The Pilot Program

Request a hardware-bound binary for technical evaluation.

Your float32 data compresses. SPCMP also gives it a Merkle commitment: the same size reduction, plus tamper-evidence. Four commands to verify.

Before You Read the Terms

Your security team will review this program. Their review is better spent on evidence than on my pitch. Start with the divergence results.

Already reviewed

Have you gone through the divergence results?

Then I want your read before any paperwork. Objections, reproductions that disagree with mine, and a fourth result are all useful. Send them to me directly.

  • Environments you ran, with CPU and Python versions
  • The hex view from each
  • What your team challenged or could not reproduce
Email Your Thoughts
Not yet

Run it first. It takes a minute.

One Python file, run on two different machines your team already has. Compare the hex view against the record. If your result matches nothing I have published, that is the more interesting outcome.

  • Pick two machines that are actually different (e.g., an Intel laptop and an AMD cloud VM)
  • Compare the hex view against the record
  • Come back and tell me what you saw
Run the Script Download

If your test machines return the same hash, they behaved identically on this broad test. For path-dependent divergence on GPUs, see the GPU harness. Full record.

The test & CAGM

A matching loss is not proof. A matching hash is proof.

Same script, same seed (42), four different environments. The final loss was identical on all four. The model weights were completely different.

CAGM is a fixed-point ML classifier that removes floating-point math from the training loop and emits a SHA-256 proof of the entire training trajectory. Run across the same four environments, every run produced the exact same proof.

  • Run the 40-line PyTorch script on your own machines
  • Compare your weights SHA-256 against the four recorded
  • CAGM pilot tier is listed under the Evaluation Tiers below
Run the Script Download

torch.use_deterministic_algorithms(True) only guarantees run-to-run repeatability on the same machine. It does not fix cross-hardware drift. Full results.

Forward this page to whoever signs off on security. The terms below (hardware binding, online verification, pricing) are written for them to read after the results, not before.

1. What You Get

The pilot is a fully functional, cryptographically bound evaluation of the Samayuktam stack. It is not a SaaS subscription. It is a local binary that proves the architecture works on your specific hardware.

2. The Evaluation Tiers

SPCMP (Float32 Container)
$2,500
10 runs · 30 days

Full canonical-ordering, compression, and fail-closed verification pipeline.

Request SPCMP
CAGM (Deterministic ML)
$2,000
10 runs · 30 days

Fixed-point classifier with full Integrity Engine and trajectory-proof output.

Request CAGM
The Bundle (Full Stack)
$3,500 (Save $1,000)
10 runs each · 30 days

Both binaries, both fingerprint-bound.

Request Bundle

All prices are visible without contact. No asterisks. A custom trial of 5 uses / 1 day may be granted at no cost, at sole discretion, via email request.

Production & Enterprise Licensing

The pilot is designed for isolated evaluation. For teams integrating SPCMP or CAGM into live production pipelines, I offer custom licensing tiers with higher usage limits, dedicated support, and integration assistance.

Standard Production
Let’s discuss your volume
Up to 5,000 runs / month
  • Support: standard email support
  • Response target: within 48 hours
Enterprise Production
Custom annual contract
Above 5,000 runs / month (custom volume)
  • Support: direct email and debug responses, answered within one business day
  • Integration: I will help fit the product into your specific pipeline (CI/CD, MLOps, or data lake)

Note on infrastructure: Production licenses require a static, hardware-bound environment. For cloud deployments, this means configuring a static MAC address on your instances. Additionally, the binary requires outbound HTTPS access to the Samayuktam licensing server (drm-license-worker.samayuktam.workers.dev) on every run; there is no offline mode. If the server is unreachable, the run fails.

Note on support: Samayuktam is a solo-founder operation. You will not be routed to a generic support tier. You will be talking directly to the engineer who built the protocol.

Request Enterprise Consultation

3. How to Get Bound (The Handshake)

To request a pilot, you do not fill out a form. You email me the hardware fingerprint of the machine you intend to test.

hardware fingerprint · run all four
# 01 · Machine ID
cat /etc/machine-id
# Persistent OS-level machine identifier.

# 02 · MAC address
ip link show | grep "link/ether" | awk '{print $2}' | head -1
# Primary NIC hardware address.

# 03 · CPU model
grep "model name" /proc/cpuinfo | head -1 | cut -d: -f2 | xargs
# CPU model name from /proc/cpuinfo.

# 04 · CPU physical ID
grep "physical id" /proc/cpuinfo | head -1 | cut -d: -f2 | xargs
# Physical socket ID from /proc/cpuinfo.

Why these four values?

When security teams hear "hardware binding," they worry I am asking for sensitive machine secrets or TPM hashes. I’m not.

The four commands above (/etc/machine-id, MAC address, CPU model, physical ID) read standard system identifiers, not secrets. I do not read your files, your environment variables, or your private keys.

Think of it like a shipping address. I need to know which machine to route the license to so that no one else can use it. You are giving me the ZIP code and the street name, not the keys to the house.

Send all four values to: [email protected]

Setup

Environment Configuration

The binary binds to the MAC address and Machine ID of the host.

  • Supported
    WSL2: Supported with a static MAC. Under [wsl2] in .wslconfig, set networkingMode=bridged, vmSwitch, and macAddress.
  • Single session
    Google Colab: Supported for single-session evaluation. Runtimes are reassigned, so a Colab binding does not persist across sessions.
  • Supported
    Enterprise VMs (VMware, AWS EC2, GCP): You must configure a static MAC address for the instance.
  • Not supported
    Snapshot/Revert VMs: Reverting a snapshot breaks the hardware fingerprint by design.
Trust

Privacy & Storage

These four values are sent once to generate your license. My server hashes them immediately into a single cryptographic binding key.

I do not store the raw MAC address, Machine ID, or CPU serials in my database as plaintext. Once your license is issued, the original email containing the raw hardware identifiers is permanently deleted from my inbox. If my database were ever breached, there is no readable hardware data to steal.

Licenses are issued manually by a single operator, with no automated pipeline in the path.

4. Technical Constraints (Read Before Applying)

Ready to start the handshake?

Email Hardware Fingerprint