Legal notice & terms of use
This page sets out how the Tetracta site and its materials may be used, and how we handle your data. It is a plain-language summary and is not legal advice. Last updated: 2026-07-20.
Who we are
Tetracta is the data controller: a stealth deep-tech engineering team based in Türkiye, operating with global reach. During this pre-launch period we remain anonymous; the formal registered entity name and full postal contact details will be published at launch. For any enquiry or data request, email [email protected].
Research results are provisional
All figures, curves and benchmarks on this site and in the downloadable materials are provisional, diagnostic-scale results (sub-Chinchilla, single-seed) shared for evaluation only. They are not production-validated guarantees of performance. We describe measured effects; we do not disclose the underlying mechanism, formula or recipe.
No investment advice
Nothing on this site is investment, financial, legal or tax advice, an offer of securities, or a solicitation. Any quantitative or market-research capability is offered only to licensed institutions on a business-to-business basis; it is never a retail trading signal, recommendation or a promise of returns.
Materials provided “as is”
Downloadable evidence, methodology notes and other materials are provided “as is”, without warranty of any kind. To the maximum extent permitted by law, Tetracta accepts no liability for any loss arising from reliance on them.
Intellectual property
The Tetracta name, logo, site content and materials are the property of the founders / their future company. “Tetracta Rational Attention” is the subject of a method patent application in preparation; the mechanism, kernel and training recipe are protected trade secrets and are never disclosed publicly. You may read and cite the public results; you may not reproduce the site wholesale or imply endorsement without permission.
Privacy notice (KVKK / GDPR)
Analytics. We run privacy-first analytics with no tracking cookies: aggregate page views, time-on-page, which sections are viewed, and approximate country. We do not store your raw IP address and use no cross-site tracking; a daily-rotating, non-reversible token counts a visit and is deleted within 90 days (legal basis: legitimate interest in understanding site usage).
Contact form. If you choose to use the contact form, we store what you submit — name, e-mail, organization, role and message — together with a daily-rotating, non-reversible anti-abuse token (not your IP address, not your browser signature), solely to respond to your enquiry (legal basis: your consent, given on the form; and our legitimate interest in responding). Using the form sets a single first-party, security-only session cookie (CSRF / anti-bot protection — never used for tracking, expires when you close the browser). We keep submissions for up to 12 months after the conversation ends, then delete them. If you prefer, you can always e-mail us instead.
Legal bases. We process contact/application data under GDPR Art. 6(1)(b) (pre-contractual steps taken at your request) and Art. 6(1)(f) (our legitimate interest in responding); KVKK m.5 equivalents apply. No sensitive-category data is processed.
Your rights. You may request access, rectification, erasure, restriction, objection, and data portability for anything you submitted (GDPR Art. 15–22; KVKK m.11): e-mail [email protected] with "data request" and we log it and act within 30 days. We never sell or share your details with third parties. You also have the right to lodge a complaint. As Tetracta is established in Türkiye, our supervisory authority is the KVKK (Kişisel Verileri Koruma Kurumu); if you are in the EU/UK you may also complain to your own national data-protection authority.
Meetings. Secure meetings (by invitation) are peer-to-peer and are not recorded by Tetracta; we process your name and audio/video only to host the call, and signaling metadata auto-expires within minutes.
Model scans (X-Ray for LLMs). When you run a scan, the model artifacts you provide — weight files (safetensors only) and/or tokenizer files, whether by public/private Hugging Face reference or direct upload — are fetched to a single-purpose worker, used only to produce your report, and then wiped, with an Ed25519-signed deletion/provenance attestation issued per job. We do not train on your weights, tokenizer or outputs, and there is no telemetry. Private jobs run exclusively on a wipe-guaranteed cloud worker that is destroyed after the job. The Language Economy panel reads only your model's tokenizer (vocabulary / merge rules, not the trained weights) transiently to compute per-language token counts for the report; nothing from it is retained beyond the report you own. Your reports are yours to delete at any time — the dashboard Delete action erases the report, its metrics and its records immediately and irreversibly, with no copy kept. Legal basis: GDPR Art. 6(1)(b) (performing the scan you requested) and Art. 6(1)(f) (our legitimate interest in operating and securing the service); KVKK m.5 equivalents apply. Full technical policy: Model Handling & Security.
Confidential / NDA materials
The technical brief, raw-log evidence chain and team introductions are shared only under a non-disclosure agreement. No mechanism or recipe is shared before an NDA.