A local cryptography startup announced Thursday that it has developed what it calls the industry's first encryption algorithm that is “constant time on average,” a breakthrough engineers describe as “operationally indistinguishable from secure.”

“Classical constant-time implementations guarantee identical runtimes for all inputs,” said the company’s lead cryptographer during a press briefing conducted in a reassuring tone. “Our model reflects real-world usage. While individual executions may vary with the secret key, ciphertext, microarchitecture, and mood of the branch predictor, the mean runtime remains beautifully stable.”

The team’s preprint, reviewed by The Constant Times, formalizes this claim under what it calls a “distributionally calm adversary.” Under this model, timing variance is permitted so long as it integrates to zero over a sufficiently polite sample size.

“In expectation, no bits are revealed,” the paper states, clarifying that adversaries observing fewer than ten million traces are “unlikely to detect meaningful structure without excessive commitment.”

Outside experts expressed cautious admiration. “It is certainly true that the average runtime is constant,” noted one professor of applied cryptography. “However, attackers tend to condition on inputs rather than average across them.”

The startup disputes this characterization. “Our threat model assumes adversaries drawn from a smooth distribution,” said the CTO. “Highly adversarial adversaries are, by definition, out of scope.”

An appendix addresses so-called “statistically unpopular keys,” which may induce slightly longer runtimes. These keys are described as rare under “normal, well-adjusted deployments,” and are not expected to occur in practice unless deliberately selected.

Concerns about cache-based side channels were resolved in a footnote specifying that the implementation is constant time “with respect to the primary channel,” and that cache effects should be regarded as “adjacent phenomena.”

When asked about power analysis, engineers emphasized that while power is a continuous signal, their guarantees are discrete. “We provide binary assurances,” one developer explained. “Either the average is constant, or it isn’t.”

At press time, the authors had introduced a new complexity class, O(1̄), defined as “constant under charitable interpretation.” The notation is accompanied by several centered graphs and a firm commitment to composure.