Method: no-tuning derivation, grading, and the magnitude firewall

Each disease is mapped to an R19 bistable switch emerged from its promoter DNA, with stiffness gamma measured, never fitted. The cusp geometry forces the sign of the corrective drive, graded [F]. No dose or effect magnitude is ever emitted (the magnitude firewall), and the whole corpus is seed-fixed and bit-for-bit reproducible.

One pipeline, run identically for every disease: LOCK → Derive → Gate. Nothing is tuned to a desired answer; the corrective direction falls out of measured geometry.

LOCK — the switch is emerged, never fitted

For each disease the proximal-promoter DNA of the lead gene is read, and its measured stiffness γ fixes an R19 bistable double-well. γ is taken from sequence; it is never fitted to make a result come out. This is the no-tuning rule, and it is what makes a recovered match (next chapters) meaningful rather than circular.

Derive — the direction is forced by the cusp

Given the well the disease occupies and the healthy branch, the sign of the corrective drive is not a choice: it is forced by the cusp. That signed direction (e.g. raise the drive / clear the drive) is the output. It is graded [F]. Crucially, the framework asserts the sign and nothing about the size.

Gate — grading and the magnitude firewall

Every statement is graded honestly:

The mapped agents and candidate leads are graded [O]: their direction matches the derived lever, but no dose, efficacy, potency, or safety number is ever emitted. This is the MAGNITUDE_FIREWALL, and a machine check confirms zero magnitude leaks before release.

Reproducibility

The build is deterministic: a fixed seed (SEED = 19) and a 2×SHA-256 hash chain make the whole corpus bit-for-bit reproducible. The register chain-head for this release is fc8598ec13f74a9e; the same inputs always regenerate the same bytes. Reproduction code is on GitHub.