← Research

Research

Provenance over persuasion

June 28, 2026

Every score in a Kairos design report is a deterministic computation you can re-run over the printed sequence. The ranking is multi-objective — never one fragile aggregate.

Most design tools report a single score and ask you to trust it. Kairos reports how the score was computed. The distinction matters: a number you cannot reproduce is an assertion; a number you can re-derive is evidence.

This is the principle behind provenance, not persuasion — the standard Kairos is built to. It shows up in three places.

1. Deterministic, re-runnable scores

Every score on every candidate is a deterministic computation. Given the printed coding sequence, you can re-run the scoring yourself and get the same numbers Kairos reports. There is no model-in-the-loop adjusting the score — the axes (manufacturability, host-likeness, 5′ initiation, protein developability) are computed by explicit, inspectable procedures.

2. Multi-objective ranking, not one aggregate

The panel is ranked by Pareto dominance across all four axes — not collapsed into a single weighted sum. Why? Because a single score hides trade-offs. Candidate A may be better on manufacturability; candidate B on host-likeness. Pareto ranking surfaces both as non-dominated, and lets you choose based on which trade-off you accept.

Each candidate's rank comes with a per-axis explanation: which axes favored it, which flagged it, and what those flags mean. The verdict is a band (strong / moderate / weak) plus a pass/flag count — qualitative, never a number without context.

3. Flags that argue against the winner

A design report that hides the loser's flaws is persuasion. Kairos surfaces risk flags on the recommended candidate. In the public example run on Cytochrome c (UniProt Q6Q4H8), the top candidate carries an extreme_pi flag — and the report says so. That flag is an intrinsic property of cytochrome c (a highly basic protein), not a design defect. Naming it is the honest thing to do.

Why this is the brand

“Provenance, not persuasion” is not a slogan. It is a design constraint on what Kairos is allowed to do: it may compute, verify, and explain — it may not assert, obscure, or auto-win. The sequence is withheld in the public report (available to invited collaborators), but the scorecard is fully visible, because the scorecard is the product.

References

The example protein, ranking method, and codon-bias baseline cited in this article are real, traceable sources.

  1. UniProt Q6Q4H8 — cytochrome c (Komagataella phaffii). UniProt:Q6Q4H8
  2. Sharp PM, Li WH. The codon Adaptation Index — a measure of directional synonymous codon usage bias. Nucleic Acids Res, 1987. PMID:3547335
  3. Gustafsson C, et al. Codon bias and heterologous protein expression. Trends Biotechnol, 2004. PMID:15245907
Scope. Kairos scores dry-lab design quality with deterministic, re-runnable computations. It does not predict expression or titer. Absolute benchmark numbers are not published — only qualitative verdicts and pass/flag counts.