Control-by-control review with evidence, cryptography inventory, findings and actions. Technical audit by the developer of the build; not an independent or legal assessment.
Scope: Tribro-Omics CRM V11 and TribroLIMS V5 as single-file browser applications, their sign-in, storage, backup, audit trail, report sign-off and the CRM ↔ LIMS link. Method: review of the code paths and the automated tests listed in the test report (same date). Evidence references (SEC, LNK, NEG) point to that report.
The controls that can exist in a browser-only application are in place and behaved as designed in every test: 99 of 99 security and link checks passed. The two largest remaining risks are structural, not defects: the initial password is still in use (F1), and nothing is enforced by a server (F2). Twelve findings are listed in section 6: 2 high, 5 medium, 4 low, 1 informational.
| Control | How it works | Evidence | Assessment | Residual risk |
|---|---|---|---|---|
| Sign-in | User name and password checked against a salted PBKDF2-SHA256 hash (310,000 rounds); constant-time comparison. | Wrong password refused; no data readable without it. | Effective | Depends on password strength; the hash is inside the public file. |
| Two-step sign-in | TOTP (RFC 6238), required by policy; 8 one-time recovery codes; a code cannot be used twice. | SEC tests “password alone does not open the app”, “wrong sign-in code refused”, “correct code signs in”, in both applications. | Effective | Checked in the browser, see finding F2. |
| Step-up for sensitive actions | Authenticator code asked again for sign-off, backup export, case deletion, identifiable cohort export, link code change, switching the policy off. | LNK-n.08 and LNK-n.09 in all five rounds; backup and link-code tests. | Effective | None beyond F2. |
| Lock-out | Increasing delay after five wrong passwords. | Earlier build tests; failed attempts are shown in “Recent sign-ins”. | Effective | Counter is stored in the browser and can be cleared by someone with device access. |
| Idle lock | Automatic lock after 5–30 minutes (default 15). | Configuration test; behaviour unchanged from earlier builds. | Effective | A 30-minute setting weakens it. |
| Encryption at rest | All data in one AES-256-GCM vault; key derived from the password, never stored. | Storage inspected after use: no readable case data. | Effective | Unlocked session exposes data in memory. |
| Encrypted backup | AES-256-GCM with a separate passphrase; plain export removed. | Backup file contains no readable data; wrong passphrase refused; restore works. | Effective | Passphrase loss makes the backup useless. |
| Audit trail integrity | Each entry stores SHA-256 of the previous entry and its own content. | Chain verifies; an edited entry is detected; chains still valid after five link rounds. | Effective against casual change | See finding F9. |
| Report integrity | SHA-256 fingerprint at sign-off; released reports locked; amendments are new versions. | Fingerprint verified; altered text detected (V3 workflow test). | Effective | None. |
| Link confidentiality and integrity | AES-256-GCM with a key from the shared link code; message type and sender bound into the encryption; delivery receipts. | Wrong link code, altered package and unknown case all refused; mailbox contains no readable data. | Effective | See finding F5. |
| Role restrictions | Five laboratory roles limit actions; researcher mode in the phenotype module is read-only and de-identified. | Technician cannot sign (V3 workflow test). | Partly effective | See finding F3. |
| Browser hardening | Content-Security-Policy; no third-party scripts; refuses to run inside a frame. | Present in both builds. | Effective | Fonts are loaded from Google, see F10. |
| Consent enforcement | Research exports blocked when research consent is declined or withdrawn. | Phenotype module tests. | Effective | None. |
| Purpose | Algorithm | Key or parameters |
|---|---|---|
| Password check | PBKDF2-HMAC-SHA256 | 310,000 rounds, 16-byte random salt, 256-bit output |
| Data vault | AES-256-GCM | Key from the password (PBKDF2, own salt); 12-byte random IV per save |
| Encrypted backup | AES-256-GCM | Key from the backup passphrase (PBKDF2, 310,000 rounds, random salt) |
| Link packages | AES-256-GCM with associated data (type and sender) | Key from the link code (PBKDF2, 200,000 rounds, fixed application salt) |
| Two-step codes | TOTP, HMAC-SHA1, 6 digits, 30 s | 160-bit random secret kept in the vault; ±1 step tolerance; no re-use |
| Recovery codes | SHA-256 of each code | 8 codes of 10 characters, single use |
| Audit chain, report fingerprint | SHA-256 | Chain over previous hash and entry content; fingerprint over report content |
| Random values | Browser cryptographic generator | Salts, IVs, secrets, identifiers |
All primitives come from the browser’s Web Crypto interface except the synchronous SHA-256 used for the audit chain, which was compared with a reference implementation on seven inputs.
| No. | Severity | Finding | Effect | Recommended action |
|---|---|---|---|---|
| F1 | High | The initial password is the same in the CRM, TribroLIMS and TribroPheno, and its hash is inside the downloadable files. | Anyone who obtains the initial password can open a fresh copy of each application; a weak password could be guessed offline. | Change the password in every application at first use to a long passphrase (20+ characters), different per application. The security check flags this until done. |
| F2 | High | All checks run in the browser. There is no server that enforces them. | Someone who can change the files on the server, or who uses an unlocked device, can bypass sign-in logic for new data or read data in memory. Existing data stays encrypted without the password. | Restrict hosting (FTP) access, enforce HTTPS and the delivered headers, use managed devices with screen lock. For a team, plan the server edition. |
| F3 | Medium | One sign-in per application; staff roles are not separate credentials. | The audit trail records the selected operator, which a signed-in user could switch. | Use one device profile per person where possible; sign-off already needs password and authenticator. Individual accounts need the server edition. |
| F4 | Medium | The applications share the web origin tribro-omics.com with the public website. | A script flaw on any public page (for example the chatbot page) could read or delete the encrypted vaults and the link mailbox. It could not decrypt them, but it could destroy data or show a false sign-in page. | Move the CRM and TribroLIMS to their own sub-domain (for example lab.tribro-omics.com) so that browser storage is isolated from the public site. |
| F5 | Medium | The link uses a shared code kept in both vaults; packages wait in browser storage. | A weak or leaked link code would let someone with access to the browser profile read waiting packages. | Use the generated 15-character code or longer; change it if a device is lost; import packages promptly. |
| F6 | Medium | TribroLIMS/demo.html (the earlier tour) shows its own demo password and stores settings unencrypted. | It holds only demonstration data, but it looks like a sign-in and may confuse users or auditors. | Remove demo.html from the production server, or keep it only on a separate demonstration address. |
| F7 | Medium | All data lives in one browser on one device. | Clearing browser data or losing the device loses everything since the last backup. | Create an encrypted backup after every session; store it on approved secure storage; test a restore. |
| F8 | Low | Unencrypted backup files from older versions can still be imported. | An old plain file could be restored by mistake and such files may exist on disk. | Delete old plain backup files after making an encrypted one. |
| F9 | Low | The audit chain is kept inside the same vault as the data. | A person with the password and technical skill could rebuild the chain after a change. | Download the audit CSV and the security report regularly to independent storage; the hashes then act as an external anchor. |
| F10 | Low | Fonts are loaded from Google Fonts. | Each page load tells Google the user’s IP address, which matters under the GDPR. | Host the three fonts on tribro-omics.com. |
| F11 | Low | Two-step sign-in can be switched to “optional” by a signed-in user (after an authenticator code). | The policy depends on the user keeping it on. | Leave it on “Required”; the security check shows the state. |
| F12 | Info | Not verified in this audit: live HTTPS configuration on STRATO, a penetration test, a physical authenticator app, legal review. | — | Check the response headers after upload; commission a penetration test before clinical use. |
© Tribro-Omics · Security audit — CRM V11 and TribroLIMS V5 · Version 1.0 · 12 November 2026