Skip to content
pandora-open-corePublic

About

A stochastic kernel formalised in LEAN4 for pandora-open-core/inferno gradient convergence and a closed-form Schrödinger bridge via Langevin Dynamics on path spaces of Gibbs particle ensembles (Author and Mathematician: Liam Dziga Douglas Giles --Pandora AI Founder)

Topics

Resources

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Repository files navigation

Pudding

Pudding is an alpha, open-core Rust SDK for runtime mathematical consistency checks, finite-support Schrödinger transport, and portable numerical diagnostics. The public core is open source under MPL-2.0 and has no external Rust dependencies. APIs and artifact formats may change during alpha.

Created and contributed by Liam Dziga Douglas Giles, founder and CEO of Pandora-AI Proprietary Ltd.

Install

Use a local source checkout:

[dependencies]
pudding = { path = "../pudding-open-core" }

The package is not published to crates.io. Rust 1.95.0 is pinned in rust-toolchain.toml.

Check a computation

use pudding::{Check, Engine, TensorSnapshot};

let values = TensorSnapshot::new(vec![2], vec![0.25, 0.75]);
let mut engine = Engine::new();
engine.check("finite values", Check::Finite(&values));
assert!(engine.finish().passed());

Engine collects named checks on owned snapshots. Checks cover shape, finiteness, probabilities and agreement within explicit tolerances. Inspect Report::passed() to determine the outcome; an empty report does not pass. See the evidence contract for numerical and formal limits.

Visualise transport

cargo run --locked --bin pudding -- visualise --output artifacts/transport
cargo run --locked --bin pudding -- visualise --mode equilibrium --output artifacts/equilibrium

Open the generated index.html. Each output includes an interactive graph, SVG and JSON measurements, with no web service required. Supply your own strictly positive masses through equilibrium::AtomReference and research::Study. See the transport workflow for input requirements, convergence behavior and artifact formats.

Use with Inferno

Inferno imports this SDK directly and exposes it as inferno::pudding. inferno::diagnostics::{snapshot, check_operator, check_solution} provides Candle tensor and linear-operator checks. The adapter lives in Inferno, so Pudding remains independent of tensor backends. See Inferno.

Containers

The numerical CLI and formal audit have separate images:

podman build -f Containerfile -t pudding .
podman run --rm -v "$PWD/artifacts:/output:Z" pudding visualise --output /output/transport
podman build -f Containerfile.proofs -t pudding-proofs .
podman run --rm pudding-proofs

The runtime image uses an unprivileged user. Ensure the mounted output directory is writable by UID 65532, or use an appropriate Podman user namespace mapping. The proof image installs pinned Lean/mathlib dependencies and rechecks the bundled corpus at build time and on execution.

Formal scope and development

The Pudding Lean library in proofs/lean contains conditional mathematical statements and transitive axiom audits. It does not certify a running binary, floating-point solver, checkpoint or GPU execution. The finite-support equilibrium statements assume the declared feasibility and positivity conditions; they do not establish general continuous bridge existence or Rust solver correctness. See the mathematical contract.

cargo fmt --all -- --check
cargo test --locked --all-targets
cargo clippy --locked --all-targets -- -D warnings
cargo doc --locked --no-deps
# Also requires the pinned Lean/mathlib dependencies:
sh scripts/release-check.sh

The open-core boundary is this public source tree. Private application training, production orchestration and model weights are distributed separately, if at all. See provenance, contributing and security.

About

A stochastic kernel formalised in LEAN4 for pandora-open-core/inferno gradient convergence and a closed-form Schrödinger bridge via Langevin Dynamics on path spaces of Gibbs particle ensembles (Author and Mathematician: Liam Dziga Douglas Giles --Pandora AI Founder)

Topics

Resources

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages