Mean of a small declared dataset
A deliberately small example with a manifest, CSV input, Python computation, exact expected value, receipt, paper, and reproduction instructions. It teaches the format; it is not a scientific validation benchmark.
public and inspectable
Run the evidence before relying on the explanation.
Both examples expose their receipt and source path. The research starter demonstrates the mechanics on a tiny open dataset; the release example shows how Certiv exercises its own failure modes.
Certiv Core release checks
Executable claims cover pack/check/tamper behavior, nondeterminism refusal, receipt field boundaries, and other release invariants. These are author-local checks, not an independent audit.
Inspect
Read the manifest, declared inputs, exact expectation, receipt fields, and evidence boundary.
Re-run
Use a disposable environment, verify downloaded hashes, then execute the documented command.
Try to break it
Change an input, command, expectation, or receipt field and confirm the discrepancy is visible or refused.
Never run an untrusted example on your everyday machine.
Author commands can execute arbitrary code with the current user’s access. Use a disposable VM with no credentials, valuable files, or unnecessary network access.