The Proof Layer
How a sealed result sits alongside your AI rather than in front of it, and how an orchestrator running many agents leaves a chain a reviewer can walk.
Read the proof layer →For reviewers and regulators
Reactor-physics evidence · one file, checked offline
Your team computes a reactor-physics number in whatever code it already uses. SMRFORGE seals that result, the inputs that fed it, and the provenance that produced it, into one SHA-256 manifest with one Ed25519 signature. The bundle is self-contained, so anyone downstream can confirm it offline and re-derive it, without trusting the model, the vendor, or us.
Not “trust us.” Check us.
The seal proves the number is unaltered and re-derives from its own recorded inputs. Fidelity is labeled on every result, and every result is advisory.
Why a reviewer wants it
A reviewer has to defend a number they did not produce. The seal delivers each of these. The fourth is the one the person signing cares about most, so it carries the warm mark.
Verify is a hash check and a signature check. It runs offline on the reviewer’s own machine, with no cluster and no re-run of the pipeline. If anything inside the bundle changed, the checker names the section.
The expensive Monte Carlo runs once. Downstream, a reviewer re-runs only the cheap deterministic step on the sealed upstream constants, which were already verified. A laptop does it.
The tolerance is committed before the run, not fitted after it. The seal covers the fingerprints of every input and output, so a doctored value is caught while the signature stays valid. Nothing rests on the producer’s word.
The reviewer’s own validated codes and V&V still produce the number. SMRFORGE adds a checkable record beside the existing process and never becomes the authority over it. Adopting it subtracts no rigor.
The bundle verifies offline years later with just the file and the public key. It outlives the author, the cluster, and the tool that made it.
Go deeper
Three pointers. None of them asks you to trust us first.
How a sealed result sits alongside your AI rather than in front of it, and how an orchestrator running many agents leaves a chain a reviewer can walk.
Read the proof layer →The record holds more than k_eff. See what a sealed result carries, and which shapes of number the seal covers today.
See the sealed results →Verify a real sealed result in your browser, then tamper with one byte and watch the seal catch the change. No install, no server, no SMRFORGE code.
Check one now →Check us
Founder-led, and reachable. Download the two-page brief for the people who have to defend the number, or skip the brief and verify a sealed result in your browser in about a minute. Questions, bundles you want checked, and review requests go to support@smrforge.io.