The operator.

Modelometer is operated by Manj Chenna.

Modelometer is currently operated by Manj Chenna as a natural person, in personal capacity. A registered entity is being formed. This page will name it on the day it exists.

Amsterdam, the Netherlands. The measurement runs against publicly available commercial model endpoints.

The measurements, pages and reports are produced by an automated pipeline running a frozen protocol: fixed items, fixed thresholds, deterministic scoring, and no model judging another. Manj Chenna operates it, holds every key, approves anything that pairs a vendor with a judgment, and answers for what is published. Manj Chenna is building three products around one question: how humans keep authority over AI systems. Modelometer measures what the models do; Sanctity and HumanChain are the other two. His site is manjchenna.com.

Every run is written to an append-only chain. Published runs carry their own timestamps and are independently verifiable from the public chain. Runs are never edited after sealing; corrections are issued as new entries and the superseded entry stays visible.

The chain is append-only and independently verifiable, so the record can be checked by anyone holding a copy, without Modelometer. A formal succession arrangement, including a second custodian and key escrow, is not yet in place. It is being arranged and this page will state it when it is. Until then, readers relying on the continuity of this record should treat that as an open risk.

Modelometer is owned and run by Manj Chenna, who also builds Sanctity, which publishes opinion and advisory work on AI governance. That relationship is stated here openly: the two are kept separate in what they are permitted to say. Sanctity holds opinions; Modelometer publishes measured differences and does not interpret them into recommendations. Sanctity is a brand and a body of work, not a registered company; nothing on this page implies an entity that does not exist.

Modelometer does not accept payment from any organisation whose models it measures, in any form, for any purpose. No measured provider can buy coverage, buy the wording of a finding, buy an interpretation, or buy a delay. This is stated as a rule and not as an intention: if it is ever breached, the breach is published here.

A contact submission carries seven form fields, each capped at four thousand characters. It is written first to the bureau's own append-only store, so a copy is retained whether or not delivery succeeds, and that retention is deliberate. It is then delivered over a single authenticated encrypted connection to one mailbox the operator controls. No third-party service handles the content, no account is created for the sender, and if delivery fails, only the class of the error is logged, never the message itself.

Modelometer is not an accreditation body and not an auditor in any regulated sense. It publishes measured differences in model behaviour over time. It does not approve a model or state that a model is safe, compliant, or fit for any particular purpose. This page carries no founding date; the chain publishes its own timestamps and the record's age is verifiable there rather than asserted here.