Use a Solution

Two ways to use PWM-validated solutions: (a) SpecLab — a chat interface where you describe your data in plain English, upload measurement files, and get reconstructions back; or (b) self-serve — match against PWM benchmarks, download sample data, run the reference solver locally. SpecLab is recommended for non-developers; self-serve is recommended for reproducibility-first workflows.

SpecLab — chat interface (recommended)

● live · pwm.platformai.org

Talk to PWM in plain English. Four modes: Reconstruct (recover a clean signal from a noisy measurement), Mismatch (analyze calibration drift), Design (propose a new measurement system), and Scientific Simulation (run physics-faithful forward models). Upload .npy / .npz / .mat data, get reconstructions + spec.md back. No coding required.

Open SpecLab →

SpecLab lives on the main platform site (pwm.platformai.org) — the full product UI with chat history, file upload, and account management. The explorer (this site) is the read-only protocol view. See /contribute for the platform-vs-explorer split.

Or use the explorer's self-serve path below — three steps: (1) describe your data, (2) preview the matched benchmark's example samples, (3) run the reference solver locally. Quicker to verify reproducibility, no account required, results are byte-stable across runs.

Describe your data

The faceted matcher is deterministic (no LLM). It checks your prompt against benchmark cards using keyword + filter scoring. A score ≥ 2.0 returns matching benchmarks; below that you'll be invited to propose a new spec via the Contribute path.

Or browse what's already validated

Two anchors live on testnet today. Three more (MRI, CT, TEM) arrive after the G4 gate — see the roadmap.

CASSI

L1-003 → L3-003

Coded Aperture Snapshot Spectral Imaging

Reconstruct a 28-band hyperspectral image cube from a single 2D coded-aperture snapshot.

Input shape: 256 × 256 (snapshot)
Output shape: 256 × 256 × 28 (HSI cube)
Reference solver: GAP-TV
Samples available: 10 (KAIST-10)

Try this benchmark →Read 4-layer walkthrough

CACTI

L1-004 → L3-004

Coded Aperture Compressive Temporal Imaging

Reconstruct an 8-frame video block from a single 2D temporal-coded snapshot.

Input shape: 256 × 256 (snapshot)
Output shape: 256 × 256 × 8 (video)
Reference solver: PnP-ADMM
Samples available: 6 (SCI Video Benchmark)

Try this benchmark →Read 4-layer walkthrough

Full matcher →All 16 demo samples →All benchmarks →Browse principles →

Run the reference solver locally

Once you've picked a benchmark, copy-paste the snippet on its Get this benchmark card. For CASSI/L3-003 it looks like:

git clone https://github.com/integritynoble/Physics_World_Model.git
cd Physics_World_Model
python3 pwm-team/pwm_product/reference_solvers/cassi/cassi_gap_tv.py \
    --input  pwm-team/pwm_product/demos/cassi/sample_01 \
    --output /tmp/out
cat /tmp/out/meta.json
# Expect PSNR around the value in the demo's meta.json (byte-stable
# across runs at the same git SHA)

Found nothing that matches?

PWM's catalog is the Genesis 500 (locked) plus translator-proposed principles (post-mainnet). If your problem isn't covered, you can stake to propose a new principle / spec / benchmark — that's the contributor path.

See the contributor paths →