Privacy Policy

Quantametron is a math tutor for kids in Grades 4–8, currently in invite-only private preview. “We,” “us,” and “Quantametron” refer to the operator of this service. This policy explains what we collect, how we use it, and the rights a parent or legal guardian has over their child's data.

Quantametron is for parents or legal guardians and their children in Grades 4–8. Kids don't sign up. A parent creates the account, defines kid profiles, and supervises use. Kids then use the parent's signed-in session on a shared family device.

Because Grades 4–5 are typically ages 9–10, we know some of our users are under 13. We treat this as a service subject to the Children's Online Privacy Protection Act (COPPA) and similar state laws.

From the parent:

• Email address and password (for parent sign-in)
• A parent PIN (to unlock the parent admin area)
• A household display name

About each kid (entered by the parent):

• Display name
• School grade (4–8) and optional accelerated “math grade”
• Optional “About me” note — hobbies or interests you'd like the tutor to reference
• Timezone (for scheduling the daily plan boundary)

From the kid as they use the tutor:

• Math questions they type and their answers
• Which steps they expand, which hints or walkthroughs they use, whether they asked the tutor a clarifying question, and how many attempts each question took
• Placement-diagnostic responses (6–10 short questions on first use) that seed the kid's starting point — skippable
• “Prove It” short-quiz answers — 5 harder questions per topic, no hints, used to mark a topic Mastered when passed
• Grade-level test answers — a 30-question checkup across every topic in the kid's grade, same no-hints rules as Prove It
• Guided-lesson conversations (when they start a “teach me…” lesson)
• Plan-item interactions (Start, Resume, Not today) on the daily plan
• The last subtopic they opened on the Campaign — one timestamp per (kid, topic), used to render the “Resume” card on the Campaign map
• Issue reports (“this question is confusing”) they file

Computed from the above (derived data, not typed by the kid):

• Per-topic mastery and readiness scores — our “skill graph” that drives frontier recommendations, review suggestions, and the Campaign map
• Mastery-promotion events — a single timestamp per topic recording when the kid passed the Prove-It quiz (or had a passing topic promoted inside the grade-level test). Lets the UI flip a topic to “Mastered” deterministically instead of waiting for the rolling clean-win average to drift past the threshold.
• Misconception labels — when the tutor's answer judge detects a known pattern in a wrong answer (e.g., “added the denominators”)
• A freshness / decay signal per mastered topic, used to suggest reviews without dropping the mastery count
• A “revisit” queue — the set of questions the kid has attempted but not yet answered correctly, surfaced on the Campaign page so they can clean up unfinished questions

Automatically collected (standard for any web service):

• Timestamps of sessions and questions
• IP address (for rate-limiting and abuse prevention only)
• Browser user-agent (for diagnostic logging)

Parent → kid messages.When a parent clicks one of the preset nudge buttons on the dashboard (e.g., “Send nice work today”), we store: the sender (parent profile id), the recipient (kid profile id), the preset key or custom message text (capped at 280 characters), the send timestamp, and an optional topic context. The kid sees the message as a banner the next time they open the math homepage, and dismisses it with an X. Dismissed messages stay in the database (audit trail) but stop showing as banners.

We do notcollect the kid's real name, school, precise location, photos, voice recordings, contacts, or anything not listed above.

• To run the math tutor:questions, grade, and the optional “About me” note are sent to our AI provider (Anthropic, see “Who else sees it” below) to generate step-by-step explanations and to judge answers.
• To compose a daily practice plan:we derive 2–4 items a day from the kid's skill graph (frontier topics, topics due for review, a guided-lesson suggestion). Composition is deterministic and does not use Anthropic.
• To offer targeted remediation:when the answer judge detects a known misconception, we may offer a short refresher on the relevant prerequisite topic. The refresher is generated by Anthropic; it receives the misconception label, the prerequisite topic name, and (optionally) the kid's exact wrong answer.
• To adapt a one-time placement diagnostic:6–10 short questions drawn from our question catalog, selected adaptively from the kid's responses. No Anthropic calls are made during placement.
• To show the parent dashboard:progress, topics mastered, recent attempts, today's plan, and a weekly narrative summary. The weekly summary's one-to-two narrative sentences are generated by Anthropic from aggregated stats about the kid's week (not the kid's raw typed content).
• To send transactional email (invites, password and PIN reset codes, waitlist confirmation) via our email provider.
• To prevent abuse (rate-limits keyed to IP and user account).

We don't show ads to kids. We don't sell personal data. We don'tuse kid data to train general AI models. We don't profile kids for anything outside the tutor.

We use a small number of third-party services to run Quantametron. Each receives only the data it needs to do its job:

• Anthropic(Claude) — receives tutor inputs: each math question, the kid's grade, the optional “About me” note, the kid's typed answers (for judging), hint / walkthrough requests, guided- lesson conversation turns, remediation prompts (misconception label + optional wrong answer), and, for the parent weekly summary, aggregated non-identifying stats about the kid's week. No raw typed content is sent for the weekly summary — only aggregated numbers and topic names. Anthropic's privacy terms apply.
• Resend — receives recipient email addresses and the content of emails we send (invites, password / PIN resets, waitlist confirmation).
• Supabase — hosts our database and handles parent authentication.
• Vercel — hosts the application.

None of these providers are permitted to use your data for their own purposes beyond delivering the service to us.

• Kid-typed math questions: 60 days, then the text is scrubbed (the aggregate — which grade, how many attempts, what topic — is kept).
• Kid-typed answers: 60 days, then scrubbed.
• Saved questions (“My questions” list): 60 days, then the row is deleted.
• Shadow-judge comparison log: 30 days, then deleted.
• Aggregate progress (topics solved, attempt counts, time-series): kept as long as the kid's profile exists.
• Parent sign-in data: kept as long as the account exists. Delete the account and it's wiped.

Anything you delete via the parent dashboard (see below) is removed within 7 days.

Each curriculum topic (e.g., “Multi-Digit Division”) has a short structured lesson — intro, key concepts, and two worked examples — generated once by Anthropic and cached in our database. Every kid working on that topic sees the same cached lesson content. These rows are keyed by topic, not by kid, and contain no personal data, no questions the kid typed, and no answers. They don't change when a kid's profile is deleted because they aren't linked to any kid.

Before we scrub the 60-day content described above, we copy certain pedagogically useful items — wrong answers, detected misconception patterns, and free-form math questions — into a separate de-identified corpus. These copies do not contain your child's name, your household, your email, a user ID, or any link back to your family. They record only: the typed text (wrong answer or question), the topic or catalog question, the misconception label, and the original timestamp.

We use this corpus to improve the tutor's explanations and grow our library of common misconceptions — the same way a teacher might notice “a lot of kids trip on this step” and adjust their lesson, without needing to know which kid. Because the corpus is de-identified at copy time, deleting your account or your child's profile does not delete these anonymous examples, but they can't be traced back to you.

You can, at any time:

• Reviewyour kid's data — in the parent admin area or via the data export tool.
• Correct any data — edit your kid's profile directly.
• Deleteyour kid's data — remove a single kid or the whole household from Settings.
• Refuse further collection — close the account.
• Export your kid's data as JSON — from Settings.

If you can't sign in or need help, email privacy@quantametron.com and we'll respond within 30 days.

Parents can review what their kid asks the tutor.When a kid types a question into the tutor — through the “Ask the tutor” affordance, the guided lesson chat, or the answer box — the conversation is stored on the kid's profile and visible to the parent who owns the household.

This is a deliberate product choice, not a side effect. Under COPPA (US, kids under 13), parents are the consent-giver and have the legal right to review their child's personal information collected by the service. We extend the same review path to all ages, not just under-13. Parent visibility is the default and is not togglable.

On a kid's first visit to the math homepage we show them a one-time notice explaining this in plain language: “Your grown-up can see what you ask the tutor.” We believe transparency at the start is better than discovery after the fact.

What we do NOT do:

• Show one kid's conversations to a different kid's parent.
• Show conversations to anyone outside your household.
• Use kid conversations to train external models. (See “Sub-processors” — we use the Anthropic API with no-training defaults.)

We run a small, deterministic safety check on every message a kid sends to the tutor before the message reaches the AI. It looks for language patterns associated with self-harm, abuse disclosure, eating-disorder behavior, or violence ideation. The check is a regex pattern list — not an AI classifier — so it's fast, predictable, and doesn't send your kid's words to a separate model.

When the check fires:

1. The tutor does not respond with its usual answer. Instead, the kid sees a short, warm message acknowledging what they wrote, plus a list of crisis-line resources they can reach right now (988 Suicide & Crisis Lifeline; Crisis Text Line; Childhelp National Child Abuse Hotline).
2. We record the event — an 80-character snippet around the matching phrase, the category, the surface where it was typed, and the timestamp. The kid's full message text is NOT stored.
3. A red banner surfaces at the top of the parent dashboard with the kid's name, what category was flagged, and the snippet, so the parent knows to follow up.

We tune the detector toward false positives over false negatives — a kid who isn't in distress occasionally seeing crisis resources is a smaller harm than a kid in real distress not seeing them. If you find the detector triggering on benign content, please tell us at privacy@quantametron.com.

This is not a crisis service. Quantametron is a math tutor. If your child is in immediate danger, call 911. The detector is a safety net, not a substitute for professional support.

Passwords and parent PINs are bcrypt-hashed. Sessions are cookie-based and expire. Our database is hosted in a Supabase project with row-level security enabled. We use HTTPS everywhere.

We're a small preview. We don't claim military-grade security. If you have a security concern, email privacy@quantametron.com.

• We collect the minimum information needed to run the tutor.
• We don't condition participation on providing more than is needed.
• We obtain parental consent at signup (a required checkbox) and again through the email-based invite process — a parent must own the email the invite was sent to.
• We do not link children's data across services, send them behavioral advertising, or enable them to make anything public.
• Parents can review and delete their child's data at any time.

If we materially change this policy we'll bump the version and ask you to re-accept next time you sign in. Non-material changes will post with a new version stamp but won't require re-acceptance.

For any privacy question, data access request, or concern: privacy@quantametron.com.