Source-term platform implementation status

Tracking issue: #3860. Implementation scope and consumer validation as of 2026-11-02; this does not confer qualification or replace domain review.

Delivered foundations

Architecture proposal #3861 is merged. Physics #3867, exchange contract #3868, and live integration #3870 reached master through merge commit b08d8bb59a95cf52e0565fab5c368c5548a9f27f. The latter two merged into the physics branch first.

The current implementation includes:

• Opt-in homogeneous-equilibrium short-opening physics, immutable inputs/results, explicit thermodynamic stations and required closure checks.
• A separate prescribed-slip short-opening model retains the HEM thermodynamic station while partitioning kinetic energy between exactly one gas and one liquid phase using a caller-declared velocity ratio. It exposes phase densities/velocities and enforces phase-area and kinetic-energy closure. It does not infer slip, entrainment or finite-rate phase transfer and remains unqualified.
• A bounded vertical-upward drift-flux short-opening model predicts gas/liquid slip from the Zuber-Findlay relation and Harmathy bubble drift velocity while retaining HEM thermodynamics. It requires one gas and one liquid phase, explicit positive interfacial tension in N/m, gas area fraction at most 0.80, and simultaneous drift-flux, phase-area and kinetic-energy closure. It remains unqualified and does not model entrainment, finite-rate phase transfer, annular jets or solids.
• A separate finite-rate phase-split sensitivity model applies an exact first-order relaxation from the upstream gas mass fraction toward the equilibrium station value before using the same bounded drift-flux closure. It requires explicit relaxation time, residence time, interfacial tension and parameter provenance; retains overall component fractions and stagnation-energy closure; emits timestep-independent analytical and refinement evidence; and fails closed outside one-gas/one-liquid bubbly/dispersed scope. It is not component-selective kinetic prediction or experimental qualification.
• A component-selective finite-rate sensitivity model applies caller-declared first-order time constants to each component’s gas-held mass, reconstructs explicit gas/liquid compositions with exact overall component conservation, and retains the same bounded drift-flux and stagnation- energy closures. Equal component times recover the uniform phase-split model. The parameters are not predicted transfer coefficients, and the model excludes interfacial heat transfer, entrainment, droplet area, annular flow, solids and experimental qualification.
• An explicit Ranz-Marshall spherical-dispersion correlation now predicts component external-film coefficients and relaxation times from caller-declared SI diameter, continuous-phase density and viscosity, relative velocity and component diffusivities. It enforces Re < 200 and Sc < 250 without defaults or clamping, recovers the stagnant-sphere analytical limit, and couples gas-bubble cases to the component-conservative vertical drift-flux release model. The correlation exposes liquid-droplet coefficients, but the release adapter fails closed because predictive droplet-in-gas hydrodynamics are not yet available. Residence time and morphology remain explicit caller inputs; interfacial heat/latent kinetics and experimental qualification remain excluded.
• A predictive vertical bubble transfer-zone adapter now derives a conservative bounded spherical- bubble $d_{32}$ proxy, resolved gas/liquid relative speed and gas residence time from explicit SI transfer-zone geometry and the existing drift-flux phase velocities. It iterates Ranz-Marshall component transfer and the component/energy-conservative release path to a checked fixed point, fails closed on correlation range or convergence violations, and retains schema-valid frames from both process containers. It remains an UNQUALIFIED screening closure without a population balance, entrainment, heat transfer, solids or experimental qualification.
• Versioned deterministic JSON, NDJSON and reduced CSV, bundled JSON Schema, source identity, provenance and machine-readable failure/lifecycle states.
• Steady runs, external capture and native dynamic stepping for ProcessSystem and ProcessModel, including real separator dynamics in tests.
• Weighted uncertainty ensembles, retaining failures and refusing unconditional statistics for incomplete ensembles, with schema-compatible case frames.
• Explicit ideal-gas and legacy-screening adapters. The ideal-gas path has no silent property defaults; the compatibility path reports SCREENING_ONLY and unresolved station semantics.
• Coupled inventory depletion through native process dynamics, with component/energy/volume closure, atomic unit updates, receiving-pressure event location, timestep-refinement evidence and schema-validated frames from both process containers. The compatibility API remains single-gas; an explicit phase-selected API supports equilibrium gas, oil, generic-liquid and aqueous withdrawal without inferring entrainment or phase fallback.
• Opt-in cylindrical inventory geometry now derives live liquid level and gas/liquid interface area from phase volumes, confirms that the declared withdrawal phase occupies the opening, and fails closed inside a caller-declared unresolved interface/entrainment band. It supports vertical and horizontal cylinders with one gas and at most one material liquid phase. It is an assessment gate, not a predictive entrainment, wave, slip or finite-rate transfer model.
• A separate opt-in retained-phase inventory keeps gas and liquid component masses independent, applies provenance-required two-way fugacity-film transfer over the live cylindrical interface, limits every donor component, withdraws the selected phase composition/enthalpy, and solves common pressure/temperature from rigid volume and total internal energy without reflashing the committed partition. It retains latent-energy and component-transfer diagnostics, atomic rollback, receiving-pressure events, both process-container paths and timestep-refinement evidence. The coefficients remain prescribed sensitivity inputs; the model assumes phase thermal equilibrium, fails closed on phase disappearance and is not predictive entrainment or experimental qualification.
• CPA inventory consumer regressions add gas/aqueous withdrawal, caller/export isolation, rejected-substep rollback and replay, nearby-temperature timestep refinement, and both native process-container paths. A prescribed test flux isolates bookkeeping; it does not qualify discharge rates or phase disappearance/reappearance. The physical-slot lifecycle in #4310 remains an explicit dependency.
• Analytical gas limits, dense-fluid conservation and flashing, multicomponent, lifecycle and contract regressions in focused CI.
• Guarded same-EOS continuation for the previously unresolved propane/butane entropy root, with nine nearby mixture cases, a separate saturation-path comparison, inventory scaling, back-pressure sensitivity and retained schema/benchmark evidence. Model version 1.2.0 preserves the acoustic warning and UNQUALIFIED evidence boundary.
• Mixture-specific, fail-closed equilibrium-solid and hydrate applicability assessment at the resolved upstream, throat and ambient-expanded stations. Detected risk is UNSUPPORTED; an unresolved required assessment is INVALID. This does not add solid-bearing flow physics.
• A separate ideal-gas Fanno model for quasi-steady, one-sided full-bore release through a constant-area pipe with explicit length and specified Darcy friction. The same immutable geometry reaches steady/live process frames and coupled inventory depletion; short-opening models reject it rather than ignoring it.
• An EOS-backed real-gas Fanno model covers the same bounded quasi-steady, one-sided, constant-area geometry while retaining EOS density, caloric properties and acoustic speed at every numerical state. It conserves mass flux and stagnation energy, matches an unchoked receiver, and fails closed on phase appearance. It explicitly excludes transient waves, line packing and non-equilibrium multiphase physics.
• A separate one-dimensional perfect-gas finite-volume process unit now resolves finite-speed decompression and line packing for a closed-end pipe discharging to a constant-pressure receiver. It conserves pipe-plus-discharge mass and total energy, uses CFL-limited substeps, exports the exact committed boundary flux through both process containers, and retains grid-refinement evidence. It is a calorically perfect-gas model, not a real-gas or multiphase qualification.
• An EOS-backed single-gas finite-volume process unit applies the same conservative transient boundary while recovering cell pressure, temperature, phase state and acoustic speed through NeqSim volume/internal-energy flashes. It retains exact pipe-plus-discharge mass and total-energy accounting, both process-container paths, dilute-limit and grid-refinement evidence, and fails closed on phase appearance or detected/unresolved solid risk. It remains unqualified and excludes heat transfer, elasticity, upstream-vessel/two-sided coupling and non-equilibrium multiphase physics.
• Immutable per-model evidence manifests now retain stable applicability and limitation codes, typed analytical/conservation/numerical/experimental references, and an explicit independent- evidence flag in schema-validated frames. Current built-in manifests truthfully remain UNQUALIFIED; no successful calculation promotes its own evidence level.
• The ideal-gas Fanno model retains an independently published NASA GFSSP nitrogen benchmark (NTRS 20070036728) with exact SI inputs, a 1% mass-flux error bound, choked-exit assertion and a machine-readable comparison receipt. This is analytical evidence for one ideal-gas case, not experimental or real-gas/transient qualification.
• The executed NeqSim-Colab safety source-term demonstration is merged through EvenSol/NeqSim-Colab #176, with retained outputs, rendered-equation/figure inspection, catalog entry and repository validation.

Legacy LeakModel behavior is preserved. The new interface now has homogeneous-equilibrium, analytical ideal-gas and explicit legacy-screening implementations. Model selection remains caller-owned; no phase-count rule silently changes the requested physics.

Remaining work before the whole issue can close

Work item	Current boundary	Completion evidence required
Broader transient inventory regimes	Rigid equilibrium inventory supports explicit phase-selected withdrawal, ordered phase-exhaustion transitions, conservative receiving-pressure events, external heat boundaries and balance/refinement tests. The opt-in retained-phase inventory adds two-way component film transfer, donor limits, thermodynamic latent-energy accounting, live interface area and inventory feedback without reflashing the committed partition. It assumes one gas/one liquid at a common temperature, prescribed coefficients and fail-closed phase disappearance.	Predictive/validated coefficient and interfacial heat-transfer models; separate phase-energy/temperature dynamics where required; predictive entrainment beyond the fail-closed interface band; qualified fire/wall boundary models rather than caller-imposed duties.
Full-bore/long-pipe and non-equilibrium regimes	Bounded ideal-gas and EOS-backed real-gas Fanno models cover quasi-steady one-sided constant-area single-gas pipe flow. Conservative perfect-gas and EOS-backed single-gas finite-volume units cover one-sided transient waves and line packing. Short-opening prescribed-slip, bounded vertical drift-flux and finite-rate phase-partition models cover one gas and one liquid within explicit screening limits. A predictive spherical-bubble transfer-zone adapter supplies bounded bubble size, phase-relative speed and residence time to Ranz-Marshall but deliberately rejects liquid-droplet hydrodynamics.	Pipe heat transfer/elasticity and upstream-vessel/two-sided coupling; measured or predictive population-balance, breakup/coalescence and entrainment models; liquid-droplet, annular/high-Weber and solid-bearing transport; independent dense-gas and multiphase validation data. The inventory’s lumped external heat boundary is not distributed pipe heat transfer.
Independent qualification and dense-fluid accuracy	Machine-readable evidence records distinguish applicability, limitations, evidence type and independence. The Fanno manifest retains one external analytical case with explicit error bounds; frames remain UNQUALIFIED.	Add independent experimental/dense-fluid datasets with range and uncertainty analysis, then obtain accountable domain review.
Solid-formation applicability	Mixture-specific solid/hydrate station assessment now fails closed and retains machine-readable diagnostics.	Assessed solid-bearing release physics where supported, plus independent mixture validation.
Multicomponent flashing qualification	The documented 80/20 propane/butane entropy root and nearby cases now close with guarded continuation; a separate same-EOS saturation path checks the maximum. Acoustic warnings remain explicit.	Independent experimental benchmarks and domain review; broader mixtures are not qualified by the regression matrix.
Colab demonstration	Executed, output-retaining demonstration merged in NeqSim-Colab #176 with rendering, catalog and validation-ledger checks.	No remaining implementation item; broader physical qualification remains governed by the rows above.

Do not close #3860 merely because initial PRs merged or focused CI passes. The NRC retains design intent; current guidance is in physics, contract, live sessions and uncertainty.