Three notebooks run entirely offline on synthetic or bundled data. Each demonstrates mechanics only and makes no geographic, hydrodynamic, or scientific claim. All three run offline on synthetic or bundled data; none requires the proprietary engine.
Notebook path
Index
- N1 —
registry_discovery.ipynb
Audience: ADCIRC scientist.
Learning goal: load the bundled manifest, search Domains, look up one Mesh Record (WNAT/hagen@v1), and walk its fields labeled authoritative-as-declared vs computed-at-publish.
Offline guarantee: fully offline by default; one clearly-markedMesh.load()cell requiring the[hf]extra is gated behind an explicit network opt-in and is skipped (never executed) in CI. - N2 —
evidence_chain.ipynb
Audience: developer / reviewer.
Learning goal: diff a tiny synthetic fort.14 pair, run Tier-1 structural gates on both meshes, evaluate a declared scope into aScopeVerdict, and (optional[crypto]) sign and verify a DSSE attestation envelope with a committed TEST-ONLY key.
Offline guarantee: fully offline; no network, no database, no backend, no proprietary engine. - N3 —
intent_guided_region_proposal.ipynb
Audience: developer / reviewer.
Learning goal: compile refinement constraints (supported / advisory / unsupported diagnostics), inspect a deterministic recipe, import precomputed synthetic selection artifacts, rank region proposals, and cross the single human approval boundary to an approved region.
Offline guarantee: fully offline; the proprietary engine is never imported.
Determinism
Every notebook runs top to bottom from a clean kernel and produces deterministic results: no current-time output, no randomness, and no network in the default/CI path. Each is validated by execution in CI, not by stored outputs — committed notebooks have their outputs stripped.
Candidate proposals are advisory. An approved region is the artifact that crosses the authority boundary and can be used for enforcement-grade scope verdicts.