Make targets

The repository splits its targets into two families: solver-* drives the Rust optimiser, research-* drives the Python validation and analysis tooling.

Solver targets (Rust)

Target	Expands to
solver-build	cargo build
solver-test	cargo test
solver-clippy	cargo clippy
solver-lint	cargo clippy -- -D warnings && cargo fmt --check
solver-fmt	cargo fmt
solver-homelab	plan against the bundled homelab dataset
solver-irina	plan against the hyperscale irina dataset
solver-irina-pareto-80	as solver-irina, with --pareto 80

All of them run from the solver/ directory.

solver-homelab

cd solver && cargo run --release -- plan \
  --infra testdata/homelab_infra.yaml \
  --workloads testdata/homelab_workloads.yaml

Small dataset, near-instant. The right target for a first run or a quick sanity check after a change.

solver-irina

cd solver && cargo run --release -- plan \
  --infra testdata/irina_infra.yaml \
  --workloads testdata/irina_workloads.yaml

The datacenter-scale benchmark — 186 nodes, 2,714 pods. Triggers the large-tier GA configuration.

solver-irina-pareto-80

The same as above with --pareto 80, solving against 80% of node capacity to leave operational headroom.

Research targets (Python)

These use uv to manage dependencies, so uv must be installed.

Target	Purpose
research-generate-testdata	Generate the irina 8-dimensional test dataset
research-homelab	Delegate to research/Makefile’s immediate_run
research-irina	Delegate to research/Makefile’s irina_stress
research-inspect	Validate a solution and render the heatmap dashboard
research-full-pipeline	Generate → solve → inspect → prove, end to end

research-inspect

uv run --with pyyaml python research/inspector.py \
  --infra solver/testdata/irina_infra.yaml \
  --workloads solver/testdata/irina_workloads.yaml \
  --solution solver/kuberina_solution.yaml \
  --output kuberina_dashboard.html

Produces kuberina_dashboard.html, an interactive cluster heatmap.

README discrepancy

The project README calls this step make solver-inspect. No such target exists — the correct name is research-inspect.

research-full-pipeline

The complete validation chain, in four announced stages:

1. Generate 8D test data — research/gen_irina_testdata.py
2. Solve against it — plan with --pareto 80
3. Inspect and validate — research/inspector.py
4. Formal mathematical proof — research/mathematical_proof.py

This is the target that reproduces the claims in the paper. Run it after any change to the solver’s optimisation logic.

Note that this target reads its test data from research/testdata/ rather than solver/testdata/, because it generates the data itself as step one.