Interactive modelling explorer
The landing page uses the website typography, plain section headings and a machine choice and explains modelling through plain-language questions. Each module shows the information received, the problem solved, and the information passed onward. Module headings pair technical names with plain-language questions; recorded claims sit behind an expandable evidence panel. An additional physics panel provides the governing relation, exchanged quantities and units, and coupling assumptions for graduate readers. It starts expanded; its open state carries between modules. It separates schematic balance equations from the actual audited numerical protocol, and distinguishes the historical hard guard from the current continuous ITER closure.
Reading structure
Each machine follows the same sequence: establish the scenario, follow a coupled time step, inspect a physics module, understand surrogate selection, and read the evidence. Section buttons move keyboard focus to the corresponding heading while retaining the machine/module URL. They also work in the portable preview, whose base URL points to the public reading links.
The scenario summary distinguishes initial data, evolving quantities, prescribed inputs or coupled responses, and work outside the demonstrated scope. MAST refers here to the MAST-U modelling route. The MAST-U summary retains static imported geometry, prescribed sources and boundaries, and the failed transport accuracy comparison. The ITER summary distinguishes current-diffusion evolution from a new free-boundary equilibrium solve.
An expandable reference defines symbols, units, cell/face locations, accepted-state and nonlinear-iteration notation, and model names. Equilibrium, heating and boundary interface descriptions change with the selected machine: reading prescribed CHEASE geometry is not described as computing a free-boundary solution, and prescribed MAST-U edge settings are not described as a predictive pedestal calculation. Shared physics panels retain the broader coupling explanation and link to technical methods.
The guard demonstration labels its binary routing and assumed finite inputs explicitly. Its machine-specific note distinguishes the current ITER continuous closure from the hard-switch example. The evidence guide distinguishes numerical convergence, comparison against a reference, and formal local robustness; none substitutes for the other checks.
What the visuals mean
The machine drawings and dependency map are schematic, not equilibria, device engineering drawings or evidence of completed coupling. The route labels distinguish implemented work, limited evidence and planned work. The topics-viewed count records reading progress only. The offline JOREK stability branch is hidden from the map and module traversal while it is outside the active coupling route. Its technical documentation remains available. Existing stability bookmarks open the selected machine’s core solver instead.
The guard switches demonstrate routing: acceptable inputs and outputs use a fast response; a failed check uses a fallback. They do not execute a surrogate, a plasma solver or a formal proof. The guard contract describes the actual implementation.
Animated information flow
Play/pause, previous/next and restart controls walk through a schematic time step. A moving packet and directional path identify the sender and receiver; a text panel names the exchanged variables. An accepted-state ledger remains unchanged while trial profiles are formed, then advances only at the illustrated acceptance stage. No synthetic temperature, density or power values are presented as simulation output.
The sequence is an explanation of dependencies, not an instrumented solver call trace. Nonlinear iterations reevaluate modules; the displayed order does not prescribe their internal call order. The ITER path describes the continuous transport closure and its acceptance checks. The MAST-U path explicitly retains static imported geometry and prescribed sources and edge conditions. It does not animate a completed live equilibrium feedback or validated MAST-U trajectory. Stability is kept outside the animated core loop.
Playback stops at the end, pauses on manual topic selection or a hidden page, and resets on machine changes. Reading progress is independent of playback. Reduced-motion settings remove packet motion while preserving directional paths, explanations and manual steps. The palette follows the owner’s published website theme: white and neutral surfaces, black text and a blue primary accent for the active exchange.
The ITER convergence chart reads the same public CSV as the existing convergence figure. It shows maximum endpoint change across the checked quantities for each original ensemble case, with adjacent maximum-step comparisons selectable. C-305 and C-306 define the result and scope. The proof tiles use C-326 directly, including the unresolved and out-of-domain states. MAST-U’s evidence route preserves the static equilibrium result (C-019), failed closed-loop comparison (C-179), and later scope limitation.
Export and checks
The exporter embeds the explorer in the home page and retains a standalone HTML version for preview. Styles are scoped to the explorer. Browser code needs no external libraries, network services, cookies or local storage. Hash links restore a selected machine and module. The remaining build log and technical pages stay accessible without JavaScript.
The evidence asset is generated from an explicit allowlist of public claim fields and the public convergence CSV. Private evidence paths and reproduction commands are not exported into that asset. Missing claims or nonfinite chart values fail generation. The disclosure scrub covers browser scripts as well as prose; link checks cover literal HTML and script links.
Focused export tests check source parity, missing claims, link handling and disclosure checks. The browser check exercises both machines, module traversal, guard fallback, chart controls, deep links, section focus without navigation away from the portable preview, machine-specific interfaces, persistent physics-panel state, per-machine reading progress and narrow-screen layout with the notation and evidence guides expanded. The preview does not replace the owner’s Quarto render and publication review. The browser check also accepts a rendered landing-page path as its final argument. The explorer stylesheet lets that page fit Quarto’s grid on narrow screens while older static diagrams keep their own horizontal scrolling.
Changelog
- 2026-11-04: Structured machine explorer separates scenario scope, interfaces, guard policy and evidence interpretation.