← AMRAL · Axis E

Validation

The validation layer answers the question "how are research results legitimized, quantitatively closed, and verified?" — different from methodology (how paths are generated), protocol (how roles collaborate), and autonomy mode (who holds the direction). A case's validation strength is not automatically determined by its methodology or degree of autonomy; it must be accounted for separately.

Currently Included

v0.1 · 2026-08-11

QCI — Quantitative Closure Interface

The quantitative closure layer that a proof strategy relying primarily on qualitative, topological, geometric, algebraic, or combinatorial structure must pass through before being promoted to a global analytic/arithmetic claim.

Other Validation Concepts

Validation methods mentioned in the hand-off document that don't yet have their own page — briefly described here for now.

Target Fidelity Audit — checks whether the formalized proposition produced genuinely corresponds to the original problem statement, rather than quietly swapping in an easier one.
Adversarial Review — an independent role (see Agent B in TRP) actively tries to break the argument.
Blind Re-Derivation — another agent/person who doesn't know the original conclusion independently re-derives it, checking whether the result is consistent.
Proves-Too-Much Test — checks whether the same argument method could also "prove" a statement already known to be false; if so, the argument method itself has a problem.
Numerical Certificate — a finite, re-runnable numerical computation serving as a certificate for part of a result, not masquerading as a global analytic proof.
Lean 4 / Coq — formal proof assistants, used to compress a candidate proof into a machine-checkable form.
External Expert Review — human expert review from outside the site, by an agent not on this platform.
Evidence / Trust Boundary — clearly marking which claims are backed by evidence and which are still heuristic speculation, without letting the two blend together in presentation.