Close the transient/dynamics parity gap: sustained severe slugging, liquid-rich stability, and a published controls benchmark

[EvenSol]

Summary

NeqSim's dynamic infrastructure is strong — 4 integrators, 45 equipment classes with runTransient, 7 control structures, MPC, and a DynamicCapability audit framework that has no commercial equivalent. Steady-state TwoFluidPipe reached reference parity in #3077/#3078 (+1.4 / +1.6 / +0.1 / −2.7 % dP, −2.4…+3.5 % hold-up across the 4/7/10/12 MSm³/d sweep).

What is not at parity is transient multiphase physics. Three named defects block it, and they are the reason olga-simulation-agent exists to delegate. This epic closes them.

Scope explicitly excludes re-litigating steady-state closures — that work landed and is validated.

Current state (measured, not estimated)

Working

• Steady TwoFluidPipe at reference parity (fix(pipeline): real gas viscosity in the stratified level solver, Bendiksen drift velocity, reachable annular criterion #3077, fix(pipeline): give the slug closure its gravity term, and make the physical defaults the defaults #3078); three-phase dP +16.9 % (was +189.5 %)
• TwoFluidPipe.runTransient is a genuinely conservative scheme: null test 0.00 bar drift over 4 h; mass balance closes to the digit (−197.9 t both by inventory and by ∫(ṁ_in − ṁ_out)dt); inlet settles within −0.13 % of reference on a rate step
• 4 integrators incl. adaptive RK45 and BDF; transactional stepping; capability audit

Blocking defects

D1 — Severe slugging limit cycle is damped, not sustained. (highest value)
On the Tengesdal (2002) benchmark, the instability starts ≈30 s, decays to flat by ≈100 s, then blows up numerically after ≈265 s.

quantity	reference sim	NeqSim	measured (experiment)
riser-base swing	34.9 kPa (0.278 heads)	no limit cycle (P10–P90 1.96 kPa)	98 kPa (0.78 heads)
cycle period	21.7 s	none	38 s

Root cause is identified, not speculative: calcOutletFlux clamps reversed phase velocities to 0 with no diagnostic (isTransientOutletBackflowClamped() is true, liquid_outlet_min == 0.0 exactly). Severe slugging requires liquid fallback at the riser top — the clamp forbids the mechanism, so the riser packs to the 0.95 hold-up clamp and stays. The fix is the outlet BC / interfacial-pressure coupling, not a hold-up closure. The flowline already agrees to 1 %; the disagreement is entirely in the riser.

D2 — Liquid-rich transient packs without bound.
The interfacial-pressure term is correct but explicit, forcing CFL ≈ 0.05. Folding it into the IMEX implicit pressure solve couples the phase momentum equations to the void-fraction wave. Currently @Disabled in TwoFluidPipeTransientNullTest.

D3 — CoupledPressureMomentumSolver stalls with a silent no-progress return.
runTransient returns getElapsedTimeSeconds() == 0. Isolation (50 × 0.1 s steps):

config	steps ok	t_end
legacy / interfacial only / backflow only	50	5.00 s
coupled (± interfacial ± backflow)	8	0.77 s

Mechanism: iters == 12 == maximumIterations, residual ≈6e-7 > relativeVolumeTolerance 1e-7 → converged=false → adaptiveDtFactor *= 0.5; currentStep--; continue; without decrementing timeRemaining, so every substep is rejected and the call silently advances nothing. Halving dt does not recover.
API gap: setMaximumIterations / setRelativeVolumeTolerance exist on the solver but there is no accessor from TwoFluidPipe, so the gate is untunable by a user. Same for pressureCorrectionLimited, computed then never surfaced.

Secondary

• Regime misclassification at high hold-up: 39 % hold-up classified ANNULAR; D=0.30 m → ANNULAR 20/20 vs D=0.50 m → STRATIFIED_WAVY 20/20 at near-identical hold-up. A friction form cannot repair a misclassified regime — an annular-film friction model was implemented and reverted (made it worse: +13.2 % vs +1.4 %).
• Three-phase free water steady solve does not converge (4078 iters / 1200 s wall-clock; profile is stationary to 0.01 bar but the criterion never trips).
• 4 @Disabled transient tests; 6 UNCLASSIFIED_DYNAMIC capability declarations unaudited.
• PipeBeggsAndBrills.runTransient has no mass-storage term — documented as not fixable in that class (needs a second boundary condition); route users to TwoFluidPipe.

Workstreams

WS1 — Sustained severe slugging (unblocks the headline capability)

• Replace the calcOutletFlux one-way clamp with a signed outlet BC permitting liquid fallback
• Surface a sticky diagnostic mirroring isSteadyStatePressureFloorLimited() — never clamp silently
• Couple interfacial pressure into the implicit solve at the outlet
• Accept: Tengesdal test 3 produces a sustained limit cycle; riser-base swing within ±30 % of 98 kPa and period within ±30 % of 38 s; flowline hold-up unchanged (0.342 ±1 %); no blow-up over ≥600 s

WS2 — IMEX implicit interfacial pressure (unblocks liquid-rich)

• Fold the interfacial-pressure term into the implicit pressure solve
• Re-enable both @Disabled TwoFluidPipeTransientNullTest cases
• Accept: liquid-rich null test bounded over 4 h; CFL ≥ 0.5 usable; mass balance still closes to the digit

WS3 — Coupled-solver robustness (prerequisite for WS1/WS2)

• Never return zero elapsed time silently — throw or expose a sticky diagnostic
• Expose solver gate (maximumIterations, relativeVolumeTolerance, pressureCorrectionLimited) through TwoFluidPipe
• Make the tolerance achievable within the cap, or make the cap adaptive
• Accept: coupled config runs 50/50 steps on Tengesdal; outlet rate compared against the stored reference trace (0.375–4.03 kg/s) — note the existing ledger test satisfies itself by construction; the outlet rate is the real gate

WS4 — High-hold-up regime map

• Fix the annular/slug branch above ~0.3 hold-up
• Acquire an inclined/high-hold-up regime-map validation basis before touching closures
• Accept: D=0.30 m and D=0.50 m at equal hold-up classify consistently; no dP regression on the reference sweep

WS5 — Controls parity benchmark (the un-evidenced half)

There is no published NeqSim controls benchmark. We have MPC, cascade, split-range, override, ratio, feed-forward and anti-surge — and no artefact a reviewer can cite.

• Define 5–8 canonical loops (level, pressure, cascade temperature, split-range, anti-surge recycle, compressor speed/recycle coordination)
• Publish setpoint-step and disturbance-rejection metrics via ControllerPerformanceMetrics (IAE, overshoot, settling time)
• Wire into AgentBenchmarkSuite
• Accept: a citable, CI-run controls benchmark page

WS6 — Capability audit closure

• Audit the 6 UNCLASSIFIED_DYNAMIC declarations into real categories
• CI gate: no new UNCLASSIFIED_DYNAMIC without an ADR
• Accept: DynamicCapabilityReport shows 100 % audited

Sequencing

WS3 (solver robustness)  ──►  WS1 (severe slugging)  ──►  WS4 (regime map)
        └──────────────────►  WS2 (IMEX liquid-rich)
WS5 (controls benchmark)  ──  independent, parallelisable
WS6 (capability audit)    ──  independent, low effort

WS3 first — WS1 and WS2 both terminate in the coupled solver, and its silent stall will mask their results.

Hard-won constraints (do not re-litigate)

• ❌ Do not chase D1 in the hold-up closures. Defects 1 and 2 there are fixed; the remaining blocker is the outlet BC.
• ❌ Drift-flux for the slug closure is a trap. It fixes direction (ratio → 10.22) but fails TwoFluidPipePhaseDegeneracyTest: C0>1 + finite v_d keeps α_G<1 at zero liquid; the Zuber-Findlay (1-α_G)^n fix kills the inclination response again.
• ❌ Do not apply separated friction in ANNULAR. Measured +14.7 % at 10 MSm³/d and pressure floor at 12.
• ❌ Do not tune to a specific commercial trace in public artefacts. Equinor-internal comparison is licensed; public repos need an independent basis.
• ⚠️ Always run the regime probe before touching a closure. An Andritsos-Hanratty interfacial friction model was implemented, measured byte-identical, and reverted — because the line was 100 % ANNULAR and the closure was inert.
• ⚠️ Dimension-check every criterion. The Kelvin-Helmholtz criterion was dimensionally wrong (spurious extra h_G) and produced a fake "improvement" that vanished on fixing it.

Definition of done

• Sustained severe-slugging limit cycle within ±30 % of experiment on amplitude and period
• Liquid-rich transient bounded; both @Disabled null tests re-enabled
• Coupled solver runs the benchmark end-to-end with a tunable, surfaced gate
• Published controls benchmark in CI
• 100 % DynamicCapability audited
• Docs updated: TWOFLUIDPIPE_MODEL.md, two_fluid_model.md, neqsim-dynamic-simulation + neqsim-flow-assurance skills
• olga-simulation-agent routing guidance updated to reflect what NeqSim can now own

[EvenSol]

Campaign ledger — WS3 coupled-solver robustness

• Roadmap item: Close the transient/dynamics parity gap: sustained severe slugging, liquid-rich stability, and a published controls benchmark #3298 WS3 prerequisite; draft PR [AI-Assisted] fix(twofluid): harden coupled solver progress #3300
• Initial reproduction base: 016aed9a0a9cc8d7a970d9d9adcff26149934322
• Published base: 83a99c5c9ef7235798d4a1784adcdd00f2bf0e05 (master advanced by three non-overlapping commits before publication)
• Exact PR head: 21a1c928a941e1aba430580b0261f6d06f91641c
• Frozen concept: bounded/fail-loud coupled pressure-momentum progress plus public nonlinear controls and diagnostics
• Split reason: none; this is the complete dependency-ready WS3 numerical-progress batch. WS1/WS2 physics remain outside the stop boundary.

Changed files

• TwoFluidPipe.java
• CoupledPressureMomentumSolver.java
• TimeIntegrator.java
• new CoupledPressureMomentumTengesdalProgressTest.java
• TWOFLUIDPIPE_MODEL.md
• two_fluid_model.md
• neqsim-dynamic-simulation/SKILL.md
• neqsim-flow-assurance/SKILL.md

Result and evidence

The former 12-iteration cap stalled near a relative cell-volume residual of about 6e-7, above the 1e-7 gate, and adaptive retry could return no physical-time progress. The default budget is now 24; maximum iterations, tolerance, latest limiter state, sticky limiter/failure states, and rejected-substep count are public. An unrecoverable adaptive call throws with partial/requested time and nonlinear diagnostics.

• Public Tengesdal 16-section probe: 50/50 × 0.1 s completes, exact 5.0 s
• Refinement: 24 sections, 100/100 × 0.05 s completes, exact 5.0 s
• Every reported gas/oil/water/liquid/total discrete mass residual: < 1e-9
• Default path: no rejected nonlinear substep and no one-way outlet clamp
• Existing 1200 s coupled liquid-rich long-horizon regression: passed
• Existing null/outlet regression set: passed; two pre-existing liquid-rich null tests remain disabled
• Focused/current-base Maven totals: 14 tests run, 12 passed, 2 pre-existing disabled, 0 failures/errors
• python devtools/run_spotless.py check: passed
• python devtools/check_documentation_search.py: passed (667 Markdown + 1 HTML)
• git diff --check: passed
• Local Javadocs were attempted but could not start because the runner has a JRE and no javadoc executable; no pass is claimed and exact-head CI is pending.

Documentation impact

Updated all WS3-affected #3298 targets present here: both TwoFluidPipe model guides and the dynamic-simulation/flow-assurance skills. Controls benchmark documentation is not affected by this WS3 runtime path. No olga-simulation-agent routing file exists in equinor/neqsim, and ownership is not promoted while physical qualification remains open.

Remaining limitation / blocker

This is progress evidence, not severe-slugging qualification. The sticky pressure limiter still fires and the 50-step interval-average liquid outlet spans -18.552 to 6.883 kg/s, outside the stored 0.375 to 4.03 kg/s comparison. No public closure was tuned to the commercial trace.

Next dependency: inspect the regime first, dimension-check the outlet/implicit-pressure coupling, and validate against the public Tengesdal experiment, nearby operating points, conservation, and mesh/time-step refinement. Leave #3298 open; WS1, WS2, WS4, WS5, WS6, and the final capability/routing audit remain.

[EvenSol]

Campaign state correction after merged WS3

PR #3300 merged into master as 3a8f8e590976b1aef2e159b984d5a050d0e01335 with every exact-head workflow passing. It completes the numerical-progress part of WS3: bounded fail-loud behavior, public solver controls/diagnostics, 50/50 Tengesdal progress, refinement, conservation evidence, and the four affected WS3 documents.

The issue was closed at merge, but #3300 explicitly did not claim full completion. I reopened #3298 because the definition of done is still incomplete:

• Tengesdal liquid outlet remains -18.552 to 6.883 kg/s versus the stored 0.375 to 4.03 kg/s comparison, with the pressure limiter active;
• no sustained 600 s severe-slugging limit cycle within the experimental amplitude/period gates;
• the two liquid-rich null tests remain disabled;
• high-hold-up regime-map validation, the published controls benchmark, and 100% capability audit remain;
• final olga-simulation-agent routing guidance is not yet eligible for promotion.

Current master: b358d3c6dfff2d25967c6be48d0feb4887d4b207.

Next dependency: preserve the merged WS3 path and investigate the signed outlet / implicit interfacial-pressure boundary coupling against the public Tengesdal experiment. No hold-up-closure, drift-flux, ANNULAR-friction, or commercial-trace tuning is in scope.

[EvenSol]

Campaign ledger — WS1 outlet-face investigation (no PR)

• Roadmap item: WS1 fixed-pressure outlet/interfacial-pressure coupling, after merged WS3 increment [AI-Assisted] fix(twofluid): harden coupled solver progress #3300.
• Exact default-branch head: b358d3c6dfff2d25967c6be48d0feb4887d4b207.
• Remote implementation head / changed files: none; no branch, commit, or PR was published.
• Regime probe: ./mvnw -B -ntp -Dtest=neqsim.process.equipment.pipeline.twophasepipe.validation.SevereSluggingBenchmarkHarnessTest test — 2/2 passed on unmodified master before any numerical edit.
• Exact-master baseline: ./mvnw -B -ntp -DexcludedTestGroups= -Dtest=neqsim.process.equipment.pipeline.twophasepipe.CoupledPressureMomentumTengesdalProgressTest test — 3/3 passed; the durable [AI-Assisted] fix(twofluid): harden coupled solver progress #3300 comparison remains approximately −18.55…6.88 kg/s over 5 s.
• Candidate investigated locally: replace the last-cell mass projection with a dimensionally checked conservative fixed-pressure outlet-face flux, couple the corresponding outlet momentum correction, bound phase correction fluxes for positivity, and reject finite phase-mass undershoots before the cross-phase state limiter.
• Focused evidence: isolated coupled-solver conservation tests passed 3/3. With a bounded 64-iteration gate, resolved/refined Tengesdal progress passed over 5 s and narrowed the resolved outlet envelope to approximately −16.25…6.35 kg/s. This is still far outside the stored 0.375…4.03 kg/s comparison and therefore is not WS1 acceptance.
• Sustained limitation: extending the resolved case exposed a gas phase-balance error at interval 63 (8.19e-5 relative) when a finite negative phase mass was redistributed by the state limiter. Rejecting that undershoot preserved phase semantics but exhausted the 50,000-attempt guard while advancing only 0.0832/0.1 s in the affected interval. A broader conservative flux-limiter run did not complete before the execution environment disconnected; no pass is claimed.
• Dimension check: matrix coefficient dt² A α/(ρ Δx_cell Δx_face) is m²/Pa; multiplying by pressure correction gives the cell area residual. Face correction dt A α ∇p′ is kg/s and its accepted ledger contribution is kg.
• Documentation impact: none. No implementation increment is publishable, so changing public capability or routing documentation would overstate the evidence.
• Blocker / split reason: the outlet face cannot be published independently while its pressure correction can deplete a phase and either trigger cross-phase redistribution or collapse adaptive progress. The candidate improves the short envelope only modestly and has no ≥600 s evidence.
• Next dependency: formulate a locally conservative, positivity-preserving phase flux correction that remains represented in the pressure residual (rather than post-scaling it), then re-run 5 s resolved/refined, nearby points, and the ≥600 s public acceptance path before opening the sole WS1 draft PR. No closure, hold-up, ANNULAR friction, or regime-map change was made.

[EvenSol]

Campaign ledger — WS1/WS2 coupling diagnosis (no PR)

• Roadmap items: WS1 fixed-pressure outlet/interfacial-pressure coupling and the directly dependent WS2 liquid-rich null gates.
• Exact default-branch head: b358d3c6dfff2d25967c6be48d0feb4887d4b207; re-read immediately before this write.
• Active Close the transient/dynamics parity gap: sustained severe slugging, liquid-rich stability, and a published controls benchmark #3298 implementation PR: none. No remote branch, commit, file, or PR was written.
• Frozen concept: determine whether the existing operator-split implicit void-wave boundary can be corrected independently, then check whether the merged coupled configuration alone retires the two WS2 disables.
• Split reason: no implementation batch is publishable. A monolithic pressure/interfacial-momentum boundary formulation is now a prerequisite; independent post-split edits move the public short benchmark in the wrong direction.

WS1 evidence

Two dimensionally consistent outlet/operator variants were isolated from unmodified master. Both preserved phase mass and cell total momentum and passed focused tests, but failed the benchmark direction:

1. A backward-Euler absorbing outlet row for q = alpha_g alpha_l (v_g-v_l), with dimensionless c dt/dx, passed TimeIntegratorTest (13/13) and the 5 s coupled progress suite (3/3). The resolved liquid-outlet envelope widened to -23.248…8.330 kg/s, versus the master ledger baseline -18.552…6.883 kg/s.
2. Using the documented interfacial slip coefficient K in the Helmholtz coefficient K dt^2/dx^2 passed TimeIntegratorTest (13/13), TwoFluidPipeInterfacialPressureTest (3/3), and the focused 5 s progress case (1/1), but widened the envelope further to -24.205…7.464 kg/s.

No closure, regime, hold-up, ANNULAR-friction, or commercial-trace tuning was attempted. The previously run public regime probe remains unchanged.

WS2 evidence

The two disabled liquid-rich null tests were enabled locally only after selecting the full coupled configuration: implicit interfacial pressure, coupled pressure-momentum, signed outlet, and CFL 0.5.

./mvnw -B -ntp '-Dtest=neqsim.process.equipment.pipeline.TwoFluidPipeTransientNullTest#testLiquidRichSteadyStateIsAFixedPointOfTheTransient+testLiquidKeepsLeavingTheOutletUnderConstantBoundaryConditions' test

Result: 2 run, 1 passed, 1 failed. Liquid continued leaving the outlet through 600 s, but the fixed-point case drifted 36.837% from its own steady inventory over 1,800 s (acceptance <5%). Therefore neither a 4 h claim nor re-enabling both tests is supportable.

Documentation impact

Documentation impact: none. Every change was an uncommitted diagnostic candidate rejected by benchmark or acceptance evidence; updating capability or routing guidance would overstate the implementation.

Changed blocker and next dependency

The blocker is no longer adequately described as an outlet positivity limiter alone. The interfacial-pressure correction is applied after the pressure projection and does not participate in the same phase-flux/volume residual or fixed-pressure boundary row. Post-split outlet changes can remain conservative while worsening the signed phase-rate excursion, and the merged configuration still does not make the steady solution a transient fixed point.

Next dependency: derive and test a conservative block pressure/interfacial-momentum correction whose phase flux response is represented in the nonlinear volume residual and fixed-pressure outlet boundary. Require positivity inside that solve, then repeat 5 s resolved/refined and nearby Tengesdal points before the >=600 s public slugging gate; for WS2, require fixed-point and conservation evidence through 4 h at CFL >=0.5 before removing either disable.

[EvenSol]

Campaign ledger — block-continuity active-set diagnosis (no PR)

• Roadmap items: WS1/WS2 block pressure–interfacial-momentum prerequisite.
• Exact default-branch head: 9e4e36d4b6a59404ac9aa629740fbc312610d3c8; re-read immediately before this write.
• Active Close the transient/dynamics parity gap: sustained severe slugging, liquid-rich stability, and a published controls benchmark #3298 implementation PR: none. Previous increment [AI-Assisted] fix(twofluid): harden coupled solver progress #3300 remains merged; no remote branch, commit, file, or PR was written.
• Frozen concept: include the implicit interfacial-pressure phase-flux response in conservative continuity before the existing nonlinear pressure/volume projection, with a signed fixed-pressure outlet and phase-specific positivity at correction faces.
• Changed files: none. All diagnostic Java/test edits were reverted; the local tracked tree is clean.
• Split reason: no publishable implementation batch. The candidate exposed a stronger nonlinear prerequisite.

Baseline and dimension check

Unmodified current master retained the prior numerical baseline:

• public regime probe: SevereSluggingBenchmarkHarnessTest — 2/2 passed;
• resolved/refined coupled progress: CoupledPressureMomentumTengesdalProgressTest — 3/3 passed.

Phase momentum is kg/s, so its implicit change is a phase mass-flow response. The continuity contribution -dt * divergence(delta mDot) is kg/m, and a fixed-outlet ledger contribution dt * delta mDot_out is kg.

Candidate evidence

The block precursor applied the implicit void-wave momentum response through conservative phase face fluxes before the pressure projection. Its 5 s resolved comparison completed, but the liquid-outlet envelope was -19.072…4.566 kg/s, versus the master ledger -18.552…6.883 kg/s: the positive excursion narrowed, while the negative excursion slightly worsened and remained outside 0.375…4.03 kg/s.

Sustained probes rejected both available positivity placements:

1. Without positivity inside the pressure correction, the generic cell limiter activated by second 7 and broke gas phase conservation (relative residual no longer < 1e-9).
2. Donor-limiting the completed pressure-correction face flux preserved phase semantics but removed flux response from the nonlinear Jacobian. The solve then advanced only 0.886136/1.0 s after 1136 attempts; latest relative volume residual 0.558211, 24/24 iterations, limiter active.
3. Suppressing the implicit outlet response did not repair convergence: 0.667794/1.0 s, 863 attempts, residual 0.004403, 24/24 iterations.
4. Raising the budget to 100 was not a remedy: 0.493872/1.0 s, 639 attempts, residual 0.700647, 100/100, limiter active.

Focused numerical tests passed before the sustained rejection (TimeIntegratorTest + CoupledPressureMomentumSolverTest: 15/15), confirming this is a coupled nonlinear/active-set failure rather than an isolated algebra or API failure. No >=600 s pass is claimed.

Documentation impact

Documentation impact: none. Every source/test candidate failed a mandatory conservation or sustained-convergence gate and was removed. Updating model, skill, controls, or routing documentation would overstate capability.

Changed blocker and next dependency

A sequential “momentum response → conservative face map → pressure projection” is still not a block solve. Post-limiting either permits cross-phase inventory repair or makes the pressure residual/Jacobian inconsistent and stalls.

Next dependency: formulate one nonlinear system whose unknowns include pressure correction and interfacial phase-flux response, with the fixed-pressure outlet row and phase-positivity constraints represented by an active set/complementarity condition inside each nonlinear iteration. Its residual/Jacobian must change when a phase face becomes constrained; do not clip a completed flux. Re-run focused conservation/positivity, 5 s resolved/refined and nearby Tengesdal points, then the >=600 s WS1 and 4 h CFL >=0.5 WS2 gates before opening the sole draft PR.

[EvenSol]

Campaign ledger — feasible active-set block-response diagnosis (no PR)

• Roadmap items: WS1/WS2 block pressure–interfacial-momentum prerequisite.
• Exact default-branch head: 9e4e36d4b6a59404ac9aa629740fbc312610d3c8; re-read immediately before this write.
• Active Close the transient/dynamics parity gap: sustained severe slugging, liquid-rich stability, and a published controls benchmark #3298 implementation PR: none. PR [AI-Assisted] fix(twofluid): harden coupled solver progress #3300 remains merged; the portfolio has room, but no benchmark-qualified batch exists.
• Frozen concept: put phase-specific positivity inside the pressure residual/Jacobian, then include the implicit void-wave phase mass-flow response in the same conservative correction and fixed-pressure outlet ledger.
• Changed files: none. All diagnostic source/test edits were reverted with Spotless applied; the tracked tree is clean.
• Split reason: no publishable batch. The active set cures the previous nonlinear stagnation, but a precomputed/operator-split interfacial response still fails the public benchmark direction and WS2 fixed-point gate.

Dimension and numerical evidence

Phase momentum and its implicit response are kg/s. Mapping that response to a face contributes -dt * divergence(delta mDot) in kg/m to phase continuity, while dt * delta mDot_out is kg in the fixed-outlet mass ledger.

A phase-face active set was inserted inside each pressure matrix rebuild. Constrained phase mobility and the known interfacial face response were both removed from the affected residual/Jacobian row rather than clipping a completed flux. This eliminated the prior 24/24 or 100/100 stagnation in the exercised cases:

• CoupledPressureMomentumSolverTest: 6/6 passed, including depleted-phase positivity, nonzero response with an initially closed volume residual, per-phase conservation, and fixed-outlet ledger identity.
• CoupledPressureMomentumTengesdalProgressTest: 3/3 passed; resolved and refined 5 s cases completed with every phase residual below 1e-9.
• TwoFluidPressureMomentumLongHorizonTest: 1/1 passed through 1,200 s with the existing conservation/stability gates.

The public short-case direction nevertheless rejected the batch. The resolved 50-step liquid-outlet envelope became -23.156…8.197 kg/s, worse than master -18.552…6.883 kg/s and outside the stored 0.375…4.03 kg/s comparison. This closely reproduces the earlier absorbing-boundary regression, despite the new feasible active set.

The full coupled WS2 diagnostic also remained unchanged: 2 run, 1 passed, 1 failed. Liquid continued leaving the outlet, but inventory moved from 90,247.90 kg to 57,002.87 kg over 1,800 s, a 36.837449% drift versus the <5% fixed-point gate. The final 5 s interval remained discretely conservative (250.000 kg inlet, 250.925 kg outlet); the failure is the initial-state-to-dynamic-state relaxation, not an unreported mass-balance defect.

Documentation impact

Documentation impact: none. The candidate failed mandatory public benchmark and fixed-point gates and was removed. Updating TwoFluidPipe guides, NeqSim skills, controls documentation, or routing guidance would overstate capability.

Changed blocker and next dependency

Positivity placement is no longer the immediate blocker: an active set inside the matrix rebuild can remain conservative and converge. The remaining defect is that the void-wave response was still precomputed outside the nonlinear solve and treated as a frozen known flux. That is not the requested monolithic block system; it moves the outlet dynamics in the same wrong direction as the earlier operator-split boundary variants and does not change the WS2 fixed-point relaxation.

Next dependency: make interfacial phase flux (or the equivalent drift-flux increment) an actual nonlinear unknown coupled to pressure and phase continuity, with its own residual/Jacobian terms, fixed-pressure outlet equation, and the proven phase-face active set/complementarity update. Recompute the response as pressure, density, holdup, and slip change inside each nonlinear iteration rather than freezing it before the solve. Then repeat focused conservation/positivity, resolved/refined and nearby Tengesdal points, the >=600 s WS1 gate, and the 4 h CFL >=0.5 WS2 gates before opening the sole draft PR.

[EvenSol]

Campaign ledger — phase-resolved block requirement (no PR)

• Roadmap items: WS1/WS2 pressure–interfacial-flux prerequisite.
• Exact default-branch head: 9e4e36d4b6a59404ac9aa629740fbc312610d3c8; re-read immediately before this write.
• Active Close the transient/dynamics parity gap: sustained severe slugging, liquid-rich stability, and a published controls benchmark #3298 implementation PR: none. PR [AI-Assisted] fix(twofluid): harden coupled solver progress #3300 remains merged with successful exact-head workflows, no reviews, and no unresolved threads. Seven autonomous implementation PRs are open, so portfolio capacity was not the blocker.
• Frozen concept: replace the previously frozen void-wave response with an evolving interfacial drift-flux unknown inside the nonlinear pressure iteration, including conservative phase-face transport and a fixed-pressure outlet ledger.
• Changed files: none. All diagnostic Java/test edits were restored byte-for-byte to the authoritative master blob SHAs; no branch, commit, file, or PR was written.
• Split reason: no benchmark-qualified implementation survived.

Dimension and structural result

The drift unknown q = alpha_g alpha_l (u_g-u_l) has units m/s. Its phase-face response rho_k A q is kg/s, its continuity contribution -dt div(rho_k A q) is kg/m, and its fixed-outlet ledger contribution is kg. The gas and liquid volume transports are +Aq and -Aq, so they cancel identically in the current solver's only total-occupied-area pressure residual.

The diagnostic block iterate recomputed the backward-Euler void-wave target from the current phase areas on every nonlinear iteration, applied only the evolving q increment to conservative gas/liquid face fluxes, preserved cell total momentum, included the signed fixed-pressure outlet transfer in the mass ledger, and rebuilt pressure/density from the resulting state. Thus q was no longer a frozen precomputed response.

Verified evidence

• CoupledPressureMomentumSolverTest,TimeIntegratorTest: 15/15 passed.
• CoupledPressureMomentumTengesdalProgressTest: 3/3 passed, including resolved and refined 5 s progress/conservation.
• Resolved Tengesdal outlet envelope: -18.181959520506773…7.188233911003568 kg/s, versus master -18.552…6.883 kg/s. The negative excursion improved slightly but the positive excursion worsened; the public benchmark remains unqualified.
• 1,800 s liquid-rich fixed-point diagnostic: 36.837511888396996% inventory drift, effectively unchanged from the recorded master 36.837449% and far above the <5% gate.
• Restored-master regime probe SevereSluggingBenchmarkHarnessTest: 2/2 passed.
• Maven/Spotless bootstrap was run before Maven; python devtools/run_spotless.py apply completed after each Java edit. The diagnostic-only failing assertions used to expose numeric values were removed before restoration and are not claimed as passing.
• Documentation impact: none — no source, API, numerical behavior, user guide, skill, controls benchmark, or routing change survived.

Narrowed blocker and next dependency

An evolving q unknown is necessary but insufficient when paired only with total-volume pressure closure. In the flat liquid-rich null case the interfacial source is near-null, and its equal/opposite phase-volume response cannot influence the pressure residual. The next block must add an independent phase-resolved continuity residual (or equivalent conservative phase-flux equation) to the pressure/interfacial-flux Jacobian, with fixed-pressure outlet and positivity/complementarity rows. A pressure-volume plus drift-momentum pair alone is under-coupled for WS1/WS2 acceptance.

Issue #3298 remains open.

[EvenSol]

Campaign ledger — WS2 fixed-point isolation (2026-08-29)

Roadmap item: WS2 liquid-rich stability, still blocked behind the WS3 coupled pressure-momentum path.

Authoritative base/head: master 9e4e36d4b6a59404ac9aa629740fbc312610d3c8; no active #3298 PR and no branch/head created. Seven open autonomous [AI-Assisted] implementation PRs were found, so the portfolio cap was not the blocker.

Reproducer and dimension check

The unchanged public regime probe passed 2/2 before any closure experiment:

./mvnw -B -ntp -DexcludedTestGroups= -Dtest=neqsim.process.equipment.pipeline.twophasepipe.validation.SevereSluggingBenchmarkHarnessTest test

On the 5 km, 0.30 m, 40-cell, 50 kg/s liquid-rich fixture, master moved from 90247.896718 to 90210.123530 kg in the first 5 s. The conservative ledger reported exactly 250.000000 kg entering and 287.773188 kg leaving. The coupled fixed-pressure outlet-cell correction was only 0.118454 kg, so it is not the direct source of the 37.773188 kg inventory change.

Pressure-gradient momentum correction has dimensions
dt * alpha*A*grad(p) = s * m2 * Pa/m = kg/s, matching phase momentum per length; its induced mass-flow correction -dt*alpha*A*grad(dp) is kg/s.

Rejected handoff candidate

The pure-oil/pure-water branch of updateLiquidPhaseSplit returns before synchronizing the present phase velocity with the solved bulk-liquid velocity. A local two-line candidate corrected the initial oil-flow profile from 52.8078...51.2745 to 40.2796...40.3376 kg/s and reduced the maximum initial oil-mass RHS from 0.100126 to 0.000153351 kg/(m s).

It changed the five-second inventory and boundary ledger by exactly zero, so it is a real diagnostic/handoff inconsistency but not the WS2 fixed-point cause. It was not published as an isolated micro-fix.

A factorial five-second probe then isolated the momentum split:

configuration	inventory change
coupled pressure-momentum + implicit interfacial pressure	-37.773188 kg
coupled pressure-momentum, interfacial pressure off	-38.931695 kg
coupled solver off, interfacial pressure off	-12.731985 kg
coupled solver off, implicit interfacial pressure on	+46.631026 kg

The corresponding diagnostic Maven runs intentionally failed at the injected evidence assertion after producing these values; they are not claimed as passing tests. Maven was prepared with the required prepare_maven.py helper, and python devtools/run_spotless.py apply was run after every Java edit.

Changed blocker

The previous “add an independent phase-continuity row” dependency is not supported by this isolation. Once the stale pure-phase velocity is synchronized, phase-mass continuity is already within 1.54e-4 kg/(m s), while the initial phase-momentum RHS remains approximately 13.9 kg/(m s2) for gas and 12.17 kg/(m s2) for oil near the outlet. The coupled pressure corrector dominates the later physical outlet transfer through momentum, despite its tiny direct outlet-cell mass projection.

Next dependency: audit and reconcile the discrete steady/transient momentum fixed point at the fixed-pressure outlet—specifically the steady mixture pressure march, phase-resolved finite-volume momentum residual, and coupled pressure corrector—before adding another block unknown. This stays inside numerical coupling; no steady closure, hold-up closure, ANNULAR friction, or drift-flux slug closure was changed.

Surviving changed files: none. The experimental source and diagnostic test were restored byte-for-byte to authoritative master blobs:

• TwoFluidPipe.java 00c5c71825d8631837c4d5571854099c13555293
• TwoFluidPressureMomentumLongHorizonTest.java 68739f3ee2b77b350edce785b0d898a44ecfcb6d

Documentation impact: none — no source, public API, behavior, benchmark, or user guidance change survived.
Split reason: no PR was opened because the only source candidate did not improve either public acceptance path, and publishing it alone would violate the campaign's coherent-batch boundary.

#3298 remains open.

[EvenSol]

Campaign ledger — phase-resolved SLUG fixed-point isolation (2026-08-29)

Roadmap item: WS2 liquid-rich stability, still blocked behind WS3/WS1-compatible phase-momentum coupling.

Exact heads: editing began from master 4b0b533c5066ece83119bd36728f010ee24b97c4. Before this ledger write, remote master was re-read at bb497b5189d5b1b177ef0dc8b3edd180b8463ad4 (an unrelated MCP merge); the four relevant blobs remained unchanged. No active #3298 PR exists, only three open repository PRs exist, and no branch/head was created.

Reproducer and fixed-point decomposition

The required public regime probe ran before the closure experiment and passed 2/2:

./mvnw -B -ntp -DexcludedTestGroups= -Dtest=neqsim.process.equipment.pipeline.twophasepipe.validation.SevereSluggingBenchmarkHarnessTest test

On the 5 km, 0.30 m, 40-cell, 50 kg/s liquid-rich line, all 40 cells are SLUG. The initial interior total-momentum residual was 44.17–45.53 Pa/m. The steady mixture gradient was 93.51–98.09 Pa/m while the transient wall source supplied only 49.29–52.21 Pa/m. The units close as force per volume: (N/m) / m2 = Pa/m.

The transient source multiplied the SLUG wall closure's already holdup-partitioned mixture-shear shares by separate phase-segment perimeters. Referencing both shares to the full pipe perimeter reduced the interior total residual to −1.38…−1.86 Pa/m, but the 1,800 s inventory drift only improved from 36.84% to 33.7693%; this correction is not sufficient alone.

Phase-resolved decomposition then exposed the dominant remaining mode. With the total residual reconciled, gas and oil still carried nearly equal-and-opposite interior residuals of approximately −188 and +150 Pa/m. The fixed-pressure outlet volume projection was only 0.096 kg over the first 5 s, so it is not the physical inventory-loss source. The absolute Oliemans bubble-drag source is relaxing the landed SLUG unit-cell slip state toward zero slip.

Rejected closure experiment

As a diagnostic only, the resolved gas-liquid drag source was removed for homogeneous SLUG/CHURN cells while retaining the conservative full-perimeter wall force.

• Initial phase residuals fell to about −0.88 and −0.51 Pa/m in the interior.
• Five-second inventory change fell to +0.00229%.
• The full 1,800 s WS2 test passed: coupled convergence and volume tolerance at every interval, discrete mass balance within tolerance, nonzero liquid outlet transfer, bounded pressure/holdup/velocities, and no outlet clamp.
• The public Tengesdal progress/refinement suite passed 3/3 numerically, but the actual 5 s liquid-outlet envelope became −26.1407…3.83831 kg/s versus the stored 0.375…4.03 kg/s comparison.

The experiment therefore destroys the WS1 public behavior and was rejected. No closure was published.

Changed blocker

The defect is not merely an outlet half-cell or another total-volume pressure row. The landed steady SLUG unit-cell closure permits finite phase slip, while the transient phase-momentum source applies absolute bubble drag whose equilibrium is zero slip. A safe next implementation must make the transient SLUG interfacial momentum equilibrium consistent with the existing unit-cell slip state while retaining perturbation damping, conservation, positivity, and the fixed-pressure boundary. It must clear both the 1,800 s WS2 criterion and the Tengesdal outlet/refinement matrix; zeroing drag, a holdup threshold, or a drift-flux slug closure is not acceptable.

All diagnostic Maven failures were intentional evidence assertions and are not claimed as passing. The successful commands were the 2-test regime probe, the 1-test 1,800 s WS2 experiment, and the 3-test Tengesdal progress/refinement suite. Maven was prepared with prepare_maven.py, and Spotless apply followed every Java edit.

Surviving changed files: none. Experimental edits were removed and local Git blob hashes match remote master:

• TwoFluidConservationEquations.java 76167a23a3fbff2358eea86e0a730afa87a646fc
• TwoFluidPipe.java 00c5c71825d8631837c4d5571854099c13555293
• TwoFluidPressureMomentumLongHorizonTest.java 68739f3ee2b77b350edce785b0d898a44ecfcb6d
• CoupledPressureMomentumTengesdalProgressTest.java 157d0122d5053e3fea23f60cba0c4d910a3bef73

Documentation impact: none — no source, API, behavior, benchmark, skill, example, or user-guidance change survived.
Split reason: no PR was opened because the only candidate that passed WS2 failed the public WS1 envelope, and publishing either partial source change would violate the coherent-batch gate.

#3298 remains open.