Reproduce It Yourself

Everything in the Lab ran on a single machine. You can reproduce it on yours. The composition deploys the same way on any x86_64 Linux with at least 16 GB RAM.

Prerequisites

• Linux (tested on Pop!_OS 22.04 / Ubuntu 22.04)
• Rust toolchain (rustup — installs in 2 minutes)
• 16 GB RAM minimum (96 GB recommended for full NCBI data)
• Git and basic build tools (build-essential)

Optional:

• Vulkan-capable GPU for 🐟⚡ barraCuda GPU workloads
• Python 3.10+ and R 4.x for baseline comparison pipelines

Step 1: Get the Primal Binaries

# Clone plasmidBin (pre-built binaries)
git clone [email protected]:ecoPrimals/plasmidBin.git
export PLASMIDBIN="$(pwd)/plasmidBin"

# Or build from source (springs are all public AGPL)
git clone https://github.com/syntheticChemistry/wetSpring
cd wetSpring && cargo build --release --workspace

Step 2: Deploy the Composition

git clone [email protected]:sporeGarden/projectNUCLEUS.git
cd projectNUCLEUS/deploy

# Deploy full NUCLEUS (13 primals) to the current machine
bash deploy.sh --composition full --gate mygate

# Verify all primals are healthy
bash deploy.sh --health-check

deploy.sh handles seed creation, primal startup ordering, health verification, and port allocation. Primals bind to 127.0.0.1 by default.

Step 3: Run the Science Workloads

# Run a single workload
toadstool execute ../workloads/wetspring/wetspring-16s-rust-validation.toml

# Run the full provenance pipeline (all workloads + DAG + ledger + braid)
bash provenance_pipeline.sh \
    --workloads-dir ../workloads/wetspring \
    --session-name "my-validation-run"

Expected Results

Total: 235+ checks, all at tol=0.000000 (exact Python→Rust parity).

Step 4: Verify the Provenance Chain

After the pipeline completes, you have:

# Check the Merkle root (content hash of all DAG events)
cat results/PROVENANCE_MANIFEST.md | grep "Merkle Root"

# Query the loamSpine ledger
curl -s -X POST http://localhost:9700 \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","method":"spine.list","params":{},"id":1}'

# Query the sweetGrass braid
curl -s -X POST http://localhost:9850/jsonrpc \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","method":"braid.list","params":{},"id":1}'

The braid carries an ed25519 witness signature from 🐻🐕 BearDog’s key hierarchy. The Merkle root covers all data registrations and workload results in one integrity proof. Tamper with one byte and the chain breaks.

Step 5: Compare Your Results

If your workload output BLAKE3 hashes match the published hashes, the science is bit-for-bit reproduced. The Merkle root and braid URN will differ (expected — they include run-specific session IDs and timestamps), but the per-workload output hashes should be identical.

# Hash a workload output
b3sum results/wetspring-16s-rust-validation.stdout

What If Something Doesn’t Match?

File an issue or send a gap report. That’s the point — the methodology is falsifiable. If the results diverge on your hardware, that’s signal, not failure. Document the divergence, the hardware, and the environment. The gap report flows upstream through 💧🕳️ wateringHole and improves the ecosystem for everyone.

Data Dependencies

For the full NCBI pipeline (real data, not synthetic):

# Download real NCBI data (requires ~5 GB disk)
# PRJNA488170: Nannochloropsis outdoor 16S (Wageningen)
prefetch SRR7760408 && fasterq-dump SRR7760408

All synthetic workloads run without external data downloads.

Hardware Baselines

Your times will vary. The checks should not.