Roadmap

Pending work for Kratos MDPA Preview, prioritizing full meshio++ integration, a clearer UI shared with CAD-Preview, and practical mesh preparation and results-analysis workflows. Existing foundations include MMG remeshing, meshio++ WASM, replayable edit histories, Python problemtypes, field visualization, time-series playback, tracked Kratos runs, and a headless MCP server.

meshio++ has resolved its WASM-module issues and is adopted at its full, latest version. Historical binding failures, missing side channels, format defects, and build omissions are no longer exclusions from this roadmap — @meshioplusplus/wasm is pinned at ^15.4.0, the latest published release, and the extension's packaged runtime already declares and uses it. What remains is connecting its capabilities to the extension end to end, preserving Kratos semantics, and verifying the resulting workflows — the items below, not the dependency adoption itself. Keeping that pin current as upstream releases is its own standing item, Tier 0.

This page is aspirational, not a release commitment. All numbered items are pending. Effort is approximate: S = a day or two, M = roughly a week, L = multi-week. Completed features belong in CHANGELOG.md and implementation details in CLAUDE.md; remove completed items here. No tracker issues have been assigned to the items below yet.

Research baseline

Reviewed /home/vicente/src/meshioplusplus on 2026-09-22, including its changelog through 15.4.0, bindings/wasm/js_bindings.cpp, and capability documentation. The extension declares @meshioplusplus/wasm: ^15.4.0. The JS API has no breaking changes between 12.0.0 and 15.4.0 — readMeshSelective gained piece/dropGhosts (composite-dataset selection) and computeNormals was added — so this bump is a routing exercise, not a compatibility one. New format keys since 12.0.0: vtkhdf (14.0.0, read/write), pvd/pvtu/pvtp (15.0.0; .pvd is read natively through its own light index, pvtu/pvtp are routed for reading), pcd/xyz (15.1.0, read/write, no cells), lsdyna (15.2.0, read/write, geometry only), frd (15.3.0, read-only, CalculiX results), gltf/glb (15.4.0, write-only, unrouted — no web-viewer consumer). vtkhdf, pcd, xyz, lsdyna, frd, pvtu and pvtp are now routed (meshioFormats.ts); the full accounting of every live reader/writer key is mesh_capabilities' unroutedReaders, asserted in mcpTools.test.ts. The full transient audit was re-run against 15.4.0 (src/test/fixtures/transient/README.md): no existing classification changed. vtkhdf and .pvd are admitted to the in-file timelines; frd is options-aware but is deliberately not admitted, because its metadata read falls back to a full read (see that README).

Evidence in the local checkout	Roadmap opportunity
WASM bindings and JavaScript API expose repair, computeCurvature, shrinkwrap, sobolevDeform, remesh, remeshVolume, optimizeVolume, grid, voxelize, and computeSdf	Add preparation, analysis, and mesh-generation workflows beyond the operations already integrated.
The same binding exposes diff, meshesEqual, interpolate, slice, isosurface, split, partition, field-management operations, and provenance functions	Mesh comparison, sampled results, reusable derived meshes, partition exports, and traceable conversions.
WASM pipeline documentation describes runPipeline and convertSurfaceOps, which keep intermediate meshes inside C++	Reduce repeated whole-mesh transfers and enable bounded batch processing.
Partition documentation covers weights, recorded source indices, and ghost layers; bindings expose ghostLayers	Extend the existing partition-label operation into actual domain exports.
Changelog records FLAC3D group fixes in 10.36.0, curvature in 10.37.0, and repair/shrinkwrap/Sobolev deformation in 10.38.0	Refresh format fidelity and integrate concrete new operations rather than treating every unused export as a feature request.
The changelog explicitly identifies curved tessellation (10.39.0), point budgets, proximity graphs, PhysicsNeMo workflows, and PMSH/Zarr/CAE/USD format additions as Python-only work	Track these separately as optional companion or future-binding work; resolving WASM defects does not automatically port Python code.

The historical audits in src/test/fixtures/transient/README.md and CLAUDE.md remain evidence about the versions tested. During integration, replace obsolete expectations with positive capability tests and retain relevant regression fixtures. Source inspection supports this roadmap; no new live-WASM audit was performed for this documentation change. A VTK-wasm rendering-runtime evaluation (standalone-session boot, bulk data transfer, a minimal end-to-end render, all under the shipped webview CSP) was run and closed on 2026-09-18 with a drop verdict — see doc/vtk-wasm-spike.md for the full findings, measurements, and reproduction steps; vtk.js remains the renderer.

Magnusim review — 2026-09-21

Implementation review at af8d059: read the result exporters, transient-control calculation, function-object writer, pressure conversions and material library, plus test_transient.py, test_result_filter_registry.py and the relevant local integration paths. This is source/test inspection, not a run of Magnusim or OpenFOAM.

Source evidence	Concrete local gap / decision
estimate_delta_t / resolve_transient_control and transient tests	Our fluid problemtype writes a fixed time step; add transparent time-step and output-budget guidance.
surface averages and summed phi monitors	Existing field integrals are measure-weighted totals, not signed boundary flux or a conservation report.
pressure conversion	The native OpenFOAM parsed-field interface does not retain dimensions; raw p cannot safely be labelled Pa.
material library	Problemtypes define constitutive laws and numeric defaults but lack a reusable physical-material catalog with provenance and unit conversion.
volume threshold exporter	thresholdCells.ts already provides a view-only threshold. Extend it to quantitative extraction/export; do not add a duplicate filter.
streamline exporter	Add seeded trajectories from actual volume velocity; its PyVista implementation is not available automatically in vtk.js.

Delivery rules

• Keep the kernel current first. Tier 0 precedes everything else: each tier's scope is measured against a specific meshio++ release, so a stale pin makes later estimates and acceptance checks describe the wrong artifact.
• Prioritize integration and usability before breadth. Tier 1 establishes a reliable shared foundation; later tiers add workflows on top. Independent UI work can proceed alongside kernel integration.
• Full integration means access to the useful kernel surface through a consistent adapter. It does not require deleting working native implementations or exposing duplicate buttons for equivalent algorithms. Choose a backend per operation on fidelity, performance, and maintenance cost.
• Preserve the extension's data contract. Original node and entity IDs, independent Elements/Conditions/Geometries ID spaces, Properties, constraints, SubModelParts, field components, and source-cell correspondence must survive wherever the operation permits. For topology-changing operations, define generated IDs, field transfer, and metadata handling explicitly.
• MCP parity ships with every headless capability. Update src/mcp/tools.ts, src/mcp/register.ts, src/test/mcpTools.test.ts, and the tool documentation in the same implementation change. UI-only work is exempt; a UI wrapper around a new analysis or edit is not.
• Acceptance checks are part of each estimate. Use real format fixtures, numerical invariants, cancellation and undo/redo tests, and packaged-extension checks where appropriate. Retest the target package as normal integration work rather than leaving features indefinitely labelled “needs live-WASM verification.”

Tier 0 — Keep meshio++ up to date

Admission criterion: upkeep that every other tier depends on, because each tier below is scoped against a specific meshio++ release. Unlike the numbered items further down, this tier is standing work: it is never removed when a bump lands, only its status line is refreshed.

0. Track upstream meshio++ releases — S per release, recurring

Pending — currently one major behind. As of 2026-09-20 the extension declares @meshioplusplus/wasm: ^12.0.0 and has 12.0.0 installed, while npm publishes 13.0.0 (upstream changelog, 2026-09-19) and the upstream checkout is at 14.0.0 (2026-09-20, not yet on npm when this was written). A caret range never crosses a major version, so neither Dependabot's minor/patch group nor npm update will ever propose the jump; it has to be a deliberate change.

For each release, read the upstream changelog entry and classify what it touches for the WASM build before bumping:

• Formats: a new or changed reader/writer key (14.0.0 adds vtkhdf, with partition selection and transient Steps). Decide explicitly whether to route it, and extend MESHIO_READER_KEYS/MESHIO_WRITER_KEYS, MESHIO_READ_CANDIDATES, EXPORT_MENU_GROUPS, SUPPORTED_MESH_EXTENSIONS and the timeline lists. A key that the live artifact reports but nothing routes is listed as deliberately absent with a reason, never left unexamined.
• Boundary and ABI: option structs or dtypes crossing the wasm boundary (14.0.0 grows ReadOptions with piece, 13 → 14). Re-check readMeshSelective/readMetadata options handling, integer array types (BigInt64Array), and the *_components maps.
• Behavior changes that alter what the extension already relies on: 13.0.0 changes fallback semantics in the Python shims and pins C++ streams to the classic locale; whether any of that reaches the WASM build is to be measured, not assumed.
• Fixes that retire a workaround: upstream fixes (for example 12.1.0's MED and Gmsh higher-order node-ordering permutations, VTU polyhedron mixed-node-count cell_data) may make a local compensation redundant or wrong. Remove the workaround and its stale note in the same change.

Then, per bump: update package.json/package-lock.json, confirm both dist/meshio/ variants (sequential and _mt) still load and that locateFile stays name-aware, re-run the transient audit and the per-format options-awareness pins so a changed capability fails a test rather than a user, refresh mesh_capabilities expectations, and record the .vsix size change. Update the "Research baseline" date and the pinned version in the introduction, and add a CHANGELOG.md entry naming the version and the notable capability changes.

Automate the detection so it does not depend on remembering: a scheduled CI job that compares the declared range with npm view @meshioplusplus/wasm version and reports when the latest release is outside it, since that is the one case Dependabot's grouping misses.

Acceptance: the pinned version equals the latest published release (or a documented reason for staying behind is recorded here), the full test suite passes against it, every reader/writer key the live artifact reports is either routed or listed as deliberately absent, and no roadmap item cites a superseded version as its baseline. MCP: no new tool by itself; any newly routed format or operation must appear through mesh_capabilities and the existing info/convert tools in the same change.

Tier 1 — Full integration and a unified user experience

Admission criterion: work that enables multiple subsequent features or improves the everyday workflow across the extension.

1. Improve the UI and unify it with CAD-Preview — L

Pending — explicit cross-extension priority. Audit both extensions against the existing shared UI design system and converge on a common interaction vocabulary and component set. Align File/View/Advanced menus, toolbar hierarchy, navigation controls, orientation cube, layer-tree actions, selection feedback, form layouts, icons, spacing, typography, progress/cancel states, and empty/error states. Update the design-system document's stale “pending” descriptions against what actually ships before using it as the implementation checklist.

Reduce the density of mesh-operation forms with searchable actions, clear categories, progressive disclosure of advanced settings, and consistent inline validation. Establish predictable panel docking and overflow at narrow widths. Include keyboard navigation, visible focus, accessible names, high-contrast themes, and reduced-motion behavior. Share tokens and reusable components through a versioned source or synchronized copies with a drift check; keep Kratos-specific problem setup distinct within the common shell.

Acceptance: compare the same open → inspect → clip → edit → export workflow in both extensions, with visual and interaction checks in dark, light, and high-contrast themes and at small viewport sizes. MCP: UI-only exemption; any new underlying operation discovered during this work still needs parity.

2. Run expensive meshio++ work in cancellable workers — L

Pending. Extend the existing MMG worker pattern to costly meshio++ reads, analyses, and operations. Report stages, support cancellation, release WASM heaps after work, and prevent stale results from replacing a newer frame or document state. Use transferable buffers where ownership permits and benchmark end-to-end memory, including filesystem staging and webview delivery.

Evaluate runPipeline for compatible batches to avoid repeated JS/WASM copies. Preserve the operation queue's per-step history and partial-completion semantics; a faster backend must still return enough results or checkpoints to honor undo and cancellation.

Acceptance: large reads and edits leave the extension host responsive, cancellation releases resources, and batching agrees with sequential execution. MCP: use the shared execution layer where applicable and expose progress/cancellation through the MCP request lifecycle without writing logs to the stdio transport.

Tier 2 — Mesh preparation and analysis

Admission criterion: a concrete user workflow supported by the researched kernel surface, with a clear output and a bounded UI.

3. Line-probe interface — S–M

Pending — the headless half has shipped. mesh_probe samples a nodal field along a polyline (gaps where the path leaves the mesh, optionally across every step of a series, CSV output) and slices, isosurfaces and threshold regions export as mesh files from the Clip dock and the Field panel. What remains is the interactive side: a Probe line action on the Inspect panel that takes two picks the way Measure does, draws the line, and shows the distance-versus-value plot (with time-series repetition and CSV export) in a seriesPanel-style chart — which needs a host round trip for the sampling because meshio++ is host-only, so a new meshAnalysis kind rather than a new message pair.

Acceptance: the plot's numbers equal mesh_probe's for the same endpoints, a path leaving the mesh breaks the line rather than bridging it, and the chart follows the timeline step. MCP: none beyond the shipped mesh_probe; the panel is UI-only.

Tier 3 — Extension workflows and maintainability

Admission criterion: useful extension-level capabilities that build on the integrated kernel and existing document/run infrastructure.

4. Selection-driven editing and Properties authoring — L

Pending. Add multi-selection, box/lasso selection, isolate/hide/restore, selection sets, and filters by part, property, field, and quality. Use selected entities to create SubModelParts or scope operations. Add a Properties editor with explicit shared-property editing versus clone-and-reassign, including beam CROSS_AREA; this replaces the old “beam sections must stay read-only” exclusion with a canonical, undoable authoring workflow.

Acceptance: selection keeps independent entity-kind ID spaces, survives applicable edits through correspondence, and property edits update the model-emitted MDPA writer without creating a competing field value. MCP: selection predicates and property mutations are headless capabilities; pointer gestures and visibility controls are UI-only.

5. Saved view state and independent result comparison — M–L

Pending. Persist camera bookmarks, field/range settings, clipping, layout, and selected layers in a versioned view sidecar. Extend comparison views to two independent meshes or runs, with linked cameras and optional linked physical times. Offer per-pane visibility with an explicit pane scope in the outline rather than silently changing the meaning of the existing global checkboxes.

Magnusim-inspired increment: allow optional synchronized field choice, clipping and color ranges across independent runs, with a visible link toggle for each setting. Match physical time explicitly (exact/nearest with tolerance); disclose unmatched frames rather than synchronizing by frame index. Save the compared run identities with the view.

Acceptance: missing fields/layers degrade predictably on reload, views never mark mesh geometry dirty, and nonmatching time grids are labelled. MCP: view presentation is UI-only; numeric comparison reuses the existing mesh_compare MCP tool and compareField operation.

6. Reusable recipes and batch processing — M–L

Pending. Add named recipe presets, editable/reorderable queued steps, parameter summaries, and batch application to selected files or a discovered series. Show an output plan, per-file progress, completed/failed/skipped results, and resumable manifests. Keep batch execution explicit rather than running expensive operations during ordinary timeline scrubbing.

Acceptance: deterministic output naming, documented partial-failure behavior, cancellation, and reproducible parameters; a batch cannot accidentally overwrite its own later inputs. MCP: batch-transform interface using the same validated recipes and execution reports.

7. Case preflight and isolated run workspaces — M–L

Pending. Expand case validation to report missing assignments, invalid property references, unused or empty parts, field requirements, and mesh-quality concerns before generation. Add opt-in per-run directories containing the generated inputs and a reproducibility manifest, preserving the existing vtk_output layout inside each run. Enable comparison of saved run outputs and parameters without collisions between cases sharing a source folder.

Magnusim-inspired increment: persist explicit source geometry/mesh revision, study ID and run ID in the manifest; clone case settings into a new run without mutating earlier inputs. Results, logs, cancellation and any later monitor/capture artifacts resolve through their owning run, even after switching the active case. Reload must never adopt a different study's output just because its filename matches. Consume a CAD handoff manifest when available and flag stale or unresolved group assignments after remeshing.

Acceptance: validation points to actionable entities or assignments; concurrent runs use distinct output locations and remain discoverable after reload. MCP: extend case validation/generation/run tools and status discovery with the same run-directory contract.

8. Export provenance and fidelity reports — M

Pending. Connect meshio++ provenance functions to conversion, derived-mesh export, recipes, and problem archives. Record source, kernel version, operation parameters, output format, and reported losses. Add an export summary describing retained/dropped groups, IDs, constraints, fields, and companions. Embedded provenance is used where supported; otherwise provide a clearly associated sidecar.

Acceptance: reports agree with a re-read of the output, and exported recipes remain distinct from machine-local solver status. MCP: return the same structured fidelity report and provenance location from write tools.

9. Large-mesh rendering and end-to-end regression coverage — L

Pending. Build on header summaries with progressive surface preview, selective field loading, bounded frame caching, and reduced data transfer. Keep a full-resolution source for editing/export while rendering a smaller representation when selected. Add a maintained packaged-extension integration harness covering both preview providers, save/revert/hot-exit, timelines, cancellation, and sidebar/palette parity; complement the existing standalone webview screenshot tooling.

Acceptance: establish representative large-file memory/latency budgets and exercise local and Remote-SSH-style sessions. A display approximation must not silently become exported geometry. MCP: selective-read and summary options share the host implementation; rendering and UI automation are exempt.

10. Solver convergence and saved monitors — M–L

Implementation route: use runCore.ts/runFile.ts for run ownership and fieldSeries.ts for saved-time sampling. Magnusim's w27-solve.js parses iteration/residual progress while case/function_objects.py generates monitor output independently of full result frames. For Kratos, select a supported output-process/log adapter per problemtype and record its version. Add named monitor rows with last value, time/iteration, history plot and CSV; store the sampling interval independently of VTK output frequency.

Pending. Extend tracked runs with bounded residual/iteration histories and saved monitor definitions. Start with one built-in Kratos problemtype and its documented solver output; prefer structured output where available and make log parsers versioned adapters. Separate iteration count, simulation time, process completion and convergence. Reuse current point/time plots and field-integral analysis for persisted point samples and area-weighted surface averages; live values require solver output configuration, while post-run evaluation samples saved frames only. Store units, selected field/component, region or coordinates, and source run with CSV export.

Acceptance: fixtures cover converged, divergent, cancelled and truncated/restarted logs; unsupported residual output stays unavailable. An analytic field validates surface weighting and point sampling; missing frames or samples remain gaps. Switching cases cannot redirect a monitor or cancel another run. MCP: read convergence and monitor tables, configure supported monitors and export their data through the same adapters. Extend the existing run store rather than adding a second job manager.

11. Streamlines from solved vector fields — M–L

Source-driven increment: Magnusim's export_particle_trace.py resolves selected boundary faces, builds seed lattices and integrates the volume field. Offer seed selection from a SubModelPart as well as explicit points/planes, forward/backward/both directions, maximum steps and terminal-speed tolerance. Preserve seed IDs and termination reasons in exported polylines; report seeds outside the domain and streams that stop immediately. Add line/tube styling only after the numerical export is usable.

Pending. Add steady streamlines seeded from explicit points or a line/plane in a selected vector field and time frame. First establish an interpolation/integration path for the supported volume cell types using the existing vtk.js/data adapters; do not assume meshio++ supplies a streamline binding. Bound step size, length and seed count, report termination at missing data/domain boundaries, and support cancellation. Keep streamlines distinct from transient particle pathlines, which require time interpolation and are a later scope decision.

Acceptance: uniform and rotational analytic fields produce expected trajectories, zero vectors terminate safely, and native solved velocity takes precedence over geometry-only cached previews. Missing velocity disables the operation with a reason. Export polylines with sampled values and source frame identity. MCP: expose seed/integration parameters and derived geometry export through a shared numerical core; viewport styling is UI-only.

12. Transient time-step and output-budget assistant — M

Pending. Extend src/problemtype/builtins/fluid.ts, which currently writes automatic_time_step: false, with an explicit fixed/adaptive choice where the installed Kratos solver supports it. Before generation, estimate a convective time scale from a documented cell length and user-selected reference velocity, target Courant number and safety factor. Show estimated step count, output-frame count and storage range; keep solver step and output cadence separate. Magnusim's estimate_delta_t uses minimum-cell volume, sizing and bounding-box fallbacks, and flow_through_time estimates domain transit time. Preserve that explanation of inputs and fallback basis, but do not silently impose its speed floor or suggest cube-root volume is sufficient for highly anisotropic cells.

Acceptance: analytic uniform meshes give the expected size/velocity scaling; thin/sliver cells receive a conservative metric or an explicit limitation; zero velocity and missing units produce unavailable estimates. User values remain authoritative. Validate emitted adaptive parameters against the selected Kratos runtime and show a convective estimate as guidance, not a universal stability guarantee for implicit, diffusive or structural solvers. MCP: read the estimate and explicitly apply chosen case controls through the existing case tools.

13. Boundary flow balance and pressure-drop reports — M–L

Pending. Add a CFD analysis over selected inlet/outlet SubModelParts: signed volumetric flux integral(u dot n dA), optional mass flux with explicit density, area-weighted pressure, pressure difference between named sections and normalized imbalance with a documented denominator. Repeat over saved times and export CSV. Reuse the integral panel for presentation but introduce oriented surface quadrature; averaging vector components and multiplying by area is not generally a flux integral. Magnusim's function-object writer generates area averages of U/p and sums face flux phi, demonstrating why flow monitoring needs its own semantics.

Acceptance: a straight duct balances opposing inlet/outlet fluxes; reversing face orientation flips the sign; overlapping selections, internal faces, missing velocity/density and uncovered samples are reported. A zero-flow denominator yields unavailable relative imbalance, not infinity. Distinguish volumetric from mass flow and only compare compatible pressure quantities. MCP: read-only balance/pressure-drop report with optional time-series CSV output; live solver monitors can follow through the existing monitor item.

14. Field dimensions and explicit pressure conversion — M–L

Pending. Carry dimensions/units from readers through FieldData, field selectors, legends, probes, comparison and CSV. Start by retaining the OpenFOAM seven-exponent dimensions vector in openfoamFields.ts; preserve original values and add an explicit derived-field conversion from kinematic pressure to Pa using documented positive density. Do not infer pressure semantics solely from a field named p, and do not rescale Kratos PRESSURE, which the fluid case already expresses in Pa. Label gauge/absolute reference separately from units; converting dimensions cannot infer a reference pressure.

Acceptance: fixtures for dimensional pressure, kinematic pressure, unknown units and conflicting density give distinct outcomes; repeated display-unit changes leave original samples unchanged. Difference plots reject incompatible dimensions or require an explicit conversion. Conversion/export provenance records density, source units and pressure reference, with variable-density cases requiring a field-aware policy. MCP: field metadata and explicit conversion tools share the same rules. Coordinate metadata retention with the format/metadata work already shipped (MdpaModel.source, OpenFOAM field dimensions) rather than another reader.

15. Reusable material presets with provenance — M

Pending. Add a small, searchable user-extensible material catalog to the existing problemtype material-law forms. Keep the constitutive law distinct from a preset of parameter values. Each preset carries compatible laws/dimensions, canonical units, reference temperature/conditions, source/version and editable copied values; existing cases retain a snapshot when a library entry changes. Start with independently sourced fluid density/viscosity examples and user-defined entries. Magnusim's materials/library.py illustrates searchable records and reference conditions, but a catalog row (including Water) does not prove its full solver workflow is validated.

Acceptance: converting kinematic viscosity to dynamic viscosity uses mu = rho * nu once; invalid density or incompatible laws cannot generate a case. Imported user presets round-trip, editing a case does not mutate the library, and updating a preset does not alter past runs. MCP: list/inspect presets and explicitly apply a snapshot through case material assignment; generated material files remain the existing writer's responsibility.

Tier 4 — Optional companion and advanced geometry work

Admission criterion: valuable opportunities needing a new runtime boundary or a larger model change. These remain pending, but are not prerequisites for the direct WASM features above.

16. Curved high-order and native polyhedral fidelity — L

Pending. Preserve original high-order/polyhedral topology separately from display tessellation, including provenance for picking and field transfer. Investigate meshio++'s curved tessellate implementation and gather/scatter maps for visually correct quadratic cells and source-preserving exports. The researched tessellation implementation is Python-only; choose a future WASM binding or optional companion rather than assuming it is already available in JS.

Acceptance: curved shared faces remain watertight, high-order node ordering is checked per format, and display subdivision never silently replaces original cells on save. MCP: faithful read/write and explicit tessellation/export capabilities; viewport tessellation alone is UI-only.

17. Optional Python companion for datasets and surrogate results — L

Pending. Explore an explicitly configured external Python companion for meshio++ capabilities outside the WASM package: point budgets, proximity graphs, PMSH (.pmsh), Zarr/CAE dataset export, USD time-series export, and PhysicsNeMo inference. Start with exporting existing runs and loading predicted fields with model metadata and comparison metrics; training orchestration is a later scope decision. Do not assume the problemtype Pyodide runtime can host native Python, Torch, or CUDA dependencies.

Acceptance: discover companion capabilities and versions, report missing dependencies clearly, stream large datasets, and label predictions separately from solver output. MCP: mirror every enabled dataset/export/inference capability through the same optional backend.

Boundaries that still apply

These are product or runtime constraints rather than historical meshio++ WASM blockers:

• No general ParaView-style visualization graph. Prefer focused analysis panels and explicit operation recipes; richer batch processing does not require a second visual programming environment.
• No automatic expensive remeshing on timeline navigation. Keep deliberate re-apply and explicit series processing.
• No silent destructive case rewrites. Directory-based formats (OpenFOAM multi-region and decomposed cases included) need complete companion ownership and fidelity before in-place save is enabled; generated copies remain useful even with a fully working kernel.
• Timelines follow capability, not reader options. A static file stays static even when its reader accepts a time-step option; an in-file timeline needs distinct time values that can be enumerated without a full read.
• Sequence export only targets containers that can represent the series. Static-grid XDMF is not reinterpreted as a changing-topology container; a topology-changing series is refused rather than written against the first grid.
• Keep solver ownership and transport honest. The MCP server starts detached runs and uses log files; it must not claim an exit code after losing observation of the process. Windows graceful stopping needs a separate process/console design, independent of meshio++ integration.
• Rendering remains a separate runtime concern. Software-WebGL translucency, recording with a non-preserved drawing buffer, browser codec availability, and webview CSP restrictions are not fixed by a WASM upgrade. A VTK-wasm replacement was evaluated and dropped — see doc/vtk-wasm-spike.md — so the existing synchronous render/copy capture and WebM/PNG outputs stand until a materially different runtime is proposed and re-evaluated.
• Keep file ownership explicit. The empty preview remains a launcher until an independently justified session abstraction supports late file binding. Shared runs views continue to project one run store.
• Quality partitioning needs a different build. The WebAssembly artifact has no KaHIP (mesh_capabilities.partitioning reports it live): kahip is refused by name and auto resolves to a space-filling-curve cut with no edge-cut minimization. Partition export is shipped on that method; a KaHIP-enabled artifact would only widen method.
• No polyhedral subdivision or agglomeration. meshio++'s subdivide and agglomerate produce polyhedral cells that this extension can only decompose back into tetrahedra on read, so there is no topology-conversion workflow they would complete; revisit only with native polyhedral rendering and export (see item 16).
• Do not advertise Python-only or optional-backend features as bundled WASM capabilities. Track the binding/runtime needed to deliver them while keeping them eligible for future integration.