The constructions, the filings, the KEM methodology, and a source for every public figure. Papers and reductions sit under NDA.
June 2026 is the first. Programmable cryptography is next. Both sit under NDA. Request, then download.
White paper 01 · June 2026 · Under NDA
The first paper in the series: a context-aware KEM on LWE, a variable-size object cipher, and a programmable handshake. Available under NDA. Current builds supersede the early figures in this draft.
White paper 02 · Forthcoming · Forthcoming
How the handshake carries logic. Policy as a field of the connection, not a sidecar. The next paper in the series. Available under NDA when it ships.
The filings cover the constructions, the programmable handshake, the object cipher, and the factory that configures them. Specifications and security reductions available under NDA.
We do not quote a millisecond table. Early reference figures in the June 2026 paper are superseded. The public floor on this site is qualitative, at matched security, not 512 versus 768.
Public floor
Security
matched level
matched level
Public key
ML-KEM baseline
about half
Compute
ML-KEM baseline
at or below
The June 2026 paper reported an unoptimized C reference at a single security level. That draft is old relative to the current construction. Treat it as a research note, not a production claim.
A hostile reading of an early millisecond table would attack the baseline, not the mathematics. We would rather ship no exact than a number a reviewer can break.
Superscripts on the site point here. If a number cannot be sourced, it does not ship.
Note 1
A compromised identity starts 79 percent of attacks. Recovery costs $375K when backups survive and $3M when they do not.
Sophos Active Adversary reporting puts compromised identity at the start of 79 percent of attacks. Verizon's DBIR is a different series: 88 percent of basic web-application attacks in the 2025 DBIR involved stolen credentials. Recovery figures are from Sophos State of Ransomware 2024: median $375,000 with intact backups, $3 million when backups are compromised.
Note 2
The estimate for breaking RSA-2048 fell from a million qubits to about a hundred thousand in ten months.
Gidney (May 2025) showed how to factor RSA-2048 with fewer than a million noisy qubits. ON01 cites Webster et al., Pinnacle (February 2026), for a subsequent physical-qubit estimate near 100,000 using quantum LDPC codes. The two papers are ten months apart.
KEM figures: see methodology. ML-KEM is FIPS 203, Module-LWE, published reduction. ONE KEM is LWE. Reduction available under NDA.
Invite-only sandboxes. Named people and systems. Papers unlock after a request.