
Press to record your interpretation. Recording never starts on its own.FR-DP-005
The level meter is the honest signal here. Silent capture is the worst failure in this loop, so a near-zero take is rejected before scoring rather than marked as a total omission.FR-DP-010
The doctor says your blood pressure is quite high, so she'd like you to come back next Tuesday for another check before she prescribes anything.
医生说您的血压很高，所以她想让您下周二再来检查一下，可能需要吃药。
STT confidence 94% · Mandarin · Tier A医生说您的血压偏高，所以她希望您下周二再来复查一次，然后再决定是否开药。
One acceptable rendering of six. Matching is on meaning, not wording.FR-SC-007Four of six propositions conveyed. One degree distortion, one omission, one substantive addition.
Accuracy is scoped to distortions only, so it doesn't double-count omissions and additions.Q7
These five criteria are our practice rubric, not NAATI's marking scheme. NAATI deducts from a fixed total, weighted by how much an error impairs communication.BR-SC-005
Computed arithmetically from your audio and word timings — no model involved, so these are the most reproducible numbers on this page.
Speech recognition returned 41% mean confidence on this take. That is below the threshold for Punjabi, so we are not going to pretend a score derived from it is meaningful. Correct the transcript below and we'll score it properly.FR-SC-012
ਡਾਕਟਰ ਕਹਿੰਦਾ ਹੈ ਕਿ ਤੁਹਾਡਾ ਬਲਡ ਪਰੈਸ ਕੁਜ ਵਧਿਆ ਹੋਇਆ ਹੈ
Mean confidence 41% · 3 tokens below 50%Speech recognition for Punjabi is materially weaker than for Mandarin or Spanish. Rather than hide that behind a confident-looking number, Punjabi is marked Tier C: scores are advisory, and they don't move your readiness.FR-CP-006
A cluster of these in one topic usually means our content or language pack is wrong — not that you are.FR-SC-013
我上个月交了押金，可是房东到现在还没有把收据给我，我该怎么办？
No feedback until the test ends. You cannot pause, re-record or go back.FR-MT-003
NAATI banned headphones of any kind in mid-2024. Practising on them rehearses acoustics you won't have on the day.
Completed segments survive a page reload and the timer keeps running. Abandoning records an incomplete attempt rather than discarding your work — but incomplete attempts don't count toward readiness.BR-MT-003
Above the practice pass mark of 63, and above the minimum on both dialogues. This is an estimate from our own marking model — not an official or predicted NAATI result.FR-MT-007
Those segments carry up to ±4 marks of uncertainty, so treat 68 as roughly 64–72. We disclose this rather than present a clean total built on shaky parts.FR-MT-009
A 49-agent research sweep established the CCL's structure — 2 dialogues, segments of 35 words or less, a 5-second start window — but verified no marking number at all, because naati.com.au is blocked by our egress proxy and no primary source was read. Total marks, pass mark and the per-dialogue minimum are uninvestigated, not disputed. The circulating 90 / 63 / 29 are shown so they can be checked, not because they are right. One config object holds them all. Do not ship scoring against these. Q1 remains blocking.
A mock counts for more than practice, because practice conditions are easier.FR-RS-004
Every mock is logged against the readiness that preceded it. That comparison is the only thing that keeps the headline number honest.FR-RS-011
You dropped temporal qualifiers in 4 of 22 segments — the most repeated single pattern in this attempt.
Every screen in this prototype is built from your Figma. Six things in it conflicted with what the CCL research established, or with itself. C6 is now resolved — the design system follows the marketing site. None of the rest are hard to fix now; all are expensive to fix after build.
Session — Dialogue Recording gives a single 3:00 take for a three-turn dialogue. The real CCL plays segments of 35 words or less, sounds a chime, and the candidate interprets that segment before the next plays. This is the single biggest gap: it changes the memory load, the note-taking strategy, and what we can score. Practising one long take does not prepare anyone for segment-by-segment interpreting.
From Detail — Mock Test. Word-for-word transfer is what a weak interpreter does; NAATI assesses whether meaning survives, and a literal rendering that mangles idiom is penalised. Teaching candidates to aim for word-for-word would actively lower their scores.
From Detail — Mock Test. No NAATI source publishes any such weighting — the research looked and found only a single unsourced prep site claiming a split. Presenting invented weights as "what's scored in the exam" is the exact failure BR-SC-005 exists to prevent.
Test duration came back UNKNOWN. The only duration-adjacent fact established is that long pauses may cost marks, particularly past a 20-minute total recording. 45 minutes has no basis found.
The session screens say English → Hindi; Vocabulary shows Persian/Dari (ترتیبی، مترجم شفاهی). Both are plausible CCL languages, but a single candidate has exactly one. Pick one for the prototype, and make sure RTL is tested — Persian, Dari, Arabic and Urdu all need it.
The indigo modals were right and the green dashboard was the outlier. The marketing site is a slate/indigo system built on Geist — its stylesheet literally says “keep the slate/indigo system”. Green was never the brand: emerald appears there only as a success colour. This prototype is now rebuilt on the marketing tokens, and green has been demoted to what it should mean — passed, conveyed, improved.