One vocabulary, everywhere SimPilot speaks.

A read-only agent mode that can inspect, explain, and route without launching side effects.

The typed source of truth for a simulation case — solver, physics, geometry, mesh, conditions, controls, QoIs, and report inputs.

The frozen, content-addressed snapshot derived from a CaseSpec at run launch time.

A reusable geometry or mesh region handle — patch, face, edge, node, or region — referenced by BCs, recipes, and figures.

A typed piece of chat — plan, case artifact, tool call, approval, log, validation record, or report preview.

The durable typed event stream behind every chat turn — exportable as Harmony or raw Codex JSONL.

The typed boundary deciding whether an action runs, needs approval, or must be denied.

A schema-bound runtime action — inspect, mutate, run, retrieve, validate, publish.

A typed segment of the agent run — intake_plan, specify, execute, diagnose, validate, report.

A signed, scope-restricted agent spawned by the kernel for a separable concern.

The pass that summarizes cold rollout segments into typed memory entries to bound the context window.

The persistent strip of active cases pinned across the bottom of the shell.

A real file in the case workspace — config, mesh handoff, script, log, plot, or report artifact.

The maintenance pass that retires cold rollout detail into replay-safe summaries.

One of the eight named boundaries (tb-1 to tb-8) across which signed, schema-validated calls flow.

One of four labels — public, org-internal, tenant-restricted, secret — that gates every record's lifecycle.

The deny-by-default outbound network gate with frozen per-route rate limits.

The isolation unit owning a workspace's data, secrets, and approvals.

Bring-your-own-keys — tenant-managed encryption keys for regulated workloads.

A solver integration package — materializes typed case state into solver-native files, commands, parsers, and evidence hooks.

The manifest-like solver metadata read by tools, docs, help, validation, and reports.

A typed predicate over the CaseSpec that selects a solver application or template path.

An OpenFOAM runtime hook that extracts forces, field values, rates, and diagnostics during a run.

A named SU2 mesh boundary marker that must match each generated boundary-condition declaration.

A recycled set of viz leases kept ready for interactive sessions.

The runtime sidecar that tails solver logs and emits typed LiveMonitorFragment records.

The typed credibility contract: QoIs, tolerances, residual targets, benchmark comparisons, and mesh evidence requirements.

A typed plan for related runs — sweeps, mesh refinement, time-step studies, design comparisons, validation benchmarks.

Quantity of interest — a value the case must extract, validate, and report.

Captured command input, output, status, cwd, and timing proving what actually ran.

Solver-monitor evidence that a numerical iteration has reached the declared convergence target.

A pre-solve check that mesh metrics satisfy the solver's and the case's fidelity assumptions.

y+ — a near-wall resolution metric that gates turbulence-model assumptions.

The content-addressed record produced by a completed (or failed) run.

The typed hypothesis ledger written before any rerun after a failure.

Numerical verification evidence — did the discretization converge — separate from physical validation.

Three monotonically refined meshes with matching QoI definitions used to compute GCI or Richardson extrapolation.

Tracks calibration and validation data sets to enforce their disjointness invariant.

Captures Monte Carlo, polynomial chaos, or deterministic perturbation evidence with explicit applicability ranges.

The typed result of a design-exploration sweep — geometric, parametric, or optimization.

The provenance chain from typed state and commands through files, figures, QoIs, validation, and reports.

The publication-pass output for a generated report — links the audience profile, gate, acceptance, and invalidation.

The replay-safe, content-addressed package containing every artifact needed to reproduce a published report.

The typed bundle-distribution approval — structural anti-self-approval, reviewer registry, audit trail.

A typed declaration of who the report is for and what to include or redact.

The OG image and metadata generated for a public share link, at request time.

The typed projection that share, fork, export, and notify routes pass through.

A typed report-policy bundle that adds mandatory sections, checklists, overlays, and signatures.

The typed registry binding figure roles to solver fields, viewpoints, and renderers.

A typed reference to a knowledge-pack chunk used to support a claim.

A signed workflow guide the agent loads when it needs solver- or method-specific procedure.

A durable, classification-aware typed claim with applicability bounds, citations, and validation lineage.

The typed scope inside which a memory record, validation result, or recommendation holds.