Privacy policy

Scaffolding is an editorial AI tool that critiques college applications. It is operated as a solo project; there is no parent company, holding company, or third-party controller behind it. References to “we”, “us”, and “Scaffolding” mean the same single operator.

Scaffolding is for adults. You must be 18 years of age or older to create an account, and we ask you to confirm this at signup. We do not knowingly collect personal information from anyone under 18. If we learn that an account holder is under 18, we close the account and delete the associated data.

We collect three categories of data:

• Account data — email, password hash, your adult-age confirmation, the consents you granted at signup, login timestamps.
• Application data you upload — essays, activities, transcripts, supplementals, the schools you target. We need these to run the committee.
• Operational telemetry — error reports (via Sentry, scrubbed of PII before send), aggregate latency and cost stats, audit-log rows of every privileged data access we perform internally.

• To run the committee of agents that produce your First Read and revisions.
• To serve the workspace UI and keep your draft history.
• To bill you (via Stripe) and remember your subscription state.
• To send you operational email (report-ready notices, payment receipts).
• To send you occasional email about getting the most out of Scaffolding — a welcome note, a nudge if you stall, a heads-up when your free thoughts run low. Every one of these has a one-click unsubscribe; operational email is separate and always sends.
• To detect product regressions (anomaly detection, eval gates).
• To train and improve our own AI models, unless you opt out — see Training our models below.

We do not sell your data. We do not show targeted advertising. We do not build an advertising profile of you.

Scaffolding trains its own AI models on the application work people submit — essays, supplemental answers, and profile fields — so the agents become better readers over time. This section explains exactly what that means.

• What is used. Application text you create in the workspace on or after 2026-09-02. Anything you submitted before that date is never used for training, regardless of your opt-out state. Account data, payment records, and support messages are never used for training.
• Raw text. Training uses your writing as you wrote it — we do not require it to be de-identified first. It stays inside Scaffolding.
• Our models only. We do not sell training data, we do not license it to other companies, and we do not send it to our large-language-model vendors for their training. The models we fine-tune are used only to run Scaffolding.
• This is opt-out. By default, application text created on or after the effective date is eligible for the training set. You can opt out at signup, or at any later time from /account/consent. Opting out stops any new work from entering the training set going forward.
• Opt-out is not retroactive. Once a model has been trained on a piece of text, that learning cannot be isolated and removed from the model. Opting out prevents future use; it cannot undo training that already happened.

Sensitive content.College essays often disclose sensitive personal information — health, religion, ethnicity, immigration status, sexual orientation. Some state privacy laws treat processing sensitive data for a secondary purpose such as model training as something that should be opt-in rather than opt-out. Whether Scaffolding's opt-out approach is sufficient for this category is an open question flagged for the legal review this document is pending. If that review concludes opt-in consent is required, we will switch to opt-in for the affected data and re-ask before using it.

We use the following processors. Each receives only the data necessary for its function. We have data-processing agreements where vendors offer them.

• Supabase — our database and auth provider. Stores account, application, and consent data with row-level security.
• Vercel — application hosting and edge runtime.
• Inngest — background job runner for committee orchestration and retention sweeps.
• Anthropic — the large-language-model provider that runs the agents. Your essay text passes through Anthropic when an agent runs. Anthropic contractually does not train on API traffic. Our own fine-tuning, described above, is separate from Anthropic and stays inside Scaffolding.
• OpenAI — outage fallback for the agents only. If Anthropic is unavailable mid-run, the request retries against OpenAI so your report still completes; in normal operation no traffic goes here. OpenAI contractually does not train on API traffic.
• Stripe — payment processing.
• Resend — transactional email.
• Upstash Redis — rate-limit counters and short-lived caches.
• Sentry — error monitoring with PII scrubbing applied before send.
• Plausible — privacy-respecting site analytics (no cookies, no cross-site tracking).

• Account data — retained while your account is active. After deletion, the account row is soft-deleted for 30 days (so accidental clicks are reversible) and then hard-deleted by the daily retention cron.
• Essays and supplementals — retained while your account is active, and deleted on account deletion.
• Training set — application text collected for training (from accounts that have not opted out) is retained for as long as we maintain the model generations trained on it. Deleting your account removes your text from the training set going forward; it does not retrain existing models.
• Audit log (records of our internal privileged access) — retained for two years, then hard-deleted by the daily retention cron.
• Agent output logs — retained for the duration required to investigate quality regressions (typically 90 days), then hard-deleted.

• Access. You can export everything we hold about you from /account/data.
• Deletion. You can delete your account from /account. Deletion is hard-deleted after a 30-day grace window.
• Training opt-out. You can stop your submitted work from being used to train our models at any time from /account/consent. The opt-out applies going forward.
• Consent revocation. You can revoke any consent you granted at signup from /account/consent. Revoked consents stop applying immediately; future processing of your data respects the revocation.
• Correction. Email support@scaffolding.app and we will fix factual errors in your account record.

Scaffolding uses first-party cookies for authentication only. We do not use advertising cookies, retargeting pixels, or cross-site trackers. Plausible analytics is cookieless by design.

Material changes to this policy bump the consent version on the affected categories, which means we ask you to re-consent before continuing to use the product. Minor wording fixes do not bump the version.

Email privacy@scaffolding.app for any privacy question. For accessibility issues, email support@scaffolding.app.