Security

Every TVCRpro security claim is enforced by infrastructure, not by policy. If a claim is not on this page, it is not made anywhere.

Security Brief

Two PDF briefs are available for download. Both are version-stamped, source the same evidence, and use the same sub-processor table reproduced below. Author and contact details are on the cover page of each PDF.

Sub-processors and infrastructure

Operator-pasted text submitted to the analyzer at /analyzer leaves the TVCRpro environment only to reach Anthropic for the rubric pass. The composite weighting that produces the Token Value Conversion Ratio is applied on a TVCRpro-controlled server after the rubric agent returns; those weights are never transmitted to the inference vendor and never shipped to the operator's browser. See /methodology for the full architectural split.

Function Provider Jurisdiction Verification
Email, calendar, contacts, encrypted mail Proton AG Switzerland Proton Trust Center
Domain DNS, edge / proxy Cloudflare U.S. (multi-region) Cloudflare Trust Hub
Document workflow / e-signature Adobe Inc. (Acrobat Sign) U.S. Adobe Trust Center
Productivity suite, workflow automation, BI Microsoft 365 (Canada tenant) Canada Microsoft Trust Center
Inference for the eleven-dimension rubric agent at /analyzer Anthropic, PBC U.S. Anthropic Trust Center
Custom domain tvcrpro.com on Proton Unlimited MX, SPF, DKIM, DMARC, catch-all all green (April 24, 2026).

Encryption

Data retention

Tier 1 — anonymous Light Analyzer

Tier 1 inputs are not retained. The claim is enforced by the absence of a database in the Light Analyzer architecture — there is no datastore that could retain Tier 1 inputs. The architectural absence is the enforcement, not a written policy.

Tier 2 and Tier 3 — NDA-gated reports and pilots

Retention is governed by the executed NDA and pilot agreement on a per-engagement basis. The default posture is minimum-necessary retention with documented deletion-on-withdrawal via a Power Automate workflow.

Deletion-on-withdrawal

A withdrawal request triggers a documented Power Automate workflow; a logged deletion confirmation is the verification artifact. Workflow documentation is internal; verification of execution is provided to any party that requests it under NDA.

Data residency

Data typeResident inProvider
EmailSwitzerlandProton AG
Workspace documentsCanadaMicrosoft 365 (Canada tenant)
E-signature documentsU.S.Adobe Inc.
DNS / edgeMulti-regionCloudflare

Access control

“No AI training” claim

TVCRpro’s scoring engine does not train any model on user data. The claim is enforced by the engine’s deterministic algorithmic design — there is nothing in the codebase that could train, even if the inventor wanted to. No training pipeline exists.

What TVCRpro does not collect

The website itself runs on first-party server logs only. No third-party analytics SDK is loaded. No client-side error tracking SDK. No A/B testing framework. No heatmap or session replay. No cookies except an optional theme-preference cookie (deferred post-v31).

The Do-Not-Track header is honored at the server log level (logs strip user agent if DNT is present), and by the Light Analyzer API per its locked specification.

What is not yet in place

SOC 2, ISO 27001, and a third-party security audit are not pursued at v31 launch. Pilot data does not yet exist. The full Limitations page enumerates this in detail. Pre-pilot, design-intent.