State-of-the-art source-term platform for live safety-simulator input

Problem

NeqSim has useful release-screening utilities, but a safety simulator needs a stable, model-explicit source-term boundary that can be driven directly by both steady-state and dynamic process models. Current calculations do not yet expose every thermodynamic station, calculation quality, transient composition, validation provenance, or a versioned neutral exchange contract.

Objective

Create a supplier-neutral source-term platform that:

consumes NeqSim ProcessSystem, ProcessModel, streams, and trapped inventories;
supports steady snapshots and sequenced dynamic/live frames;
separates release physics from process coupling and serialization;
exposes upstream, critical/throat, orifice-exit, and ambient-expanded states;
carries explicit SI units, composition, phase mass fractions, model/version identity, assumptions, diagnostics, and calculation quality;
fails closed in strict mode and never hides property or flash fallbacks;
supports uncertainty ensembles and qualification evidence;
produces deterministic, schema-validated input for downstream safety simulators without depending on proprietary software.

Proposed PR sequence

NRC (proposal only): agree architecture, safety boundary, contract lifecycle, validation expectations, and rollback.
Release-flow physics: add a pluggable model API; preserve the legacy screening path; add ideal-gas and homogeneous-equilibrium models, explicit stations, immutable results, strict/screening policies, tests, and model-selection documentation.
Exchange contract: add a versioned ConsequenceScenario/source-term package and JSON Schema with deterministic JSON, CSV, and newline-delimited streaming representations.
Process/live coupling: connect streams and inventories from both ProcessSystem and ProcessModel; support steady snapshots, dynamic stepping, calculation IDs, simulation time, sequence numbers, staleness, isolation timing, and time-varying composition.
Qualification and advanced regimes: add uncertainty propagation, benchmark evidence, full-bore/transient inventory coupling, and dense-fluid/solid-formation applicability controls.
Demonstration: publish a NeqSim-Colab notebook after the stable Java APIs merge.

Initial model boundary

The first physics PR will provide:

a backward-compatible screening model matching current LeakModel behavior where possible;
a homogeneous-equilibrium, isentropic mass-flux model using EOS enthalpy/entropy and the equilibrium acoustic diagnostic;
explicit model applicability and limitations;
no claim that passing software validation qualifies a method for a specific facility or engineering decision.

Non-equilibrium slip, delayed flashing, finite-rate phase transfer, full-bore pipe decompression, dry-ice/solid transport, atmospheric dispersion, fire, explosion, and toxic consequence calculations remain separate models or later work.

Acceptance criteria

Java 8 compatible public API with complete Javadoc.
Input thermodynamic systems are unchanged by calculations.
No silent default molecular weight, heat-capacity ratio, density, phase fraction, or flash recovery in strict mode.
Gas, liquid, aqueous, and solid mass fractions are computed by mass and sum to unity within tolerance.
Choked/subcritical decisions and critical-state diagnostics are explicit.
Every output value has declared units and every frame has stable identity and provenance.
Invalid, stale, unsupported, and screening-only results are machine-readable.
Deterministic serialization and schema compatibility tests.
Steady and real dynamic-path tests for both process container types.
Benchmark matrix spans gas, flashing/two-phase, high-backpressure, multicomponent, and dense-fluid cases.
Documentation includes equations, assumptions, decision tables, examples, validation status, failure modes, and integration guidance.
No runtime dependency on external safety software and no supplier-specific terminology in the new contract.

Governance

This work introduces public safety-model semantics and a new serialized contract. It therefore starts with an NRC and requires independent safety/domain review. API stability and successful regression tests are not engineering qualification.

feat(safety): add failure-aware weighted source-term ensembles

Status checked against master 5c0d6f4663e94117d764d3c417642775210e70bf.

The foundation is now merged: NRC #3861, physics #3867, contract #3868 and live integration #3870. The latter two were stacked into the physics branch and reached master through b08d8bb59a95cf52e0565fab5c368c5548a9f27f.

I continued the implementation in #3878 (9ed6e347be7a25190370716b69395ce40ce0ce80):

weighted joint-input uncertainty cases with retained failure probability;
unconditional mean/quantiles blocked whenever any case is invalid or unsupported;
deterministic v1 case frames with explicit ensemble identity and case-versus-time sequence semantics;
broader gas/high-backpressure/dense-CO2 benchmark coverage and retained CI CSV evidence;
executable documentation and an explicit remaining-work table.

Local validation: 41 focused tests passed, 10 emitted schema fixtures passed, 10 deliberate schema corruptions rejected, Java 8 API compile passed, formatting/pre-commit/pre-push/documentation checks passed. Gas-reference maximum rate deviation was 0.2604%. The dense-CO2 check establishes conservation, not independent rate accuracy. Hosted CI is pending.

This issue should remain open. Remaining work:

Explicit ideal-gas and legacy-screening adapters for the new model interface.
Two-way release/inventory depletion coupling with component/energy balance and timestep convergence.
Full-bore/non-equilibrium models and independently reviewed qualification evidence.
Mixture-specific solid-risk applicability and unresolved multicomponent flashing validation.
The executed Colab demonstration and its catalog/validation ledger.

Current live source terms remain hypothetical openings without inventory feedback, and all exported model evidence remains UNQUALIFIED. Draft #3878 still requires the independent review specified by GOVERNANCE.md.

Campaign ledger — 2026-09-21

Authoritative base: 5c0d6f4663e94117d764d3c417642775210e70bf. Sole active implementation PR: #3878, exact head d2c330e8f983409938d4821b0cae1a68e316d6c7 (open, draft, mergeable). The existing repair adds unmodifiable defensive copies to two getters; both inline review threads are resolved. No independent owner/domain approval is present.

New verified evidence

Java 21 fast suites now pass on Ubuntu and Windows: each reports 14,782 tests, zero failures/errors, 209 skipped. All four Java 21 slow shards also pass.
The exact-head source-term gate passes 41 tests, validates 10 emitted frames and rejects 10 invalid mutations. Retained benchmark/schema/test evidence.
Java 8/21 model specifications, Javadoc, formatting, pre-commit, documentation and code-analysis gates pass.

Changed blocker and action
Java 8 Ubuntu job 106237826916 compiled 3,407 production and 2,221 test sources, then failed before test execution because Maven could not resolve the artifact descriptor for org.apache.maven.surefire:surefire-api:3.5.6. This is dependency-resolution infrastructure, not an observed source-term assertion failure. Attempted a single-job rerun; GitHub rejected it with “The workflow run containing this job is already running.” Java 8 Windows is still running. Next action: after that workflow completes, retry only the failed Ubuntu job and inspect the exact-head result. Do not cancel the running Windows suite or make speculative dependency/source changes.

Scope and documentation
No source files or branch refs changed in this repair/evidence run; existing exact-head CI was reused instead of duplicating local suites. Updated the PR description to replace stale platform-pending claims with the actual results and infrastructure blocker. The frozen nine-path uncertainty/benchmark concept and existing completion-status documentation remain unchanged. No new PR, master synchronization, merge or readiness declaration.

Acceptance/next dependency
Foundations #3861/#3867/#3868/#3870 remain merged; uncertainty in #3878 is still unmerged. Keep #3860 open. Explicit ideal-gas/legacy adapters, two-way component/energy-conserving inventory coupling and timestep convergence, appropriate full-bore/non-equilibrium capabilities, independent qualification, mixture solid-risk controls, the unresolved multicomponent flash and the clean-executed Colab demonstration remain outstanding. No evidence level is elevated: exports remain UNQUALIFIED.

The immediate dependency is successful remaining Java 8 platform validation plus required independent safety/engineering review of #3878. After a verified merge, choose the next dependency-ready model increment; do not mix new release physics into this uncertainty PR.

Campaign ledger — exact-head CI complete

Current master is 30a6cbafda1780ca3f9ef9e18ab9b8f4eafd670d. The sole active implementation PR remains #3878 at exact head d2c330e8f983409938d4821b0cae1a68e316d6c7, open and draft.

Material change since the previous ledger

The targeted Java 8 Ubuntu retry passes: 14,782 tests, 0 failures, 0 errors, 209 skipped. Java 8 Windows also passes.
All 31 exact-head check runs are complete: 30 successful and one not-applicable PaperLab matrix job skipped. Java 21 Ubuntu/Windows fast suites, all four Java 21 slow shards, Java 8/21 model specs, Javadoc, the 41-test source-term gate, schema validation, formatting, documentation, CodeQL and code-quality gates pass.
The earlier conflict classification was stale. The two master commits after [AI-Assisted] feat(safety): weighted source-term uncertainty and benchmark coverage #3878's recorded base affect only refinery and unit-conversion paths, with no overlap against [AI-Assisted] feat(safety): weighted source-term uncertainty and benchmark coverage #3878's frozen nine-path source-term diff. GitHub reports [AI-Assisted] feat(safety): weighted source-term uncertainty and benchmark coverage #3878 mergeable. The exact head was therefore preserved without routine master synchronization.
Both inline threads are resolved. No independent safety/engineering domain approval is recorded.

Scope, documentation and evidence
No source, branch, schema, model or evidence-level changes were made in this qualification follow-up. The PR description and qualification comment now contain the final exact-head results and corrected mergeability state. Existing uncertainty, source-contract, live-session and completion-status documentation remains the applicable documentation-impact record. Exports remain UNQUALIFIED.

Blocking dependency and next increment
#3878 is CI-qualified but not READY TO MERGE: GOVERNANCE.md requires independent safety/engineering domain review, and the author/automated checks cannot provide it. Keep #3860 open. Do not add new release physics or the Colab demonstration to this uncertainty PR.

After independent review and verified merge of #3878, select the next dependency-ready coherent model increment. The outstanding acceptance criteria remain explicit ideal-gas/legacy-screening adapters; two-way component/energy-conserving inventory coupling with timestep convergence; appropriate full-bore/non-equilibrium capability; independent applicability/qualification evidence; mixture-specific solid-risk controls; the unresolved multicomponent flash; and the clean-executed, rendered, cataloged NeqSim-Colab demonstration.

[AI-Assisted] feat(safety): weighted source-term uncertainty and benc…

Campaign ledger — adapters published after uncertainty merge

Verified #3878 merged as commit 1cd66357557ab045ca03afe86bf6f81f0a3bb2e0. Weighted uncertainty, failure-aware statistics and benchmark evidence are now on the default branch. The campaign lane advanced to draft #3889, exact base 6746e5f185c69c995de3532040222f046dc552da, exact head e4e4e012fbc8455e5a2f9abacd504315a92e6283.

Delivered in the new seven-path increment

IdealGasReleaseModel: explicit constant-gamma isentropic gas model with analytical density, velocity, choked/subcritical decision and station states; finite NeqSim molar mass and gamma are mandatory and no default/model fallback is permitted.
LegacyScreeningReleaseModel: exact numeric compatibility with LeakModel.calculateMassFlowRate for choked, subcritical and no-flow cases. Usable results are always VALID_WITH_WARNINGS with SCREENING_ONLY, unresolved-station and fallback-policy diagnostics.
Existing model selection remains caller-owned. Legacy scalar/blowdown behavior and schema v1 are unchanged.
The source-term CI lane now runs the adapter suite and retains its emitted schema fixtures.
The release-model guide records equations, units, station semantics, selection/applicability and migration behavior; the platform status moves the adapter criterion to delivered.

Verified local evidence

47 focused tests passed with zero failures/errors/skips.
Analytical choked rate/critical pressure and Mach identity, critical-ratio continuity, subcritical/no-flow limits, unsupported liquid behavior, input immutability and Java serialization pass.
Legacy rate equality is exact in all three flow regimes.
12 Java-emitted frames passed Draft 2020-12 schema validation; 10 corruptions were rejected.
Java 8 compilation produced class-file version 52.
Javadoc doclint, repeated Spotless apply/check, pre-commit, pre-push, documentation search (726 Markdown + 1 HTML) and whitespace checks pass.
All seven published blobs were verified byte-for-byte against the validated local files. Hosted CI is active and not yet claimed.

Boundaries and next dependency
This adapter PR remains draft, UNQUALIFIED, and requires independent safety/domain review. It adds no automatic selection, inventory depletion, full-bore/non-equilibrium physics, solid transport, dispersion or evidence-level elevation. Current autonomous implementation occupancy is 4, below the target of 20 and ceiling of 24.

Keep #3860 open. After #3889 merges, the next dependency-ready implementation is two-way transient release/inventory coupling with component and energy conservation plus timestep-convergence evidence. Full-bore/non-equilibrium capability, independent qualification, mixture-specific solid risk, the unresolved multicomponent flash and the clean-executed/rendered Colab demonstration remain separate later increments.

Campaign ledger — coupled gas inventory published (2026-09-21)

Verified #3878 merged as 1cd66357557ab045ca03afe86bf6f81f0a3bb2e0 and #3889 merged at 15:16:28 UTC as 03de75e5965263a596ccd4c08f84a947dc3d56ef. Weighted uncertainty and both explicit adapters are on master.

The sole active campaign PR is now #3893, branch codex/source-term-inventory-3860, exact base 03de75e5965263a596ccd4c08f84a947dc3d56ef, exact head f28a78eeda3bef8ac2b36cabf6d0004bac7e6be3. GitHub reports draft and mergeable; no reviews or inline threads are present. Hosted CI is running. No readiness declaration or independent domain approval is claimed.

Implementation advanced

Adds ReleaseInventory: a rigid, adiabatic, well-mixed single gas-phase inventory that removes component mass and upstream stagnation enthalpy through native process transient calls.
Couples the changed inventory back into release physics, with explicit volume/internal-energy solution, strict per-step and cumulative closure, bounded integration, immutable accounting and same-UUID duplicate-withdrawal protection.
Adds addInventorySource for both ProcessSystem and ProcessModel, using the inventory's own geometry/model and recording coupled provenance in existing v1 frames.
Separates physical opening isolation from frame-export disabling. A failed unit step commits no inventory/accounting/clock/UUID change; whole-process rollback is not claimed.
Rejects screening results, incompatible release state/composition/energy, unsupported phase changes and receiving-pressure overshoot.
Nine-path concept includes implementation, 13 tests, retained benchmark/schema CI evidence, executable guide, live-session documentation and navigation/status updates.

Verified evidence

The entire local baseline tree was reconstructed from connector reads and matched master tree 8fddb1fe5c9cc3544b50c88c38f9e544384dbd28; 17 unmodified adapter/session tests passed.
All 60 focused tests pass, zero failures/errors/skips; 17 emitted schema frames pass and 10 corruptions are rejected.
Mixture component/energy/volume closure, cooling and decreasing rate, nearby subcritical cases, no flow, input immutability, serialization, physical isolation/reopening, export disabling, stale identity, mid-step rollback and both native process containers pass.
Nitrogen at 300 K and initial 2.0/2.2 bara, with 0.2/0.1/0.05 s substeps: pressure-difference refinement ratio ~0.499; finest pressure error against the closed-form constant-gamma blowdown reference <0.006%; maximum cumulative energy residual 0.000243 J. Initial gamma/molar mass use NeqSim. This is analytical/conservation evidence, not independent experimental qualification.
A reproduced native VU residual (0.999999563375 m3 for a 1 m3 target) was too large for strict accounting. The coupling explicitly refines the same VU equations to 1e-11 relative residual, bounded at 32 solves, and counts these solves in provenance. No core flash modification or weakened acceptance.
Java 8 API compile/class version 52, focused Javadoc doclint (no errors; private-field warnings), direct stable Spotless apply/check, both pre-commit stages, documentation search (730 Markdown + 1 HTML) and whitespace checks pass.
Publication tree 5d5242c40b5bc9d7e6a026656e5613d53caf8f22 matches the local validated tree, and all nine remote blobs were independently verified.
Provenance clarification for the preceding adapter run: its local adapter files matched the published blobs, but its supporting checkout was not a fully reconstructed current-master tree. The present full-tree reconstruction and 60-test run validate the merged adapter/support combination. Do not interpret the earlier local “exact base” wording as full-tree verification.

Acceptance and split boundary

Criterion group	Verified status
Architecture, explicit physics, neutral schema, immutable SI snapshots, steady/native dynamic sampling	Foundation merged (#3861/#3867/#3868/#3870); all evidence remains UNQUALIFIED.
Weighted uncertainty, failure-aware statistics, benchmark matrix	Merged #3878.
Explicit ideal-gas and legacy-screening adapters	Merged #3889; no silent model selection.
Two-way inventory release and conservation/refinement	Implemented and locally validated in draft #3893 for a rigid adiabatic single gas phase; not yet delivered on master.
Broader inventory, full-bore/non-equilibrium and solid-risk applicability	Outstanding.
Independent qualification/domain review and unresolved multicomponent flashing validation	Outstanding. Merge events do not supply missing independent review evidence.
Complete executed/rendered Colab demonstration and catalog/ledger	Outstanding.

The gas-only split is a physical/validation boundary: condensation, multiphase storage and phase-selective withdrawal require additional model assumptions and conservation/flash validation. Continuous receiving-pressure event location also remains unimplemented; current overshoot fails closed. No wall/fire heat, inflow or long-pipe model is implied.

Next dependency: keep #3893 as the only active implementation PR; inspect its exact-head CI and justified feedback, then obtain the required independent owner/domain review. After verified merge, choose the next dependency-ready broader-inventory/phase-boundary increment, coordinating with flash, dynamics and safety work. Do not duplicate the open solid-check repair #3892. Current owned autonomous occupancy is 4, below target 20 / ceiling 24.

Keep #3860 open. Closure still requires the complete merged implementation, reviewed evidence, current documentation and executed Colab demonstration.

Campaign ledger — PR #3893 review repair (2026-09-21)

Draft #3893 advanced from initial head f28a78eeda3bef8ac2b36cabf6d0004bac7e6be3 to exact repaired head f66fbef296be05c001210531f303cd677f916d39.

Code-quality review identified four missing @Override annotations on interface methods in two anonymous ReleaseFlowModel test doubles. The feedback was technically correct. The single scoped repair adds exactly those annotations; no production source, physical model, schema, documentation, numerical result or applicability boundary changed. All four review threads are resolved.

Repair validation

ReleaseInventoryTest: 13 tests pass.
Test source compiles with javac --release 8.
Direct Spotless apply repeated to stability; direct Spotless check passes.
Documentation search passes: 730 Markdown pages and 1 standalone HTML page.
Pre-commit and pre-push stages pass under the required Maven bootstrap.
Whitespace check passes.
The repaired remote blob 6c5b6d6051e74cc5750bac760be27df397bf8bfc matches the validated local test file.

Initial-head CI had reached 25 successes plus one not-applicable skip, with three platform jobs still active when superseded. Repaired-head CI is running; current completed gates include formatting, pre-commit, documentation detection and submission checks. GitHub reports the PR draft and mergeable. CodeQL aggregate is temporarily neutral while individual analyses are active, not a recorded source failure.

#3893 remains the sole active implementation PR for #3860. It is not READY TO MERGE: repaired-head CI and the independent safety/domain review required by GOVERNANCE.md remain pending. Keep the issue open. The frozen gas-inventory scope and later dependency order are unchanged.

Campaign ledger — PR #3893 dynamic-capability repair (2026-09-21)

Draft #3893 is open, mergeable, and remains the sole active #3860 implementation PR at exact head 991c8052fd81a5dbeccbba2c4f6349e2c7089af1 (tree eb06ec5e9d826d52a42d969dbee0e24bfa4ebc5b).

Branch-attributable CI defect repaired

The preceding head f66fbef296be05c001210531f303cd677f916d39 completed the Ubuntu fast suite with 14,850 tests and one failure: DynamicCapabilityBuiltInInventoryTest correctly found that ReleaseInventory was absent from the built-in dynamic-capability registry.
The failure was reproduced locally. The repair maps the inventory to DYNAMIC_LUMPED because it owns component masses and internal energy, while preserving UNVERIFIED runtime activation and UNQUALIFIED source-frame evidence.
Added explicit registry/runtime-report tests for both ProcessSystem and ProcessModel, isolation and steady requests; expanded the focused workflow trigger/gate; documented the mapping and applicability boundary.

Validated repair evidence

79 focused tests pass with zero failures/errors/skips.
Changed Java compiles with javac --release 8.
17 emitted frames validate and 10 corruptions are rejected.
Stable Spotless apply/check, documentation search (730 Markdown + 1 HTML), both pre-commit stages and whitespace checks pass.
The GitHub publication tree exactly matches the locally validated tree.

Current hosted state

25 checks have passed and three non-applicable jobs are skipped; source-terms, Javadoc, model specifications, formatting, documentation, notebooks/examples and all code analyses pass.
The Java 21 Windows fast job terminated its forked JVM in unrelated ReactiveElectrolyteEquilibriumTest after 119 tests with zero failures/errors (one skip). [AI-Assisted] feat(safety): couple gas releases to inventory depletion #3893 changes no chemical-reaction path, so this is classified as infrastructure/flakiness rather than a source defect.
Fail-fast cancelled the Ubuntu fast job. One slow shard is still active; the failed Windows job will receive one safe retry after the workflow completes. No source change is justified.

#3893 is not READY TO MERGE. Exact-head CI completion and the independent safety/engineering domain review required by GOVERNANCE.md remain blockers. Keep #3860 open; the later broader-inventory/full-bore/non-equilibrium, qualification, mixture solid-risk, multicomponent flashing and executed Colab dependencies are unchanged.

[AI-Assisted] feat(safety): couple gas releases to inventory depletion (

Completion audit and next implementation — 2026-09-21

Verified current master e4e9f27534870f510203512f912dd3aa2b9cde1d (tree a0a06655362a2181f2d576bda1ad3233567ce39a). The previous ledger is stale: #3893 has merged as 9c3ea2b7c01fa67b9665163d0583967c9413c4e7; conservative gas-inventory depletion and dynamic-capability registration are delivered. The merged PR's source-term, Java 21, model-specification, Javadoc, formatting and documentation checks pass; Java 8 platform checks were still running at audit time. The review API contains only an automated COMMENTED review, not independent domain approval.

The documented multicomponent flashing blocker remains reproducible on master. For SRK/classic 80 mol% propane + 20 mol% n-butane at 300 K, 20 bara to 3 bara, the HEM calculation returns INVALID at 823118.4346510068 Pa. Cold TP phase selection discards an incipient vapour phase; a nearby candidate also assigns the liquid cubic root to that trace phase, violating fugacity closure. Three new regression tests fail against the unmodified master implementation.

I am completing one bounded follow-up: guarded same-EOS vapour/liquid continuation within the HEM entropy root, retaining strict component/entropy/energy/fugacity checks, a Gibbs-energy comparison, explicit numerical diagnostics and model-version provenance. No core flash defaults or schema change. The original case and nine nearby composition/temperature cases now run; a separate saturation-path algorithm agrees with the critical pressure/rate. Full focused validation and publication gates are in progress.

This does not establish completion of #3860. The existing status document still requires broader inventory/pressure-event handling, separately assessed full-bore/non-equilibrium regimes, mixture solid-risk controls, independent experimental qualification/domain review and the executed/cataloged Colab demonstration. The full-bore/non-equilibrium work was expressly a separate model boundary in the original issue; this numerical fix does not claim those models. Frames remain UNQUALIFIED.

fix(safety): resolve incipient-phase mixture release entropy roots

Campaign ledger — multicomponent flashing repair published

The next implementation is draft #3903, branch codex/source-term-flashing-3860, parent e4e9f27534870f510203512f912dd3aa2b9cde1d, exact head 54512c1f3f82230babcc7a85d0496c69027b1902, tree daf989a853f8bb1226e8c0357a0c3b9fc9ff1dc3. GitHub confirms open/draft/mergeable; hosted CI is running, with no submitted reviews or inline threads at inspection. It is the sole active implementation PR for this campaign.

Implemented and validated

Guarded same-EOS phase-split continuation resolves the documented multicomponent entropy-root failure and trace-vapour cubic-root mismatch.
Explicit solver diagnostics, Gibbs/phase/component/fugacity checks, model version 1.1.0, unchanged schema v1 and retained UNQUALIFIED evidence.
Three new regression tests fail against the unmodified implementation. Final focused suite: 82 tests, zero failures/errors/skips.
Nine propane/butane composition/temperature cases pass, plus inventory scaling and choked/subcritical backpressure checks.
Original case: 1.617323766 kg/s, throat 823128.570 Pa, ambient gas mass fraction 0.205318976. A separate bubble-pressure/entropy algorithm gives 1.617323782 kg/s and 823128.538 Pa (relative rate difference 9.62e-9; pressure difference 0.0323 Pa).
18 emitted frames validated; 10 corruptions rejected. Java 8 class version 52; Javadoc doclint, repeated stable Spotless, both pre-commit stages, documentation search and staged/unstaged whitespace checks pass.
All six uploaded blobs and the complete publication tree match the locally validated files. Documentation explains the numerical method, rejection rules, evidence and remaining qualification limits; CI retains mixture/saturation-reference CSVs and the emitted frame.

The acoustic warning remains explicit: phase appearance gives a nonsmooth maximum with THROAT_MACH_MISMATCH (approximately Mach 0.1215). This is numerical/conservation evidence using the same SRK EOS, not independent experimental qualification.

Closure decision
Keep #3860 open. Merged gas-inventory work #3893 is delivered, but #3903 remains unmerged. Other remaining criteria are broader inventory/pressure-equilibration handling; separately assessed full-bore/non-equilibrium regimes; mixture-specific solid-risk controls; independent experimental qualification and safety/domain review under GOVERNANCE.md; and the executed, rendered, cataloged Colab demonstration. No approval or qualification is inferred from a merge or passing CI.

Next dependency: inspect #3903's exact-head CI and justified review feedback, obtain the required independent review, then merge before selecting the next physical-model increment. Do not mix broader release physics into this six-path numerical-repair PR.

@EvenSol

Campaign ledger — PR #3903 exact-head CI complete (2026-09-22)

Draft #3903 remains open and mergeable at exact head 54512c1f3f82230babcc7a85d0496c69027b1902, based on master e4e9f27534870f510203512f912dd3aa2b9cde1d.

Hosted qualification status

All 31 exact-head checks are complete: 30 passed and one non-applicable PaperLab matrix job was skipped.
Java 8 and Java 21 platform suites, all four Java 21 slow shards, the 82-test focused/local evidence suite through the source-term gate, model specifications, Javadoc, Spotless, documentation, examples/notebooks and CodeQL pass.
GitHub reports the PR mergeable. There are no submitted reviews or unresolved inline threads.
The PR description now records the exact hosted evidence; no source, numerical method, schema, documentation or evidence-level change was made in this follow-up.

#3903 is not READY TO MERGE because GOVERNANCE.md requires a safety/engineering owner plus an independent domain reviewer for this method. CODEOWNERS and MAINTAINERS.md currently list only @EvenSol for safety and leave the backup role unassigned, so automation cannot satisfy the independent approval requirement.

Keep #3860 open. Once #3903 is independently reviewed and merged, the next dependencies remain broader inventory/pressure-event handling, separately assessed full-bore/non-equilibrium regimes, mixture-specific solid-risk controls, independent experimental qualification, and the clean-executed/rendered/cataloged Colab demonstration.

Lifecycle prerequisite advanced: transactional cubic-root and amount isolation — 2026-10-21

Current master is fee802870c80a012ad3b59e4f7f87bc3a0088bc0. #3878/#4419 remain merged and no #3860 implementation PR is open. The two master commits after the owner's exact base 484759c96b65f1e58742372087f48092a77080f8 affect labyrinth-seal/refinery documentation and implementation, not SystemThermo, TPflash or TPmultiflash.

The new #4310 transactional root/amount investigation, with #4362 disposition, materially narrows the A4b prerequisite:

At the 0.01 mol, 194 K / 81.00 bara return, phase compositions can be correct while one phase retains the wrong cubic root. Evaluating fixed-composition root pairs identifies LIQUID/LIQUID as the consistent pair: max |Δ ln(f)| = 2.02e-14, G = -22.8586 J, versus a homogeneous-liquid reference of -22.8279643 J.
The split advantage is only 0.0306035 J at 0.01 mol. The existing absolute 1 J spurious-phase threshold can therefore hide the defect at 0.01 mol while collapsing the same intensive state at 1 and 100 mol, explaining the previous inventory-scale topology change.
A beta trial made through SystemInterface.clone() can still alter live active-beta storage on this path. Rejected or diagnostic trials require an explicit live-state snapshot/restore and immutable accepted values.
A bounded local transaction separates root selection from amount synchronization, requires normalized feed/phases and strict component lever-rule closure, restores live betas after every rejected trial, and limits accepted beta correction to 0.05. It passes 20/20 focused methods: the three-inventory nearby/return lifecycle, six sour-gas consistency cases, condensate conservation including the 750-point map, all four large-volatility controls and both tiny-aqueous controls.

This is strong local isolation evidence, not a published or campaign-qualified repair. Repository-wide Spotless, Javadocs, complete Java 8/21 and slow-shard regressions, documentation/search, security/CodeQL and hosted exact-head CI have not run; the reconstruction runtime lacked javadoc. No performance or accountable domain-review claim exists.

Next dependency: #4310 retains shared-flash ownership. Implement the same snapshot/restore plus bounded root/amount transaction in a full exact-master checkout, add the frozen lifecycle regression, run every repository gate and publish a draft only if the complete matrix remains green. The source-term campaign must not duplicate this fix or start broader inventory coupling before that assessed result merges.

The frozen next source-term increment remains phase-level interfacial heat/latent and component-transfer feedback with geometry/entrainment, strict component/energy/volume closure, positivity/events, atomic rollback/refinement and both native process containers. Independent experimental qualification and accountable safety/domain review also remain open. Keep #3860 open and UNQUALIFIED.

Documentation impact: issue evidence only; current completion-status and guide boundaries remain accurate. No source, API, schema, model, notebook, branch, PR, readiness, merge, closure or automation-state mutation was made by this run. Target 20 / absolute ceiling 24 remains the governing user policy; capacity is not the blocker.

Lifecycle prerequisite published as draft #4502 — 2026-10-21

Current protected master is 74a0529d19db577cb41034685905067a8ac9d0c6. The merged interfacial-tension repair #4499 is included in that base; its SystemThermo.clone() change deep-clones and rebinds interfacial properties, while the lifecycle repair itself remains isolated to TPflash.java plus regressions and generated coverage.

The previously unpublished #4310 transaction has materially advanced to draft #4502, exact head 8b8a233254382ff784b60e24b68a01198f805a72, exact base 74a0529d19db577cb41034685905067a8ac9d0c6. The current five-file batch:

independently qualifies fixed-composition cubic-root pairs using strict fugacity equilibrium;
snapshots/restores live phase fractions around clone trials;
applies a separate bounded lever-rule amount transaction only for eligible neutral, nonchemical two-phase endpoints;
preserves the frozen 1e-12 normalization, 1e-10 material, 1e-8 fugacity and 0.05 maximum-beta-correction gates;
adds the three-inventory pressure-return regression, PR/SRK/CPA root-reuse/isolation checks, and corrects the independently adjudicated 194 K / 78.5 and 81 bara expectations from homogeneous oil to the lower-Gibbs conserved split.

Unmodified master fails the new inventory regression by recovering methane 0.7000037987108007 instead of 0.7. The exact draft reports 55/55 focused methods passing on the default build and Java-8-target profile, covering the 750-point condensate map, sour gas, large volatility, cubic roots, liquid-liquid, trace-water and aqueous controls. The optimization reduces the local root-comparison helper from 32.63 to 17.34 microseconds and allocation from about 199000 to 99560 bytes; 45 before/after endpoint snapshots match exactly. Complete-flash timings vary across forks, so this is not claimed as a significant end-to-end speedup.

At this ledger update the exact-head check set has 26 successes, one conditional skip and two active Java 21 fast lanes, with no failure. All eight Java 21 slow shards and their aggregate, Javadocs, Java 8/21 model specifications, benchmark, Spotless/pre-commit, coverage and CodeQL have passed. Actual Java 8 runtime execution has not been reported for this head. There are no submitted human reviews or inline review threads. Therefore #4502 remains draft / VALIDATION PENDING EXACT-HEAD CI and is not READY TO MERGE.

Dependency boundary: do not duplicate #4502 or start the broader phase-level inventory increment while its two fast lanes, accountable thermodynamics review and merge are outstanding. After a verified merge, rerun the source-term consumer/conservation matrix on the resulting master before resuming the frozen interfacial heat/latent/component-transfer inventory feedback with geometry/entrainment, atomic rollback/refinement and both native process containers.

#3860 remains open and UNQUALIFIED. Independent experimental qualification, solid-bearing/multiphase applicability evidence and accountable safety/domain review remain separate completion gates. Documentation impact: issue evidence only; the current completion-status page remains accurate.

Exact-head Java 8 timeout blocker and targeted retry — 2026-10-22

Protected master remains 74a0529d19db577cb41034685905067a8ac9d0c6. Draft #4502 remains open, mergeable and unchanged at exact head 8b8a233254382ff784b60e24b68a01198f805a72 on that base; its frozen five-file scope and the previously reported 55/55 focused local result are unchanged.

Hosted exact-head validation is now materially resolved from “pending” to a reproducible blocker:

five of six workflow groups pass; Pre-commit, Engineering Capability Coverage, Spotless, PaperLab and CodeQL are green;
Java 21 Ubuntu and Windows full suites pass with 17,693 tests per platform, all eight Java 21 slow shards and their aggregate pass, and Javadocs, Java 8/21 model specifications, benchmark accuracy and workflow guards pass;
both Java 8 full-test lanes completed with the same single error and no assertion failures: DoeBigHillAtmosphericVacuumIntegrationTest.connectedVacuumSensitivityQualifiesIndependentPoints exceeded its 300-second JUnit timeout;
each Java 8 lane reports 17,693 tests, 0 failures, 1 error, 209 skipped. The failing refinery integration test is not among [AI-Assisted] fix(tpflash): conserve multiphase roots and amounts #4502's changed files, but the common cross-platform timeout must still be adjudicated rather than assumed to be infrastructure-only.

A targeted rerun of the failed Windows Java 8 job has been started on the unchanged exact head. GitHub rejected simultaneously queuing the Ubuntu job because the workflow run was already active; Ubuntu remains to be retried after the current attempt becomes idle. No source, branch, tolerance, workflow or evidence-level mutation was made. If the timeout reproduces, the next action is focused Java 8 isolation and performance/root-path evidence, not another blind retry.

There are still no submitted reviews or inline review threads. #4502 therefore remains draft, exact-head CI red, and not READY TO MERGE. Accountable flash/thermodynamics review is independently outstanding.

Dependency boundary remains unchanged: do not start the broader #3860 phase-level inventory increment until #4502 passes its mandatory exact-head gates, receives accountable review and merges. #3860 remains open and UNQUALIFIED; the current completion-status document remains accurate.

#4820 merged; premature issue closure reversed after default-branch audit — 2026-11-03

Exact state. Protected master is 8830c7141f9561227e8f08a9c97cbc8c05d26eeb. #4820 merged exact head d926432553877c01e36bc9f6774450988429d0a2 as 41147116d5668b161e62c5afd15cb01b690bc08b. Its exact-head CI is 10/10 workflow groups and 19/19 Core jobs green. GitHub records 0 reviews / 0 unresolved threads. #3860 was closed with the merge at 2026-11-02 21:32 CET, but the required completion audit found open acceptance rows and no independent review; the issue has therefore been restored to OPEN / UNQUALIFIED.

Merged capability. Opt-in NonEquilibriumReleaseInventory retains gas/liquid named-component state, evaluates forced homogeneous EOS phases at solved common pressure/temperature, performs provenance-required donor-limited fugacity-film transfer, withdraws the live opening phase through supplier-neutral ReleaseFlowModel, and atomically closes component mass, rigid volume, released enthalpy, external heat and total internal energy. Deterministic evidence, rollback/replay, pressure events and real dynamic frames cover both ProcessSystem and ProcessModel; the equilibrium default and neutral schema remain unchanged. The merged repair registers the inventory as DYNAMIC_LUMPED; closed-opening activation/evidence remain UNVERIFIED/UNQUALIFIED.

Evidence. The head retains 73/73 focused tests, stable Spotless apply/check, generated coverage, documentation search (828 Markdown + 1 HTML), pre-commit/pre-push and git diff --check. Refinement at 0.04/0.02/0.01/0.005 s decreases pressure/partition differences; maximum energy residual is 2.503e-9 J. Documentation covers equations, use, applicability, limitations and status.

Why closure failed. Current default-branch status still requires: predictive/validated transfer coefficients and interfacial heat models; separate phase energy/temperature where needed; predictive entrainment; distributed pipe heat/elasticity and two-sided coupling; population-balance/breakup/coalescence plus liquid-droplet/annular/high-Weber/solid-bearing transport; independent dense-gas/multiphase datasets; independently validated solid-bearing physics; broader experimental flashing benchmarks; and accountable domain review. The executed Colab demonstration is complete, but cannot substitute for these rows. GOVERNANCE.md requires an independent domain reviewer; CODEOWNERS names only the author and MAINTAINERS.md leaves the safety backup unassigned. The merge has no submitted review.

Next coherent dependency. The next campaign batch is predictive/validated interfacial transfer with separate phase-energy treatment, retaining the merged component/volume/total-energy transaction, explicit SI/provenance, phase-disappearance failure, both process containers and timestep convergence. Predictive entrainment/population balance and solid-bearing/full-bore transport remain separate high-risk boundaries. No new branch was published in this run because the execution workspace contains no exact-master checkout for mandatory Maven, conservation/refinement, Spotless, documentation and hook gates; remote GitHub state remains authoritative.

Static distributed-pipe geometry evidence added; lumped source-term tranche remains separate — 2026-10-30

Exact lease refreshed immediately before this ledger write. Protected master is dd5e650ed6107234084517cd0147297dae825d65; #3860 remains OPEN / UNQUALIFIED.

Documentation/test-only #4745 merged unchanged from final head a6fd504f9545cc45c7dc58c5ee8e801d7aa38d6a as dd5e650ed6107234084517cd0147297dae825d65. Its exact head is terminally green (8/8 workflow groups) and executes the literal Java 8 process-safety quick start. It has 0 submitted reviews / 0 review threads and changes no production physics, source-term contract, notebook, ReleaseInventory, or completion status.
New draft #4750 is mergeable at exact head 17a0a6a14af5ee587a9103210807e50d00acdc29 on base dd5e650ed6107234084517cd0147297dae825d65. It adds a read-only distributed-TwoFluidPipe hydrostatic cell reconstruction with immutable start/end/min/max depths, integrated liquid volume, horizontal interface and wet-wall areas, minimum gas area, explicit units and residuals. The PR reports 3,060 independently regenerated static reference rows plus 125 focused tests; current exact-head CI is 8/9 workflow groups green, 1 active and 0 failed. GitHub records 0 submitted reviews / 0 review threads.
feat(pipeline): qualify read-only hydrostatic cell depth and area #4750 is useful full-bore geometry evidence, but its declared boundary is static and non-invasive: it does not alter transient inventory, phase appearance/deletion, force/transport closure, or mass-transfer laws. It explicitly leaves stable transient pools, mesh convergence, retained gravity, predictive evaporation/entrainment, heat supply and independently qualified transfer rates unqualified. It therefore does not own or complete State-of-the-art source-term platform for live safety-simulator input #3860's opt-in lumped non-equilibrium inventory coupling.
Default-branch campaign blobs are unchanged: ReleaseInventory e1473d835ec61144587ca9c52ec3a85dea160590, phase geometry e954c426444ffb789778e47d7ebe5e3a56bae7d2, coupled-inventory guide f03e90b949cfe48ba448ca5ff2fc4a3873731adb, and completion status 3aa7ccdf609f20ebf9b587028f5c05bc214463ad. No open PR matches ReleaseInventory, State-of-the-art source-term platform for live safety-simulator input #3860 or source-term scope.
Required next batch remains a separate opt-in lumped non-equilibrium ReleaseInventory using the merged fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711 film/EOS solver and fix(physicalproperties): preserve liquid diffusion pair invariants #4725 transport correction: conservative per-phase component and enthalpy state, geometry-derived live area, explicit film/heat-supply provenance, donor limiting, adaptive atomic steps, phase appearance/disappearance diagnostics, deterministic frames, both process containers and refinement/independent validation. Distributed TwoFluidPipe, Safety Shutdown, CPA, PVT and reporting-agent work remain outside that batch.
Maven bootstrap succeeds, but the allocated execution workspace still contains no NeqSim checkout, pom.xml, wrapper, Spotless runner or hooks. Mandatory compilation, conservation/nearby-case/refinement tests, Spotless, documentation search and hook gates cannot be run here; no unvalidated campaign branch was published.

#4732 exact-head CI is terminally green after merge; source-term completion remains unchanged — 2026-10-30

Exact lease refreshed immediately before this ledger write. Protected master is fd3d28142b2139691ce606e4cba5ab8241469486; #3860 remains OPEN / UNQUALIFIED.

Merged #4732 final head b84efa4bdc39df3fbd971c88d92f341ff9f08b07 now has 9/9 successful workflow groups. Core Build/Test/Javadoc is success: both Java 8 platform suites have finished successfully, alongside Java 21 Ubuntu/Windows, all eight slow shards and aggregate, Javadoc, model specifications, CodeQL, Spotless, documentation and coverage gates. No exact-head CI failure remains.
This is post-merge software evidence, not readiness or independent qualification: GitHub still records 0 submitted reviews and 0 review threads. The disabled 600 s severe-slugging case, physical pool-height refinement, predictive pool heat/area/mass-transfer closure and independent dense/multiphase evidence remain outside the merged claim.
New draft #4745 is documentation/test-only at a6fd504f9545cc45c7dc58c5ee8e801d7aa38d6a. It compiles and runs the literal Java 8 process-safety quick start from docs/process/safety/README.md; 7/8 workflow groups are currently green and Core is in_progress. It changes no production safety physics, source-term contract, notebook, ReleaseInventory, or completion status, and has 0 reviews / 0 threads.
Default-branch State-of-the-art source-term platform for live safety-simulator input #3860 blobs remain unchanged: ReleaseInventory e1473d835ec61144587ca9c52ec3a85dea160590, phase geometry e954c426444ffb789778e47d7ebe5e3a56bae7d2, coupled-inventory guide f03e90b949cfe48ba448ca5ff2fc4a3873731adb, completion status 3aa7ccdf609f20ebf9b587028f5c05bc214463ad.
Maven bootstrap succeeded, but this workspace still contains no NeqSim checkout or validation scripts. No State-of-the-art source-term platform for live safety-simulator input #3860 source edit, branch, commit, ref or PR was published.

Disposition: terminal #4732 CI removes the last software-check uncertainty from that merged head, but it does not satisfy #3860's independent review or missing lumped non-equilibrium inventory acceptance items. The next dependency remains the separate default-off ReleaseInventory per-phase component/enthalpy coupling with geometry/film/heat provenance, adaptive atomic stepping, convergence evidence and both process-container paths. Do not close #3860.

Shared compressor repair and distributed wet-source work merge; #3860 lumped inventory scope is clear — 2026-10-30

Exact lease refreshed immediately before this ledger write. Protected master is now a01d663990a8f43de366a8c02f1ea6280b61c143. Issue #3860 remains OPEN / UNQUALIFIED.

fix(compressor): resolve sub-watt wet-gas energy closure #4739 merged first: exact head 88fd44a2b65f3e882ad527edcaab605a01a99cdb merged as 8a2438b99b9a2ac03e2d923406d3660cc98a59f8. All 9/9 exact-head workflow groups pass, including Core Build/Test/Javadoc. The repair closes differential wet-gas sub-watt EOS energy increments by refining equilibrated outlet enthalpy without changing the public API, global flash tolerances, or the existing 1e-5 W assertion. GitHub records 0 submitted reviews and 0 review threads, so the software evidence is terminally green but not independently reviewed.
fix(pipeline): transient handoff and conservative wet component sources #4732 then merged on top of fix(compressor): resolve sub-watt wet-gas energy closure #4739: final head b84efa4bdc39df3fbd971c88d92f341ff9f08b07, base 8a2438b99b9a2ac03e2d923406d3660cc98a59f8, merge commit f36187351caa620cb8e271b19ee4c6fd1f9545f9. The prior branch-attributable severe-slugging detector failure is repaired; Java-21 Ubuntu/Windows fast suites, all eight slow shards and their aggregate, Java 8/21 model specifications, Javadoc, CodeQL, formatting, documentation and coverage workflows have passed. Exact-head Core remains in_progress because Assert tests with Java 8 on Windows and Assert tests with Java 8 on Ubuntu are still active; current failed jobs: none. The PR was merged before its exact-head Java-8 suites reached terminal status.
No independent acceptance review: fix(pipeline): transient handoff and conservative wet component sources #4732 has 0 submitted reviews and 0 review threads. Its merged evidence advances conservative distributed gas/oil/aqueous component sources, trace-phase momentum, reversed-outlet traction, severe-slugging measurement, and network component/phase/time ledger consistency. It does not qualify predictive interfacial kinetics, entrainment, stationary pool area/heat supply, physical evaporation time, long-horizon 600 s severe-slugging behavior, local pool-height resolution, independent dense/multiphase data, or the State-of-the-art source-term platform for live safety-simulator input #3860 lumped inventory feedback contract.
Default-branch source-term audit remains unchanged: ReleaseInventory blob e1473d835ec61144587ca9c52ec3a85dea160590; phase geometry e954c426444ffb789778e47d7ebe5e3a56bae7d2; coupled-inventory guide f03e90b949cfe48ba448ca5ff2fc4a3873731adb; completion status 3aa7ccdf609f20ebf9b587028f5c05bc214463ad. Thus neither merge implements the missing opt-in non-equilibrium ReleaseInventory state or changes the issue completion matrix.
Portfolio/overlap: only 1 PR is currently open (test(thermo): stage bidirectional CO2-HV diagnostics and performance evidence #4744 test(thermo): stage bidirectional CO2-HV diagnostics and performance evidence), below the user's 20-target/24-ceiling. No open PR claims State-of-the-art source-term platform for live safety-simulator input #3860 or ReleaseInventory, so the separate lumped batch is dependency-ready and overlap-clear.
Local publication gate remains blocked: Maven bootstrap succeeded, but this workspace still has no NeqSim checkout, pom.xml, Maven wrapper, Spotless runner or hook configuration. Mandatory reproduction, component/energy conservation and timestep-refinement tests, direct Spotless apply/check, documentation search and hook stages cannot run. No unvalidated branch, commit, ref or PR was published.

Next coherent #3860 batch: integrate the merged gas/oil film/EOS and corrected-diffusivity foundations into a default-off lumped ReleaseInventory non-equilibrium mode with per-phase component and enthalpy inventories, geometry-derived interface area, explicit film/heat-supply provenance, donor limiting, adaptive atomic transactions, phase-appearance/disappearance diagnostics, deterministic machine-readable frames, and both ProcessSystem/ProcessModel paths. Keep distributed TwoFluidPipe, Safety Shutdown, CPA boundary and compressor-repair scope excluded.

Disposition: do not close #3860 and do not claim READY TO MERGE. Remaining acceptance gates include the lumped inventory-feedback implementation and convergence evidence, predictive/validated entrainment and interfacial heat/latent/mass transfer, broader full-bore/non-equilibrium qualification, solid-bearing release physics and mixture-specific validation, independent multicomponent flashing benchmarks, accountable domain review, and the final executed demonstration/default-branch audit.

Severe-slugging detector blocker is repaired; #4732 now waits on terminal CI, review, and shared #4739 — 2026-10-30

Exact lease refreshed immediately before this ledger write. Protected master remains 9975919e7908fe4ef34bbfeca37055d8883116fa; #3860 is OPEN / UNQUALIFIED. The source-term completion-status blob remains 3aa7ccdf609f20ebf9b587028f5c05bc214463ad, so no #3860 acceptance item moved to completed default-branch evidence.

fix(pipeline): transient handoff and conservative wet component sources #4732 materially advanced on its existing branch: exact head c57c8f863068ab21b51662e65c6c6178ee0ba4aa, base 9975919e7908fe4ef34bbfeca37055d8883116fa, draft and mergeable. The former branch-attributable slow-shard blocker was traced to the benchmark's global-extrema event detector: one large reverse-flow excursion shifted the threshold below real blowout/fallback events. The test-only repair retains the original 16-cell, 0.1 s, 180 s trajectory, pressure/conservation checks and experimental thresholds; it uses the existing 75%/125% feed-relative fallback/blowout gates, requires recovery between events, and keeps the ten-second secondary-minimum merge.
Hosted detector evidence is now green: 8/8 non-Core workflow groups pass; all eight Java-21 slow shards pass, including shard 3/8 and the aggregate gate. Java 8/21 model specifications, Javadoc, agent-skill references and the agentic benchmark pass. Core is still in_progress because Ubuntu and Windows Java-21 fast suites remain active at this snapshot. Local evidence recorded by the PR is 83 cases with 82 passes and one pre-existing disabled 600 s qualification, plus six detector controls, Java-8 compilation, stable Spotless apply/check, documentation search, Checkstyle, Javadoc and hooks.
Additional boundary-ledger evidence at the same head: the new network regression verifies named-component mass against hydrodynamic phase mass for initial/final inventory and signed inlet/outlet transfer, exact elapsed-time agreement, and accepted-substep agreement across phase/component/energy reports. It passes on unchanged master, the preceding fix(pipeline): transient handoff and conservative wet component sources #4732 head, and the final branch, ruling out canonical storage-node merging, branch boundary ledgers and repeated branch solves as the observed zero-storage-junction mechanism. This is evidence, not new source-term coupling.
Shared compressor blocker moved to fix(compressor): resolve sub-watt wet-gas energy closure #4739: #4739 is the single owner for the remaining sub-watt wet-gas EOS energy-closure failure left after fix: scale compressor head inversion tolerance for tiny shaft duties #4735. Exact head 88fd44a2b65f3e882ad527edcaab605a01a99cdb, base 9975919e7908fe4ef34bbfeca37055d8883116fa, open and mergeable. It refines differential EOS outlet enthalpy closure without changing public APIs, global flash tolerances or the unchanged 1e-5 W assertion. Local evidence reports 48 focused tests on Java 17/21 and 606 broader compressor/shaft/expander tests passing. 8/8 non-Core workflows and all eight slow shards pass; Core remains in_progress with Java-21 fast suites active.
Review and qualification remain absent: fix(pipeline): transient handoff and conservative wet component sources #4732 has 0 submitted reviews and 0 review threads; fix(compressor): resolve sub-watt wet-gas energy closure #4739 has 0 submitted reviews and 0 review threads. fix(pipeline): transient handoff and conservative wet component sources #4732's disabled 600 s experimental case, large signed-flow excursions, pressure-correction limits and independent liquid-rich/full-bore validation remain open.
Portfolio/overlap: 6 PRs are open in the repository, below the user's 20-target/24-ceiling. Searches show no open PR explicitly claiming ReleaseInventory or State-of-the-art source-term platform for live safety-simulator input #3860's separate lumped non-equilibrium inventory scope. fix(pipeline): transient handoff and conservative wet component sources #4732 remains distributed TwoFluidPipe work and fix(compressor): resolve sub-watt wet-gas energy closure #4739 remains shared compressor repair; neither replaces the next State-of-the-art source-term platform for live safety-simulator input #3860 tranche.
Local publication gate remains blocked: Maven bootstrap succeeded again, but this workspace still contains no NeqSim checkout, pom.xml, wrapper, Spotless runner or hook configuration. Required reproduction, component/energy conservation and refinement tests, direct Spotless apply/check, documentation search and pre-commit/pre-push cannot be executed here. No branch, commit, ref or PR write was attempted.

Next #3860 dependency: retain the separate opt-in lumped non-equilibrium ReleaseInventory batch using merged film/EOS and corrected-diffusivity foundations, with per-phase component/enthalpy state, geometry-derived area, explicit film/heat provenance, donor limiting, adaptive atomic steps, phase-appearance diagnostics, deterministic frames and both process containers. Do not absorb distributed pipe, Safety Shutdown or compressor-repair scope.

Disposition: the severe-slugging CI blocker changed from failing to passing on #4732, but neither #4732 nor #4739 is READY TO MERGE. Do not close #3860. Remaining issue-level gates continue to include inventory-feedback interfacial transfer and entrainment, broader full-bore/non-equilibrium capability, solid-bearing physics/validation, independent multicomponent flashing benchmarks, accountable domain review, and final default-branch demonstration audit.

CPA and Safety Shutdown foundations merge; shared compressor repair lands while #4732 remains blocked — 2026-10-30

Exact lease refreshed immediately before this ledger write. Protected master is 9975919e7908fe4ef34bbfeca37055d8883116fa. Issue #3860 remains OPEN / UNQUALIFIED. The audited source-term completion-status and ReleaseInventory/geometry/guide blobs remain unchanged from the preceding audit, so these merges do not complete the source-term acceptance matrix.

feat(thermo): qualify CPA boundaries and hydrate amount flashes #4731 merged: CPA boundary and hydrate-amount flash APIs are now on master at b7ae1f47f6e7f188f1b0bb76fd3af4079785c957 (PR head d49328a78453c3e72686b8511a1c7851d1673222). Its exact PR head retained 8/9 green workflow groups; Core was red from the inherited Java-8 tiny-compressor-duty assertion. It had no submitted reviews or review threads. This provides a thermodynamic boundary foundation, not source-term station integration, solid-bearing release physics, independent mixture validation, or domain qualification.
[AI-Assisted] docs: execute diffusivity guide example #4733 merged: the executed diffusivity guide is on master at fc837905d47db7d82e8da44df9bb2e32e91cae3d. It documents a foundation but does not implement the remaining State-of-the-art source-term platform for live safety-simulator input #3860 coupling/qualification gates.
feat(pipeline): add bounded Safety Shutdown pump-curve acoustics #4734 merged: Safety Shutdown pump-curve acoustics is on master at ff0550c0d82b57392db9dff58fd553681adcf4a8 (PR head debe3ce0219e74d087454f0b7b4881d35a1204f5). This is deliberately separate from State-of-the-art source-term platform for live safety-simulator input #3860 and does not supply the lumped release/inventory coupling.
fix: scale compressor head inversion tolerance for tiny shaft duties #4735 merged as the single shared repair: compressor head-inversion convergence now scales to tiny requested head at 9975919e7908fe4ef34bbfeca37055d8883116fa (PR head c0dd152f9fbaa97b59d5821b8de9109eddf86cbe), avoiding duplicate fixes across feat(thermo): qualify CPA boundaries and hydrate amount flashes #4731/fix(pipeline): transient handoff and conservative wet component sources #4732/[AI-Assisted] docs: execute diffusivity guide example #4733/feat(pipeline): add bounded Safety Shutdown pump-curve acoustics #4734. At this snapshot, 8/8 non-Core workflow groups are green and Core is in_progress; all eight slow shards and Java 8/21 model-specification jobs are green, while Java-21 fast jobs remain active. No test tolerance was weakened. The repair was merged before the exact-head Core workflow finished and has no independent review, so it is not reusable as a claim that every affected PR head was fully green.
fix(pipeline): transient handoff and conservative wet component sources #4732 remains the sole open implementation PR: exact head 5b813ab39104e4eb3ebf4dec3d097f5753477d54, base 9975919e7908fe4ef34bbfeca37055d8883116fa, draft and mergeable. 8/8 non-Core workflow groups pass. Core is in_progress with slow shard 3/8 deterministically failing SevereSluggingExperimentalBenchmarkTest.reproducesTheLiquidCycleAndTheRiserPressureSwing; the other seven slow shards pass, and Java-21 fast jobs are still active. This is a branch-relevant physical/numerical regression, not infrastructure, and it blocks readiness. Submitted reviews: none; unresolved review threads: none.
State-of-the-art source-term platform for live safety-simulator input #3860 implementation publication remains locally blocked: this workspace contains no NeqSim source checkout, pom.xml, Maven wrapper, or Spotless runner. The mandated Maven bootstrap helper ran successfully, but there is no source tree on which to reproduce, edit, run focused conservation/refinement tests, apply/check Spotless, run documentation search, or execute pre-commit/pre-push. No branch, commit, ref, or PR write was attempted.

Next dependency-ready #3860 increment remains unchanged: a separate opt-in lumped non-equilibrium ReleaseInventory coupling using merged gas/oil film/EOS and corrected diffusivity foundations, with per-phase component and enthalpy inventories, geometry-derived interface area, explicit film/heat provenance, donor limiting, adaptive atomic steps, phase-appearance diagnostics, deterministic frames, and both ProcessSystem/ProcessModel integration. Distributed TwoFluidPipe, Safety Shutdown water-hammer/acoustics, and CPA boundary implementation remain outside that PR.

Acceptance disposition: do not close #3860 and do not report readiness. Remaining gates still include predictive/validated entrainment and coupled interfacial heat/latent/mass transfer with inventory feedback; broader full-bore/non-equilibrium capabilities; solid-bearing release physics and mixture-specific validation; independent multicomponent flashing benchmarks; complete independent qualification/domain review; and a final default-branch audit of the executed demonstration.

CPA CI inherits a master Java-8 failure; distributed wet-source repair exposes a severe-slugging regression — 2026-10-30

Protected master remains 1929a43a8f43. The audited State-of-the-art source-term platform for live safety-simulator input #3860 blobs are unchanged: ReleaseInventory e1473d835ec6, phase geometry e954c426444f, coupled-inventory guide f03e90b949cf, and completion status 3aa7ccdf609f.
Draft #4731 is still open, mergeable and unchanged at exact head d49328a78453. 8/9 workflow groups pass; Build/Test/Javadoc is terminally red because both Java 8 platform jobs fail the same unrelated compressor assertion: CompressorChartPolytropicMethodTest.smallWetGasPowerRetainsFinitePressureRise expected 0.663991527050103 but obtained 0.6638522259891033. The identical two-platform Java 8 failure already occurs on merged #4721 exact head 0b1348da4829, while all Java 21, slow-test, Javadoc, model-specification and CPA-related jobs pass. This is an inherited protected-master regression, not a feat(thermo): qualify CPA boundaries and hydrate amount flashes #4731 CPA-path failure or infrastructure outage; mandatory CI is nevertheless red and GitHub records 0 reviews and 0 threads, so feat(thermo): qualify CPA boundaries and hydrate amount flashes #4731 is not ready or independently qualified.
Draft #4732 advanced from 8f6d974fbe0f to exact head 39370538f6cb without weakening thresholds. The correction now transports momentum with accepted pressure-correction mass, separates advection from pressure traction and preserves trace-phase pressure acceleration. The original slow 1/8 reverse-outlet and slow 8/8 rich-gas failures now pass, and the PR reports 723 affected tests passing locally plus focused analytical/oracle checks.
fix(pipeline): transient handoff and conservative wet component sources #4732 is still blocked: 8/9 workflow groups pass, but slow shard 3/8 fails SevereSluggingExperimentalBenchmarkTest.reproducesTheLiquidCycleAndTheRiserPressureSwing because no repeated liquid cycle appears for 16 sections, 0.1 s steps, (10^{-12}) perturbation and 180 s. The aggregate slow gate therefore fails and 2 Java 8 jobs remain active. This is a physical/numerical regression, not infrastructure. GitHub records 0 reviews and 0 threads. Do not claim fix(pipeline): transient handoff and conservative wet component sources #4732 as qualified full-bore/non-equilibrium evidence until the unchanged severe-slugging criterion has a technically justified terminal disposition.
New draft #4734 separately owns a bounded forward pump-curve acoustic boundary for Safety Shutdown at debe3ce0219e. It changes WaterHammerPipe/study paths, not ReleaseInventory, source-term frames or interfacial transfer. Its current hosted status is 8/9 workflow groups passing, with Build/Test/Javadoc active and no observed failed workflow; GitHub records 0 reviews and 0 threads. Keep this scope excluded from State-of-the-art source-term platform for live safety-simulator input #3860 implementation.
Open-PR searches still find no PR explicitly claiming State-of-the-art source-term platform for live safety-simulator input #3860 or ReleaseInventory. The distinct next State-of-the-art source-term platform for live safety-simulator input #3860 batch remains the opt-in lumped non-equilibrium inventory state using merged fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711 film/EOS transfer and terminally green fix(physicalproperties): preserve liquid diffusion pair invariants #4725 transport. Distributed pipe repair (fix(pipeline): transient handoff and conservative wet component sources #4732), CPA boundary implementation (feat(thermo): qualify CPA boundaries and hydrate amount flashes #4731) and Safety Shutdown acoustics (feat(pipeline): add bounded Safety Shutdown pump-curve acoustics #4734) remain separate owners.
Maven bootstrap succeeds, but this run's allocated workspace still has no NeqSim repository tree, Maven wrapper or validation scripts. Mandatory compilation, component/energy conservation and refinement tests, Spotless, documentation search and hooks cannot run here; no unvalidated State-of-the-art source-term platform for live safety-simulator input #3860 branch was published.

Issue #3860 remains open and UNQUALIFIED. Closure still requires merged default-branch implementation for the remaining physics plus independent experimental/dense-fluid/multiphase evidence and accountable domain review.

Liquid-transport CI completes; CPA and distributed wet-source work advance — 2026-10-30

Protected master is 1929a43a8f43. The audited State-of-the-art source-term platform for live safety-simulator input #3860 blobs remain unchanged: ReleaseInventory e1473d835ec6, phase geometry e954c426444f, coupled-inventory guide f03e90b949cf, and completion status 3aa7ccdf609f.
Merged #4725 is now terminal on exact head 032308935e80: 8/8 workflow groups pass, with no failed or active workflow. GitHub still records 0 submitted reviews and 0 review threads. Its corrected Wesselingh–Krishna liquid diffusivities are therefore software-validated for reuse by the merged fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711 film/EOS solver, but remain independently unqualified.
Draft #4731 repaired its earlier agent-lint failure and master conflict without changing ownership. It is open, mergeable and exact-head at d49328a78453 on base 1929a43a8f43. 8/9 workflow groups pass and Build/Test/Javadoc remains active; inside that workflow, 17 jobs pass, 2 remain active and 0 fail. The two active jobs are Java 8 Ubuntu and Windows. GitHub records 0 reviews and 0 threads. This removes the prior lint blocker but does not yet provide terminal CI, independent domain review, source-term station integration or solid-bearing release physics.
New draft #4732 is open and mergeable at exact head 8f6d974fbe0f on base 1929a43a8f43. It owns distributed TwoFluidPipe transient handoff and conservative gas/oil/aqueous component relaxation; it does not modify ReleaseInventory or the source-term contract. 8/9 workflow groups pass while Build/Test/Javadoc remains active with 3 failed jobs and 2 active jobs. The failures are physical/numerical, not infrastructure: the slow 8/8 rich-gas drag qualification produces an oil velocity of (1.3472\times10^{11}) m/s before rejecting the step, and slow 1/8 does not observe the required negative reverse-outlet component transfer. GitHub records 0 reviews and 0 threads. No acceptance threshold should be weakened; this distributed-pipe scope must reach a technically justified terminal disposition before State-of-the-art source-term platform for live safety-simulator input #3860 claims it as full-bore/non-equilibrium evidence.
Open-PR searches find no PR explicitly claiming State-of-the-art source-term platform for live safety-simulator input #3860 or ReleaseInventory. The State-of-the-art source-term platform for live safety-simulator input #3860 campaign must avoid duplicating fix(pipeline): transient handoff and conservative wet component sources #4732's distributed pipe/component-source work and feat(thermo): qualify CPA boundaries and hydrate amount flashes #4731's CPA boundary work. The distinct next batch remains an opt-in lumped non-equilibrium ReleaseInventory using merged fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711 film/EOS transfer and terminally green fix(physicalproperties): preserve liquid diffusion pair invariants #4725 transport, with conservative phase component/enthalpy inventories, geometry-derived interfacial area, explicit heat/film provenance, adaptive atomic substeps, phase-appearance diagnostics, deterministic frames and both process containers.
Maven bootstrap succeeded, but this run's allocated workspace still contains no NeqSim repository tree, Maven wrapper or validation scripts. Other scratch worktrees were not reused. Mandatory compilation, conservation/energy/refinement tests, Spotless, documentation search and hook gates cannot run here; no unvalidated branch was published.

Issue #3860 remains open and UNQUALIFIED. Independent experimental/dense-fluid/multiphase evidence and accountable domain review still block closure.

Acoustic inventory and liquid-transport prerequisites merge; CPA solid-risk work remains active — 2026-10-29

Protected master is ed35e0a938ce. The audited State-of-the-art source-term platform for live safety-simulator input #3860 blobs are unchanged: ReleaseInventory e1473d835ec6, phase geometry e954c426444f, coupled-inventory guide f03e90b949cf, and completion status 3aa7ccdf609f.
#4724 merged unchanged from exact head 23a94e08fc5a as 5078e1e1a0ff. Exact-head CI is terminal: 9/9 workflow groups pass. GitHub records 0 submitted reviews and 0 review threads, so the linear-acoustic kernel has complete software evidence but no independent qualification. It remains separate from ReleaseInventory, multiphase/EOS mass-energy, process coordination and source-term contracts.
#4725 merged from exact head 032308935e80 as 1715ae9c4921. It repairs liquid pair-diffusivity symmetry, insertion-order dependence, units and clone isolation using the documented Wesselingh–Krishna generalized Vignes approximation. This is a real prerequisite improvement for fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711-based multicomponent film transfer: the future State-of-the-art source-term platform for live safety-simulator input #3860 inventory adapter must consume the corrected transport rather than calibrating against defective diffusivities.
fix(physicalproperties): preserve liquid diffusion pair invariants #4725 exact-head hosted evidence is 7/8 workflow groups passing, 1 active and 0 failed; Build/Test/Javadoc remains active. GitHub records 0 reviews and 0 threads. Do not treat the corrected transport as terminally validated or independently qualified until that run and review disposition complete.
New draft #4731 is open and mergeable at 764f3f6b0660. It owns CPA liquid-water/ice/hydrate onset calculations, failure isolation and agent/MCP routing. Current hosted status is 7/9 workflow groups passing, 1 active and 1 failed; Skills and Agents Lint fails and Build/Test/Javadoc remains active. There are 0 reviews and 0 threads.
feat(thermo): qualify CPA boundaries and hydrate amount flashes #4731 does not integrate solid/hydrate onset into source-term stations, add solid-bearing release transport, provide independent mixture datasets or supply accountable domain review. It therefore cannot satisfy State-of-the-art source-term platform for live safety-simulator input #3860's solid-risk completion row in its current scope, but State-of-the-art source-term platform for live safety-simulator input #3860 should reuse its boundary API after terminal disposition rather than duplicate CPA phase-boundary logic.
Draft [AI-Assisted] test(process): qualify separator pressure-control inventory #4726 is documentation/test-only qualification of an existing separator pressure-control chain; it changes no production source and does not own ReleaseInventory or source-term physics. Open-PR searches still find no active PR claiming 3860, ReleaseInventory, interfacial-transfer or mass-transfer scope.
Next dependency: retain the frozen opt-in lumped non-equilibrium ReleaseInventory tranche, reuse merged fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711 and corrected fix(physicalproperties): preserve liquid diffusion pair invariants #4725 transport after terminal exact-head validation, and exclude Safety Shutdown, distributed pipe and CPA boundary implementation. Solid-risk station integration remains a later distinct batch after feat(thermo): qualify CPA boundaries and hydrate amount flashes #4731 reaches terminal disposition.
Maven bootstrap succeeded, but this run's allocated workspace still has no NeqSim repository tree, Maven wrapper or validation scripts. Other scratch worktrees belong to separate campaigns and were not reused. Mandatory compilation, conservation/refinement tests, direct Spotless, documentation search and hook gates could not run; no unvalidated State-of-the-art source-term platform for live safety-simulator input #3860 branch was published.

Issue #3860 remains open and UNQUALIFIED.

Safety Shutdown acoustic inventory gains a distinct active owner — 2026-10-29

Protected master is 7b31cbc4057d. Merged #4723 adds the proposal-only Safety Shutdown conservative acoustic-inventory RFC; it does not modify the audited State-of-the-art source-term platform for live safety-simulator input #3860 implementation or documentation blobs.
Adjacent draft #4724 is open and mergeable at exact head 23a94e08fc5a on base cc35f058e607. It owns cell-integrated linear-acoustic storage, physical face fluxes and conservation diagnostics inside WaterHammerPipe/WaterHammerStudy, with synthetic refinement and standing-wave evidence.
fix(pipeline): conserve Safety Shutdown acoustic cell inventory #4724's changed paths do not include ReleaseInventory, ReleaseInventoryPhaseGeometry, source-term frames/schema, ProcessSystem/ProcessModel coupling, or the merged fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711 gas/oil film solver. Its PR explicitly excludes multiphase/TwoFluidPipe, process coordination and full EOS mass/energy. The State-of-the-art source-term platform for live safety-simulator input #3860 lumped thermodynamic inventory tranche therefore remains dependency-ready, but must not duplicate or edit the active Safety Shutdown acoustic kernel.
Exact-head hosted evidence is not terminal: 8/9 workflow groups pass and 1/9 remains active. The Build/Test/Javadoc workflow still has the visible Java 21 Ubuntu and Windows fast suites in progress; no exact-head failure is recorded. GitHub records 0 submitted reviews and 0 review threads. Proposal acceptance, owner plus independent domain/contract review, compatibility classification and terminal CI remain mandatory for fix(pipeline): conserve Safety Shutdown acoustic cell inventory #4724.
Current-master source-term blobs remain unchanged: ReleaseInventory e1473d835ec6, phase geometry e954c426444f, coupled-inventory guide f03e90b949cf, and completion status 3aa7ccdf609f. Searches still find no open PR claiming 3860, ReleaseInventory or interfacial-transfer scope.
State-of-the-art source-term platform for live safety-simulator input #3860 next tranche is unchanged: a separate opt-in lumped non-equilibrium gas/oil inventory state reusing merged fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711, with conservative per-phase component/enthalpy accounting, geometry-derived area, explicit stationary-film and heat-supply provenance, donor limiting, atomic adaptive steps, diagnostics, deterministic frames and both process containers. Predictive entrainment/population balance, distributed pipe behavior, Safety Shutdown and independent qualification remain separate.
Maven bootstrap succeeded, but this run's allocated workspace still contains no NeqSim repository tree, wrapper or validation scripts. Other scratch worktrees are separate campaign state and were not reused. Mandatory compilation, conservation/refinement tests, direct Spotless, documentation search and hooks could not run, so no unvalidated State-of-the-art source-term platform for live safety-simulator input #3860 branch was published.

Issue #3860 remains open and UNQUALIFIED.

Separator V/U rollback is already merged; no new #3860 blocker — 2026-10-29

Protected master is c02e7cf52cf7. The commits after 491cabffbca9 are docs: correct Python runtime prerequisites #4713–fix(plugin): stop writing user mcp.json on VS Code 1.141+ #4717 documentation, measurement, pipeline, optimization and plugin work; they do not change the audited State-of-the-art source-term platform for live safety-simulator input #3860 source-term paths.
The later cross-campaign comment about #4583 is not a new unresolved prerequisite. Separator transient consumes rejected VU states while variance smoke test passes #4583 is closed as completed by merged #4646, which moved from exact head 8d37f0d0898e to merge 8a2b0a9dba9f. All 10/10 exact-head workflow groups pass. GitHub records 0 submitted reviews and 0 review threads, so this is software evidence, not independent qualification.
fix(separator): reject unclosed transient VU states #4646 supplies fail-closed separator V/U acceptance and identity-preserving transaction rollback for two- and three-phase separators. It does not add non-equilibrium ReleaseInventory phase transfer, stationary-film transport, interface/heat-supply closure, predictive depletion or source-term qualification. It therefore strengthens the reusable process transaction seam without changing the frozen State-of-the-art source-term platform for live safety-simulator input #3860 model boundary.
Current-master blobs remain unchanged: ReleaseInventory e1473d835ec6, ReleaseInventoryPhaseGeometry e954c426444f, coupled-inventory guide f03e90b949cf, and completion-status document 3aa7ccdf609f.
Open-PR searches still return no PR claiming 3860, ReleaseInventory or interfacial-transfer scope. Open fix(pipeline): honor exact Safety Shutdown valve travel knots #4720/fix(compressor): honor selected polytropic method in chart mode #4721/feat(pipeline): TwoFluidPipe liquid holdup calibration multiplier #4722 and documentation docs: explain analyser sample-system screening #4719 do not edit these paths. feat(pipeline): TwoFluidPipe liquid holdup calibration multiplier #4722's TwoFluidPipe holdup calibration is an adjacent pipeline calibration and is excluded from the lumped inventory tranche.
Next frozen tranche remains dependency-ready: reuse merged fix(fluidmechanics): qualify opt-in coupled gas/oil film solve #4711's default-off gas/oil film/EOS solve inside a distinct opt-in lumped non-equilibrium inventory state, with conservative per-phase component/enthalpy inventories, geometry-derived area, explicit stationary-film and wall/ambient heat-supply provenance, paired donor-limited mass/energy updates, bounded atomic adaptive steps, phase-appearance/residual diagnostics, deterministic schema-valid frames and both process containers. Preserve equilibrium compatibility. Predictive entrainment/population balance, distributed pipe behavior, Safety Shutdown and independent qualification remain separate boundaries.
Local Maven bootstrap succeeded, but this run's allocated workspace contains no NeqSim repository tree, Maven wrapper or validation scripts. Other scratch worktrees belong to separate active campaigns and were not reused as authoritative source. Compilation, conservation/refinement tests, direct Spotless, documentation search and hook gates therefore could not run; no unvalidated branch was published.

Issue #3860 remains open and UNQUALIFIED.