Privacy notice

What the tools on this site do with data, and what we keep. The mock orals store no recording and nothing a student said.

Last updated 30 July 2026.

The short version

The practice tools on this site need no account, no sign-up and no name. Nothing a student says into a mock oral is stored by us, and no recording of anyone's voice is ever kept. We do hold contact details for teachers who ask us for papers, and we hold written work that students choose to send in for marking. Everything below says which is which.

The Chinese mock orals

This covers the Chinese Continuers and Chinese Beginners mock orals.

The student's answer is recorded in their browser and sent, one answer at a time, to Google's Gemini API to be transcribed and answered. The model is gemini-3.6-flash. We have not trained or fine-tuned anything. It is the off-the-shelf commercial model, steered by a written set of instructions that holds the exam rules, the question bank and the marking criteria, and no student data has ever been used to build or tune it.

We keep no audio. The recording is never written to our storage at any point.

We keep none of the words either. What we store for a session is a random session number generated in the browser, the date, how many turns there were, how many times the student asked for a question to be repeated, how many takes had no voice in them, and whether a report was produced. That is the entire record. It contains nothing a student said and nothing that identifies anyone.

The examiner's spoken voice is synthesised by MiniMax, model speech-2.8-turbo. Only the text of the examiner's own next question goes there, never any student audio, and it is cached for two days so a repeated question does not have to be synthesised again.

We are on Google's paid API tier, not the free tier, and that distinction is the reason we pay. Google's API terms state that for paid services your prompts and responses are not used to improve Google products, and are logged only for a limited period to detect abuse. The free tier permits both product improvement and human review of API input and output.

The Japanese mock oral

This one is built differently and the difference is worth knowing. The conversation runs on OpenAI's realtime model gpt-realtime-2.1-mini, with gpt-transcribe for transcription and gpt-4o-mini-tts for the examiner's voice. The student's browser connects to OpenAI directly, so their audio does not pass through our servers at all. The written report, and the kana and romaji shown under each line, are produced by Google's gemini-3.6-flash on the same paid account described above.

We store nothing from a Japanese session. No transcript, no audio.

The Chinese and Literature Study Hub

The Study Hub at clhsc.hqsociety.com carries an AI tutor for the prescribed texts. A question typed into it is sent to Google's Vertex AI Search on our paid Google Cloud account, model gemini-3.1-pro-preview, which answers from our own study notes. We keep no record of the questions and no record of who asked them. The only thing written to storage is the usage counter described under Usage limits below, and the site itself needs no account and asks for no name.

Written work sent in for marking

This is the one place we do hold student work. If a student uses the Send for feedback button on a practice paper, their written answers are turned into a PDF and stored so that it can be marked, together with the name and email address they typed into the form. Nobody outside HQ Society sees it. Email us and we delete it.

Teachers who ask us for papers

The form on each paper page asks for a name, a school and an email address. We use them to email you the files and sometimes to tell you when new papers are ready. We do not sell these details and we do not share them. Reply to any of our emails and say so, and you come off the list.

Payments

Paid orders go through Stripe. Card details are handled by Stripe and never reach us. A school that asks for an invoice gives us the billing details its finance office needs, and nothing beyond that.

Counting visits

We use Cloudflare Web Analytics, which counts page views without cookies and without building a profile of anyone. Some practice pages also record which questions students found difficult, as plain counts, with nothing attached that identifies a person.

Usage limits

So that one visitor cannot run up the model bill for everybody, the tools count requests per network. That counter keys on a one-way hash of the IP address rather than the address itself, holds a number and nothing else, and expires after two days.

Who else handles data

Google, for the models behind the Chinese tools and the reports in the Japanese one. OpenAI, for the Japanese conversation, its transcription and its voice. MiniMax, for the Chinese examiner's voice. Cloudflare, for hosting, storage and analytics. Ghost, for this website and its email. Stripe, for payments. There are no advertising networks and no tracking services.

How long we keep things

The session counters described above are deleted after a year, and that limit is enforced by the storage itself rather than by someone remembering to run a script. Written work sent in for marking is kept so that it can be marked, and is deleted whenever you ask. Teacher contact details are kept until you ask us to remove them.

Asking us something, or asking us to delete something

Email info@hqsociety.com. If you are a school assessing these tools and you need something more specific than this page, ask and we will answer it properly. If you want every record from the dates your students used a tool removed, tell us the dates and we will remove them.

Changes

If what the tools do with data changes, this page changes with them and the date at the top moves.