Effective 2026-11-14. Security reports: elijah@trackairoi.com.
1. How the platform is built to protect data
Track AI ROI's product is a record that no party can edit. The security design follows from that: isolation between customers and immutability of the record are enforced in the database itself, by row-level security policies, triggers and separate roles, so that a mistake in application code cannot cross a tenant boundary or rewrite a verdict. Everything a customer connects is encrypted under a key the database never holds. The application and the database run in the United States (Oregon) on a managed host; the platform keeps no servers of its own. A daily job re-verifies the whole record and publishes the result at /verification.
2. Measures
- Encryption in transit. Every connection to the platform uses TLS; responses over HTTPS carry a one-year Strict-Transport-Security header. Connections from the application to the database and to every provider use TLS.
- Encryption at rest. The hosting provider encrypts the database and its backups at rest. Connected API keys and source secrets are additionally encrypted with AES-256-GCM under a master key held only by the platform process, never in the database.
- Passwords and sessions. Passwords are stored as salted scrypt hashes, never in clear. Sessions are random tokens in an HttpOnly, SameSite cookie, expire after 30 days, and are revoked on sign-out, password change and account deletion. Viewing or changing a wage requires the password to be re-entered (or a fresh Google or Microsoft sign-in); deleting an account requires the password, or for a sign-in-provider account the email address, to be typed.
- Tenant isolation in the database. Row-level security policies, enforced by PostgreSQL for every query, confine each database role to the rows of its own company or firm. Isolation does not depend on application code being correct.
- Separation of roles. The role that computes verdicts is separate from the roles that read and write workspace data, and is never exposed to a tenant. Verdicts can be written by no other role. Nothing runs as a database superuser.
- Append-only records. Verdicts, committed baselines, confirmed financial events, usage figures and the change log are append-only, enforced by database triggers. Each workspace's change log is a hash chain; every verdict and committed goal is fingerprinted onto it.
- Daily integrity check. A scheduled job re-verifies every chain and every fingerprint each day, records the result, and alerts the operators on a failure. The latest result is published at /verification.
- Hardening headers and rate limits. Every response carries X-Content-Type-Options, X-Frame-Options and Referrer-Policy headers. Sign-in, password reset and webhook endpoints are rate-limited per address or per token.
- Secrets and configuration. Secrets live in the host's environment, never in the repository. A public server refuses to start without a master key or with development sign-in enabled.
- Operator access. Platform operators are named individually in the host's configuration; only those accounts reach the operator desk. Operators access customer data only to run the service, answer a support request or investigate a problem; the same people hold the hosting and database accounts.
- Backups and recovery. The hosting provider keeps automated database backups for at most seven days; customers cannot restore from them themselves. A restore returns the whole database to a point in time, which can bring back data erased inside that window. After a restore the operator re-runs the scheduled workspace erasures and repeats, by hand, the account deletions recorded on the audit trail since the backup; there is no automatic step yet.
- Retention windows. Short-lived data (sessions, sign-in links, import drafts, webhook buffers, assistant transcripts, recordings) is deleted or emptied by a daily job on fixed windows printed in the Privacy Policy.
- Change control. Every change ships through version control with its tests; the schema is applied by numbered, idempotent migrations on every deploy.
3. What the customer is responsible for
- Who holds an account and in what role, and removing people who leave. Administrators see everything in a workspace, wages included.
- The credentials the customer issues: provider keys scoped as narrowly as the provider allows, business-system logins (a JobBOSS² Cloud API user, for example) with read access to the records connected and nothing more, webhook addresses and source tokens given to one system each and rotated if they may have leaked, the Microsoft 365 consent granted by an authorised administrator.
- The machine the JobBOSS Classic agent runs on, if the customer installs it: checking the published SHA-256 before running the script, keeping the machine patched and its administrators trusted (they can read the agent's settings, which Windows encrypts for that machine), limiting the agent's database login to the tables its guide lists, and turning on certificate verification where the SQL Server has a certificate the machine trusts.
- The systems the recorder runs in, and the staff notice the Terms, section 5, require before any per-person feature is switched on.
- Keeping its own copies of anything it needs: the service is a measurement record, not a system of record, and the workspace can be exported from Settings at any time.
- Not putting in a workspace what the service is not designed for: special categories of personal data, government identifiers, card numbers, bank details, prompt or response content.
4. Reporting a vulnerability
Security researchers and customers who find a vulnerability are asked to report it to elijah@trackairoi.com with enough detail to reproduce it. The operator acknowledges a report within two business days, keeps the reporter informed, fixes confirmed issues as a priority, and credits the reporter if they wish once the fix has shipped. No bounty programme is offered at present.
Testing is welcome within these limits, and a reporter who stays within them will not face legal action from the operator for it: test only against accounts and workspaces you created yourself (a trial workspace is enough); do not access, modify or delete another customer's data, and stop and report at once if you reach it by accident; no denial of service, no spam, no social engineering of operators or customers, no physical attacks; do not keep or publish data you obtained; give the operator a reasonable time to fix an issue before any disclosure. Testing the recorder is included on the same terms.
5. What happens after an incident
When the operator learns of a security incident it contains it first (revoking credentials, pausing a feature, isolating a workspace), then establishes what was affected from the request logs, the change log and the hosting provider's records. If personal data was affected, the affected customers' administrators are told without undue delay and within 48 hours of the operator becoming aware, with what is known, what has been done and what they should do, and the information is updated as the investigation proceeds; the Data Processing Addendum, section 8, is the commitment. A failed daily integrity check is treated as an incident until explained. After any incident the operator writes down the cause and the change made so it cannot recur, and shares that with the affected customers on request.
Where a customer's credential has leaked (a webhook address, a source token, a connected key), the customer should rotate or revoke it in Settings and tell the operator; the operator revokes it on the customer's word and helps inspect what arrived through it.
6. Assessments and what is not claimed
The platform holds no SOC 2 report, ISO 27001 certificate or penetration-test attestation yet, and does not claim one. The measures above are what the code and the hosting implement, and a customer may verify them as the DPA, section 10, provides: by questionnaire once a year, and by audit where the law requires it. A third-party assessment will be published here when one is held. The sub-processor list names every provider involved.
See also Terms of Service, Privacy Policy, Data Processing Addendum, Acceptable Use Policy, Sub-processors, Workforce notice and processing summary, how verification works and how adoption is measured. All documents: /legal.