
Run the audit yourself
Our proof, packaged. Verify it, reproduce it, run it on your data — yourself.
A signed evidence bundle and an open benchmark harness. Check the integrity and the signature, reproduce the published numbers on public data, then run the same harness against your own dataset on your own machines. Ryedore does nothing on your behalf — and nothing here is a pilot or a trial.
The whole audit — four offline steps, nothing sent to us
From the signed bundle to your own dataset. Each step is detailed below; none of them needs a conversation with Ryedore.
swipe → to see the whole diagram
Verify, reproduce, then run it on your data
In the order a sceptical engineer would do it.
1 · Integrity
Download the evidence pack and run sha256sum -c SHA256SUMS. Every file must match its recorded hash — any machine, no dependencies.
2 · Authenticity
The whole bundle is signed with RSA‑PSS/SHA‑256 under the public key shipped inside it. Run verify.sh (or the Python one‑liner in VERIFICATION.md): the manifest hash, every file and the signature must all come back true. Each per‑task report is signed independently too.
3 · Reproduce the numbers
The benchmark harness is open (Apache‑2.0), isolated, uses public datasets only, and reports multi‑seed mean ± std — no single‑seed cherry‑picking. Provenance, Python version and pinned dependencies are in environment.json. Run it; get the same numbers, or the signature breaks.
4 · Run it against your own data
The same harness runs on your machines against your own dataset: point its loader at your data, keep the same metrics and seeds, and compare. Nothing is sent to Ryedore; nobody from Ryedore touches your data. What you see is what it does.
Don’t take our word. Take the bundle.
We publish our losses on public data for the same reason we hand you the harness: a result you can’t re‑run is marketing. signed · re-runnable
The four commands, with what you should see
Same steps as above, as a walkthrough: copy each command, run it on your machine, compare the output.
1 · Download and unpack
What it proves: You are working from the same bytes we published.
curl -O https://ryedore.com/verify/audit-bundle-2026-09-01.zip unzip -q audit-bundle-2026-09-01.zip -d ryedore-audit && cd ryedore-audit
You should see: A folder with audit_report.html, audit_report.pdf, evidence/, manifest.json, manifest.sig, license_public.pem, SHA256SUMS, verify.sh.
Six straight answers
- Is this a pilot or a free trial?
- No. Ryedore does not run pilots or trials. This page gives you the signed evidence and the open harness so you can check our published numbers and run the same tests on your own data, yourself. Commercial engagements start from the published price list.
- What do I need?
- Any machine for the integrity check; Python with the cryptography package for the signature check; the public datasets (NASA C‑MAPSS, UCI Hydraulic) and the harness for a reproduction; your own data in the harness’s loader format for a run on your data.
- Will Ryedore help me run it?
- The bundle’s VERIFICATION.md and the harness README are written so you don’t need us — VERIFICATION.md is also readable on the web at /audit/docs/ before you download anything. Questions about the method are welcome on the contact page; we won’t ask for your data.
- Where do the numbers on this site come from?
- Every Ryedore figure on the site is read from a number registry whose signed entries trace to this bundle. Comparison figures are quoted from the named publications for disclosure.
- How long does it take?
- Integrity and authenticity: minutes — two commands on any machine. Reproducing the published numbers: longer, and it depends on your hardware, because the harness trains and evaluates multi‑seed on the public datasets; VERIFICATION.md states the pinned environment. Running on your own data takes as long as your dataset is large.
- Can we share the bundle inside our company?
- Yes. It is public, unsigned‑for‑nobody‑in‑particular: no NDA, no registration, no licence key. Forward the zip; anyone with it can verify it independently, which is the point.
After the audit
Ready to talk numbers — yours?
Pricing is public. When your own run convinces you, the price list is where the conversation starts.
