[Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting

[EvenSol]

Goal

Make the NeqSim MCP server a state-of-the-art, governed oil-and-gas process-engineering tool that can:

• translate a short engineering request, equipment list, facility narrative, structured data, or supported PFD/P&ID/DEXPI description into a reviewable NeqSim simulation;
• grow the same canonical model from one calculation or unit operation into a large, multi-area facility with recycles, controls, constraints, and reusable model state;
• execute the calculations routinely needed by process, flow-assurance, production-technology, operating-support, and troubleshooting engineers;
• investigate operational symptoms through explicit hypotheses, simulations, sensitivity cases, and evidence rather than unsupported diagnosis;
• return concise operator-facing conclusions while retaining detailed balances, assumptions, provenance, convergence evidence, validation findings, and selectively retrievable large-model results.

The server is an advisory engineering calculation layer. It must not autonomously control a plant, write to operational systems, claim accountable engineering approval, or hide missing data and model limitations.

Current baseline to reuse

Do not rebuild capabilities already present on current master.

• The neqsim-mcp-server module already provides governed profiles, standard response envelopes, validation, benchmark-trust metadata, traceability, STDIO/HTTP transports, schemas, resources, and guided prompts.
• runFlash, runProcess, manageModel, runPipeline, runOperationalStudy, lifecycle/session tools, and domain runners provide an extensive curated surface.
• Merged PR Fix MCP security lockout and propagate caller identity #2874 established caller identity and recoverable security enforcement.
• Merged PR Add MCP model handles, response limits and execution bounds #2875 established model handles, solved-model reuse, bounded execution, response-size control, large-process evidence, catalog coverage checks, and stronger benchmark contracts.
• Merged PR Add reusable MCP capability and process execution #3152 added runtime capability inspection/execution, canonical replayable process definitions, multi-area and equipment-design preservation, compressor anti-surge topology, capacity/bottleneck evidence, and typed detailed two-fluid pipeline results.
• Coordinate process/P&ID/DEXPI ingestion with [Campaign] Professional DEXPI and P&ID workflow #2899, dynamic studies with [Campaign] Professional reservoir-to-market dynamic process simulation #2911, flash/stability work with [Campaign] TP-flash stability, multiphase equilibrium and performance roadmap #2937, and the merged production-optimization foundation recorded in [Campaign] Reservoir-to-market production optimization roadmap #2941. Do not duplicate or compete with those roadmaps.

Engineering and product principles

1. One canonical model. Use normal NeqSim fluids, streams, ProcessSystem, ProcessModel, topology, design, and result objects. Do not create a second incompatible MCP-only simulator.
2. Progressive construction. A one-off PVT or equipment calculation must be reusable as an input to a stream, process area, facility, study, and report.
3. Explicit units and bases. Distinguish standard/actual, mass/molar/volume flow, absolute/gauge pressure, wet/dry basis, and operating/design conditions.
4. No silent invention. Facility parsing must return mapped facts, assumptions, ambiguities, missing data, unsupported items, and questions. Inferred values require provenance and confidence.
5. Evidence before diagnosis. Troubleshooting must separate observations, hypotheses, model tests, findings, uncertainty, and recommended next measurements/actions.
6. Fail visibly. Non-convergence, extrapolation, inconsistent measurements, unit errors, ambiguous topology, invalid thermodynamic choices, or unavailable validation evidence must produce structured diagnostics.
7. Bounded operation. Preserve time, memory, concurrency, payload, and security limits; make cancellation and partial-result behavior explicit.
8. Layered results. Return an operator-sized summary first and permit deterministic retrieval of area, equipment, stream, profile, balance, diagnostic, and provenance detail.
9. Validation maturity is explicit. Runtime availability is not proof of engineering validation. Every result must carry model choice, applicability, benchmark status, and limitations.
10. Backward compatibility. Evolve stable contracts deliberately, with schemas, migration guidance, and contract tests.

Roadmap

Phase 0 — Current-master audit and acceptance matrix

• Inventory every MCP tool, resource, prompt, schema, runner, supported equipment type, report path, deployment profile, test, guide, and known limitation on current master.
• Reconcile open/merged/closed MCP work and prove what Fix MCP security lockout and propagate caller identity #2874, Add MCP model handles, response limits and execution bounds #2875, and Add reusable MCP capability and process execution #3152 deliver on current master.
• Establish public synthetic acceptance fixtures at four scales: a single calculation/unit, a small train, a multi-area facility, and a large facility with recycles.
• Create a traceability matrix from each campaign criterion to source, tests, documentation, benchmark evidence, or a confirmed gap.
• Record baseline runtime, memory, response size, tool-call count, convergence, balance closure, and report usefulness.
• Define a machine-readable capability/maturity matrix by engineering discipline.

Phase 1 — Facility description to reviewable simulation

• Define a versioned facility-description contract accepting narrative facts, equipment/stream tables, structured JSON, and supported diagram/DEXPI references without mixing extracted facts with inference.
• Produce a deterministic normalized specification containing areas, fluids, streams, equipment, connections, operating/design cases, controls, constraints, data provenance, units, confidence, gaps, and questions.
• Add component/equipment/tag synonym resolution with explicit ambiguity diagnostics.
• Validate topology, port/phase compatibility, unit consistency, component coverage, boundary conditions, recycle closure, required equipment data, and thermodynamic-model suitability before solving.
• Generate a canonical ProcessSystem or multi-area ProcessModel plus editable/replayable process definition.
• Support incremental refinement: add an area, stream, equipment item, case, control, or constraint without rebuilding unrelated solved state.
• Coordinate diagram ingestion and round-trip semantics with [Campaign] Professional DEXPI and P&ID workflow #2899 rather than implementing a parallel P&ID model.
• Require an explicit review checkpoint before inferred facility content is treated as an accepted simulation input.

Phase 2 — Small-to-large composable process construction

• Provide reusable templates/builders for common oil-and-gas building blocks: well/feed, inlet separation, compression, dehydration, gas treatment, NGL recovery, stabilization, produced-water handling, export, injection, utilities, relief/blowdown/flare, and tie-in facilities.
• Preserve live stream identity, multi-port topology, area boundaries, recycles, adjusters, controllers, design constraints, equipment charts, geometry, and model revisions.
• Add predictable model composition, cloning, revision, comparison, serialization, Python replay, and selective re-solve behavior.
• Make unsupported equipment/connections explicit and extensible through reviewed adapters.
• Prove deterministic build and execution for small, medium, and large multi-area cases.

Phase 3 — Process-engineering calculation coverage

• Fluids/PVT: fluid characterization, plus fractions, EOS/mixing-rule selection, TP/PH/PS/TV flashes, phase envelopes, dew/bubble points, density/viscosity/thermal properties, standard volumes, calorific value/Wobbe, PVT experiments and regression where validated.
• Flow assurance: hydrate, water/MEG/TEG, wax, asphaltene, corrosion/scale screening, single- and multiphase pipe/well calculations, profiles, cooldown/transients, erosion and operating envelopes with applicability boundaries.
• Core equipment: streams, mixers/splitters, separators/scrubbers, valves/chokes, compressors/expanders, pumps, heaters/coolers/heat exchangers, columns/absorbers/strippers, tanks, pipelines/wells, reactors, utilities, and relevant mechanical/design calculations.
• Facility studies: mass/energy/component balances, equipment performance, capacity/utilization, bottlenecks, operating envelopes, turndown, recycle/control behavior, production allocation/optimization handoff, emissions/energy KPIs, and case comparison.
• Route eligible generic static calculations through runtime capability discovery only when the exact signature, bounds, maturity, and response contract are known; keep stateful calculations behind curated runners.

Phase 4 — Troubleshooting and operating-support workflow

• Define a structured troubleshooting case: symptom, time/operating context, observed values, expected behavior, data quality, relevant equipment/topology, hypotheses, tests, findings, confidence, and actions.
• Convert operator/support-engineer descriptions into testable hypotheses without claiming causality from correlation.
• Add reusable diagnostic patterns for separation loss, compressor underperformance/surge/recycle, valve restriction, exchanger fouling, column off-spec, hydrate/condensation risk, pipeline pressure loss, recycle non-convergence, instrument inconsistency, and changing feed composition.
• Compare current/reference/what-if states with conservation, thermodynamic consistency, equipment limits, nearby operating points, and sensitivity/uncertainty evidence.
• Distinguish process-model mismatch, bad/missing data, equipment degradation, control behavior, and numerical failure.
• Recommend the next safest measurement, check, or simulation; never issue plant-control commands.
• Build a public synthetic fault library with known injected causes and measurable pass/fail diagnostic criteria.

Phase 5 — Dynamic, operability, and scenario studies

• Support bounded steady-state, quasi-steady, and dynamic study orchestration with explicit solver/model selection.
• Compose startup/shutdown, trip, control-response, depressurization, cooldown, compressor recycle, slugging, and constraint-change cases where the underlying NeqSim model is validated.
• Coordinate underlying dynamic-model implementation with [Campaign] Professional reservoir-to-market dynamic process simulation #2911 and pipeline/flash campaigns rather than duplicating solver work.
• Preserve initial state, case definitions, time series, convergence, event chronology, safeguards, limitations, and reproducible replay.
• Keep all outputs advisory and clearly separate screening from qualified dynamic studies.

Phase 6 — Large-process execution and lifecycle

• Benchmark representative small (3–5 unit), medium (25–50 unit), large (150+ unit), and stress-scale facility models where feasible.
• Improve registration, revision, dependency-aware partial solving, warm-state reuse, caching, immutable snapshots, concurrency isolation, cancellation, timeout, recovery, and deterministic replay.
• Prevent repeated full solves for read-only result retrieval.
• Stream or page large results by model → area → equipment/stream → property/profile instead of producing context-breaking payloads.
• Preserve the standard compact response envelope and deterministic selective retrieval after truncation.
• Record runtime/memory/payload baselines and conservative regression thresholds by environment; do not hide degraded accuracy or convergence to meet a performance target.
• Add multi-client and tenant-isolation tests for mutable models and long-running studies.

Phase 7 — Engineering results and reporting

• Define typed, versioned result contracts for fluids, equipment, streams, areas, complete facilities, scenarios, troubleshooting findings, and comparisons.
• Provide summary/detail levels that retain exact units, basis, case, timestamp/revision, model/provenance, convergence, validation, warnings, and engineering-review requirements.
• Generate concise operator and support-engineer summaries: what changed, likely mechanisms, evidence, uncertainty, limits, and recommended next action.
• Generate process-engineering results: stream tables, equipment duties/performance, mass/component/energy balances, constraints/utilization, profiles, case deltas, diagnostics, and assumptions.
• Support deterministic Markdown/JSON/CSV and suitable visual/report artifacts without making downloadable files the only place results can be reviewed.
• Validate report completeness, numerical consistency, units, ordering, clipping/size, and selective-detail links for large facilities.

Phase 8 — MCP usability, schemas, and agent cooperation

• Make discovery task-oriented: “build this facility,” “run this calculation,” “explain this symptom,” and “compare these cases,” not only tool-name discovery.
• Keep input/output schemas complete, bounded, example-backed, and synchronized with the actual tool surface.
• Provide fix-oriented validation errors with field paths, accepted units/enums, alternatives, and recoverable next steps.
• Support resumable multi-step studies and explicit handoffs between facility parsing, model construction, simulation, troubleshooting, validation, and reporting.
• Add end-to-end protocol tests over STDIO and Streamable HTTP using real MCP clients or a standards-conforming harness.
• Keep NeqSim agents/skills and MCP contracts synchronized; route domain gaps to the underlying NeqSim capability instead of papering them over in prompts.

Phase 9 — Trust, security, deployment, and observability

• Preserve profile enforcement, identity, tenant isolation, audit, approval gates, response/execution bounds, OIDC/HTTP security, and advisory-only digital-twin behavior.
• Add calculation/model/revision correlation IDs, structured timings, cache/solve provenance, and privacy-safe diagnostics.
• Test malformed, adversarial, oversized, recursive, unsupported, and resource-exhausting inputs.
• Maintain deployment recipes and health/readiness evidence for desktop, CI, container, and governed team/enterprise use.
• Prove that secrets, credentials, proprietary input data, and raw plant data are not leaked through logs, reports, caches, or cross-tenant handles.

Phase 10 — Executed reference workflows and completion gate

• Execute and retain outputs for at least: one PVT/flash study, one equipment calculation, one small facility, one large multi-area facility, and several known-cause troubleshooting cases.
• Demonstrate narrative facility description → normalized specification/gap review → canonical model → execution → validation → operator summary → detailed engineering report → model revision/replay.
• Demonstrate progressive reuse from a single fluid/equipment calculation into a complete facility.
• Validate key engineering results against analytical checks, published/public benchmarks, independent models, experimental data, or defensible physical bounds.
• Complete focused and proportionate broader tests, formatting/static-analysis/documentation gates, exact protocol tests, security tests, and rendered artifact inspection.
• Publish a final capability/maturity/benchmark/limitation matrix with exact evidence links.
• Perform an independent current-master audit: no unmet definition-of-done criterion, no active campaign PR, no unresolved critical/high defect, no stale documentation/schema/example claim.

Definition of done

This campaign is complete only when current default-branch evidence demonstrates all of the following:

• A user can start with a typical oil-and-gas engineering question or facility narrative and obtain a reviewable canonical model with explicit mappings, assumptions, gaps, and questions.
• The same model can be extended from a single calculation/unit into a multi-area large facility without incompatible representations.
• Representative PVT, flash, valve, compressor, separator, exchanger, column, pump, pipeline/well, flow-assurance, balance, performance, capacity, and scenario studies execute through documented contracts where NeqSim has validated functionality.
• Known-cause synthetic troubleshooting cases produce evidence-backed, uncertainty-aware findings and useful next actions without unsafe control advice.
• Small, medium, and 150+ unit acceptance models are reproducible, bounded, isolated across clients, and usable without repeated unnecessary full solves or oversized responses.
• Operator summaries and detailed engineering results are consistent, unit-complete, selectively retrievable, reproducible, and explicit about convergence, validation maturity, assumptions, and limitations.
• Every public tool/resource/prompt/schema is covered by contract tests and current documentation; security, governance, execution, compatibility, and documentation gates pass.
• The MCP server remains an advisory calculation and decision-support tool; it never claims plant authority or accountable engineering approval.

Campaign coordination and run ledger

• Maintain at most one active implementation PR for this MCP campaign and no more than ten open autonomous NeqSim implementation PRs overall.
• Before every run, read this issue and newest comments, current master, repository instructions, current MCP source/docs/tests, related open/recent PRs, reviews/checks, and overlap roadmaps.
• Search for merged or active functionality before selecting work. Reuse it and never create a competing implementation.
• Select the largest coherent dependency-ready increment that remains one reviewable concept. Include implementation, tests, validation, documentation, schemas/examples, and agent/skill contract updates when coupled.
• Never write to master, merge, enable auto-merge, force-push, or mark a draft ready for review.
• Every productive run must append a ledger comment containing: timestamp; exact base/head; active autonomous PR count; selected roadmap item and stop boundary; files/PR; tests and engineering validation; documentation impact; limitations/blocker; and next dependency.
• Open or draft PR work remains in progress until merged. Mark roadmap boxes complete only from verified merged default-branch evidence.
• Notify only for a new/materially updated PR, a changed blocker/decision, first READY TO MERGE classification for an exact head, or final campaign completion. Keep no-op runs silent.

[EvenSol]

Autonomous campaign run ledger — 2026-08-22T15:54:02.190Z

• Base / published head: 885e3986c195ba5ed4da95b772b3ca0407e22716 → 885e3986c195ba5ed4da95b772b3ca0407e22716 (no branch or commit published)
• Open autonomous implementation PRs: 1 overall ([AI-Assisted] feat(diagrams): add exchange-neutral graphical projection #3155, unrelated); 0 active for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153
• Selected roadmap item: Phase 0 runtime MCP surface-inventory and cross-layer traceability foundation
• Stop boundary: exact published tools/resources/resource templates/prompts and deployment profiles only; schema/example/runner/equipment/report-path inventories, acceptance fixtures, and performance/conservation baselines remain follow-up
• Working-tree implementation: new McpSurfaceInventory; getCapabilities.serverSurface; response-size protection; focused core/server regressions; MCP STDIO protocol assertions; README/API reference/new surface-inventory guide
• Validation: 13 focused core tests passed (CapabilitiesRunnerTest, McpToolSurfaceContractTest, ResponseSizeGuardTest); 3 server inventory tests passed under the available JDK 17 fallback; 10 focused STDIO MCP capability checks passed under that fallback; Spotless apply was stable, Spotless check passed, documentation-search audit passed (655 Markdown + 1 HTML), Python protocol test parsed, and git diff --check passed. The pre-commit executable is unavailable; mandatory direct gates were run.
• Engineering validation: annotation-derived counts reconcile 71 tools, 7 resources, 7 templates, and 9 prompts with governance/capability registries; the compact inventory survives the 256 KiB response guard and preserves the canonical NeqSim ProcessSystem/ProcessModel. No thermodynamic or numerical model changed.
• Documentation impact: updated MCP README and API reference; added the Phase 0 inventory/reconciliation/compatibility guide.
• Blocker / limitation: the mandatory clean module gate cannot run successfully in this environment because neqsim-mcp-server requires JDK 21 and only JDK 17 is installed (release version 21 not supported). Per publication gates, no commit, branch, or draft PR was published. No roadmap checkbox was marked complete.
• Next dependency: obtain a JDK 21 runner, rerun the clean module test/package and focused STDIO protocol gate on the exact final tree, then publish one connector-authored draft PR if all direct gates pass.

[EvenSol]

Autonomous campaign run ledger — 2026-08-22T18:12:59.298Z

• Base / head: 0cc95dbe3d065094580f3d2a742ad5b373b5f52f → 0fdc94c9b5e735d7b34980bcdeec639d9ae4d995
• Open autonomous implementation PRs: 3 overall ([AI-Assisted] test(mcp): freeze published protocol surface #3162, docs: repair learning path accuracy and navigation #3161, [AI-Assisted] fix(process): serialize shared-stream consumers #3160); 1 active for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153
• Selected roadmap item: Phase 0 exact transport-facing MCP surface inventory and capability/governance reconciliation
• Stop boundary: published tools, static resources, resource templates, guided prompts, deployment-profile names, and existing tool-catalog coverage only; the broader schema/example/runner/equipment/report/test/guide/limitation inventory and acceptance/performance matrices remain follow-up
• Files / PR: neqsim-mcp-server/test_mcp_server.py, README.md, docs/API_REFERENCE.md, new docs/SURFACE_INVENTORY.md; draft PR [AI-Assisted] test(mcp): freeze published protocol surface #3162
• Tests and engineering validation: 83/83 focused real MCP STDIO protocol checks passed, including exact 71-tool, 7-resource, 7-template, and 9-prompt identities plus complete 71/71 capability-descriptor coverage; Python compile, Spotless check, documentation search (655 Markdown + 1 HTML), pre-commit, pre-push, and diff checks passed. No thermodynamic/numerical behavior changed.
• Documentation impact: added the published-surface inventory and linked it from the MCP README and API reference; compatibility and advisory-only boundaries are explicit
• Limitations / blocker: hosted exact-head JDK 21 CI and review are in progress, so the draft is not classified READY TO MERGE. The available local JDK 17 fallback was used only to package unchanged server source for protocol execution; the temporary POM override was restored and is absent from the PR.
• Next dependency: after exact-head CI/review, inventory individual schemas, examples, runners, equipment factories, report paths, tests, guides, and known limitations, then reconcile Fix MCP security lockout and propagate caller identity #2874/Add MCP model handles, response limits and execution bounds #2875/Add reusable MCP capability and process execution #3152 criterion by criterion.

[EvenSol]

Campaign run ledger — 2026-08-22T20:31:54+02:00

• Base / head: c63eaa00320ba309dc5a27ab888fe37091c32b54 → 02bd188fce0949802c04363ba1856486227c05e9
• Open autonomous implementation PRs: 3/10 overall; 1 active for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153
• Roadmap item / stop boundary: Phase 0 executable schema/example catalog traceability. Stops before runner, equipment-factory, report-path, test/guide/limitation matrices and measured acceptance baselines; no Phase 0 checkbox marked complete.
• PR / files: draft [AI-Assisted] test(mcp): verify schema and example resources #3164; neqsim-mcp-server/test_mcp_server.py, neqsim-mcp-server/docs/SURFACE_INVENTORY.md
• Tests / engineering validation: focused catalog protocol 35/35; publication + catalog + capability protocol 118/118; all 142 schema resources and 114 example resources resolved through JSON-RPC resources/read; Python compile, Spotless, documentation search, pre-commit, pre-push, and diff check passed. No scientific calculation changed, so new numerical/benchmark validation was not applicable.
• Documentation impact: records exact schema/example counts, resource evidence, compatibility, and the limit that discoverability is not scientific validation or facility authority.
• Limitations / blocker: exact-head build/test/Javadoc, pre-commit, and Spotless workflows passed; CodeQL remains in progress. No reviews, Copilot findings, comments, or unresolved threads at inspection. Not READY TO MERGE while CodeQL is pending.
• Next dependency: after [AI-Assisted] test(mcp): verify schema and example resources #3164 is merged, continue the first Phase 0 inventory with runners, equipment factories, report paths, tests, guides, and known limitations, then build the acceptance/traceability matrices without overlapping [Campaign] Professional DEXPI and P&ID workflow #2899, [Campaign] Professional reservoir-to-market dynamic process simulation #2911, [Campaign] TP-flash stability, multiphase equilibrium and performance roadmap #2937, or [Campaign] Reservoir-to-market production optimization roadmap #2941.

[EvenSol]

Scheduled campaign ledger — 2026-08-22 20:50 CEST

• Exact master/base/head inspected: f8030e066d0e875bb241404c0ca4c6ff3dae1795 ([AI-Assisted] test(mcp): verify schema and example resources (#3164)). No campaign branch or commit was created in this run.
• Autonomous/open implementation capacity: 3 open PRs observed repository-wide; no active [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153 implementation PR; below the 10-PR campaign cap.
• Roadmap item / stop boundary: Phase 0 only — next incomplete dependency is runner/equipment/report-path inventory and explicit ownership/overlap traceability. Current source confirms reuse of canonical neqsim.process.equipment.EquipmentFactory and neqsim.process.processmodel.JsonProcessBuilder; specialized DEXPI construction remains with [Campaign] Professional DEXPI and P&ID workflow #2899. No Phase 1 ingest/build, dynamics ([Campaign] Professional reservoir-to-market dynamic process simulation #2911), flash/stability/performance ([Campaign] TP-flash stability, multiphase equilibrium and performance roadmap #2937), or production-optimization ([Campaign] Reservoir-to-market production optimization roadmap #2941) implementation was started.
• Merged evidence reconciled: [AI-Assisted] test(mcp): verify schema and example resources #3164 is merged on current master and extends the test-backed MCP schema/example resource inventory; it has no unresolved review threads. Its merged scope explicitly leaves runners, equipment factories, report paths, tests/guides/limitations and broader traceability/baselines for follow-up Phase 0 work.
• Files/PR: inspection only; no repository files changed and no PR opened.
• Validation/gates: GitHub source/PR/issue inspection completed. Local checkout/bootstrap was attempted solely as a supporting validation environment, but DNS resolution for github.com failed (Could not resolve host: github.com). Consequently Maven/pre-commit/pre-push and focused MCP tests could not be executed locally. Per campaign publication gates, no commit was published and no CI pass is claimed.
• Documentation impact: none published. Proposed next coherent increment would extend neqsim-mcp-server/docs/SURFACE_INVENTORY.md plus executable MCP/source-traceability regressions, without adding a second simulator or construction model.
• Limitation/blocker: mandatory direct validation environment unavailable in this run; this is an infrastructure blocker, not a scientific or architectural blocker.
• Next dependency: when direct gates are executable, implement the Phase 0 runner/equipment/report-path inventory as one reviewable increment, prove references against current source, run the required gates, then open/update one draft [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153 PR only if all publication gates permit.

[EvenSol]

Campaign run ledger — 2026-08-22 21:46:14 CEST

• Exact base / head: current master 4cf781816f905aebafec6686457f9be750a25561; draft PR head fca8ed971aafb29166eea66cb40239b35d422a3f. The branch was created from f8030e066d0e875bb241404c0ca4c6ff3dae1795; [AI-Assisted] fix(process): serialize shared-stream consumers #3160 merged to master during publication, changes only unrelated ProcessSystem scheduler files, and GitHub reports [AI-Assisted] feat(mcp): expose implementation inventory #3166 mergeable with the intended eight-file diff.
• Autonomous implementation PRs: 3/10 overall ([AI-Assisted] fix(optimization): reject non-finite plant candidates #3163, [AI-Assisted] fix(thermo): reject stale reaction topology #3165, [AI-Assisted] feat(mcp): expose implementation inventory #3166); 1 active for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153.
• Selected item / stop boundary: Phase 0 executable implementation, canonical EquipmentFactory, and bounded report-path traceability. Metadata/contracts only; no simulator, thermodynamics, process execution, facility ingestion, DEXPI ([Campaign] Professional DEXPI and P&ID workflow #2899), dynamics ([Campaign] Professional reservoir-to-market dynamic process simulation #2911), flash/performance ([Campaign] TP-flash stability, multiphase equilibrium and performance roadmap #2937), or optimization ([Campaign] Reservoir-to-market production optimization roadmap #2941) expansion.
• Files / PR: draft [AI-Assisted] feat(mcp): expose implementation inventory #3166 adds McpImplementationInventory, additive getCapabilities/schema metadata, Java and full stdio protocol regressions, and synchronized API_REFERENCE.md / SURFACE_INVENTORY.md.
• Tests and engineering validation: focused Maven 33/33; complete packaged MCP JSON-RPC suite 309/309 with task-local writable user.home; exact 71 tools → 60 loadable classes, 205 factory types equal the process contract, and 2 report paths; representative flash/process/valve/compressor/separator/PVT/pipeline/report flows remained green; ResponseSizeGuard passed. Spotless apply/check, pre-commit, pre-push, documentation search (655 Markdown + 1 HTML), Python compile, diff check, and Maven Javadoc passed.
• Documentation impact: updated API reference and Phase 0 surface inventory with compatibility, discovery, canonical-model, report-persistence, advisory-only, and validation-maturity boundaries.
• Limitations / blocker: no scientific code changed, so no new benchmark claim applies. Local runtime has Java 17 only; MCP packaging used a restored, local-only release-17 fallback, while hosted Java 21 remains authoritative. Exact-head CI is running (build/test/Javadoc, Spotless, pre-commit, CodeQL, PaperLab); head is not READY TO MERGE while checks are pending.
• Next dependency: after [AI-Assisted] feat(mcp): expose implementation inventory #3166 is merged on master, verify and inventory remaining Phase 0 tests, guides, and known limitations, then define the four acceptance scales and full traceability/maturity matrices. No Phase 0 roadmap criterion is marked complete in this run.

[EvenSol]

Campaign run ledger — 2026-08-23 02:38:58 CEST

• Exact base / head: e537b6c24eb8a5137e79bdd8956820804a37bcc6 → 911f63124e7869f523ff78f9b1aaf8d05fdfc499; branch was re-read at that exact head after publication and GitHub reports the intended eight-file diff mergeable.
• Autonomous implementation PRs: 2/10 overall ([AI-Assisted] feat(dexpi): adapt graphical projection to Core #3167 and [AI-Assisted] feat(mcp): expose Phase 0 evidence inventory #3173); 1 active for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153.
• Selected item / stop boundary: remaining Phase 0 test, guide, and known-limit/trust evidence inventory. Additive discovery metadata and regressions only; no new tool, simulator, numerical behavior, facility content, benchmark claim, or overlap with [Campaign] Professional DEXPI and P&ID workflow #2899/[Campaign] Professional reservoir-to-market dynamic process simulation #2911/[Campaign] TP-flash stability, multiphase equilibrium and performance roadmap #2937/[Campaign] Reservoir-to-market production optimization roadmap #2941.
• Files / PR: draft [AI-Assisted] feat(mcp): expose Phase 0 evidence inventory #3173; new McpEvidenceInventory, additive getCapabilities/schema metadata, Java + full real-STDIO protocol regressions, and synchronized MCP README/API/surface-inventory documentation.
• Tests / engineering validation: focused Java 17 34/34 and Java 8 target 34/34; broad MCP JUnit 480, 0 failures/errors, 1 expected skip; packaged MCP JSON-RPC 317/317, including flash, PVT, process, valve, compressor, separator, pipeline, reporting, catalog, and evidence reconciliation. Spotless, documentation search (655 Markdown + 1 HTML), pre-commit, pre-push, Javadocs, Python compile, and diff checks passed. Inventory reconciles 67 MCP test classes, 94 protocol scenarios, 4 guides, 71 published tools, 20 explicit trust pages, 51 generic fallbacks, 64 known-limit entries, 30 validation cases, and 5 concrete verifiedBy links. No scientific code changed.
• Documentation impact: updated neqsim-mcp-server/README.md, docs/API_REFERENCE.md, and docs/SURFACE_INVENTORY.md; MCP_CONTRACT.md assessed unchanged because the tool/request/envelope contract is unchanged. No companion agent/skill handoff is required.
• Limitations / blocker: inventory deliberately remains complete: false; generic TESTED fallback is not validation, and per-result provenance/convergence/assumptions/limitations remain authoritative. Local JDK 17 packaging used a restored local-only release fallback; hosted JDK 21 is authoritative. Exact-head CI is in progress with no reviews, Copilot findings, comments, or unresolved threads yet, so the head is not READY TO MERGE. No roadmap criterion is marked complete before merge.
• Next dependency: after [AI-Assisted] feat(mcp): expose Phase 0 evidence inventory #3173 merges, reconcile Fix MCP security lockout and propagate caller identity #2874/Add MCP model handles, response limits and execution bounds #2875/Add reusable MCP capability and process execution #3152 criterion-by-criterion on current master, convert generic trust fallbacks into evidence-backed capability or confirmed-gap records, then define the four acceptance scales and traceability/maturity baselines.

[EvenSol]

Scheduled campaign ledger — 2026-08-23 06:33 CEST

• Exact base / head inspected: current master 5a851750f8d12b3a598a87da0acadb3faef8b4e6 → same; no branch or commit published in this run.
• Autonomous implementation PRs: 0 open user-authored implementation PRs observed in equinor/neqsim; 0 active for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153, below the 10-PR cap.
• Selected roadmap item / stop boundary: Phase 0 merged-foundation reconciliation for Fix MCP security lockout and propagate caller identity #2874/Add MCP model handles, response limits and execution bounds #2875/Add reusable MCP capability and process execution #3152 against the now-merged surface/schema/example/implementation/evidence inventory. Stop before converting the 51 generic trust fallbacks, defining acceptance scales, or adding performance/scientific behavior.
• Merged evidence reconciled: Fix MCP security lockout and propagate caller identity #2874 establishes caller identity/security recovery and tenant/subject-aware authorization; Add MCP model handles, response limits and execution bounds #2875 establishes model handles, response-size/execution/concurrency bounds, capability catalog and bounded execution infrastructure; Add reusable MCP capability and process execution #3152 establishes reusable capability discovery/execution plus canonical ProcessSystem/ProcessModel construction/replay and process/pipeline/report workflows. [AI-Assisted] feat(mcp): expose Phase 0 evidence inventory #3173 is merged and current master now exposes 67 MCP test classes, 94 protocol scenarios, four guides, 20 explicit trust pages, 51 generic fallbacks, 64 known-limit entries, 30 validation cases and 5 concrete verifiedBy links. The current Phase 0 audit therefore remains explicitly incomplete.
• Overlap boundary: DEXPI/P&ID construction remains [Campaign] Professional DEXPI and P&ID workflow #2899-owned, dynamics [Campaign] Professional reservoir-to-market dynamic process simulation #2911-owned, flash/stability/performance [Campaign] TP-flash stability, multiphase equilibrium and performance roadmap #2937-owned, and production optimization [Campaign] Reservoir-to-market production optimization roadmap #2941-owned; no competing implementation was started.
• Files / PR: inspection/reconciliation only; no repository files changed and no PR opened.
• Tests / engineering validation: current GitHub source, merged PR evidence, repository AGENTS.md, and MCP inventory were inspected. A fresh local checkout/bootstrap was attempted for mandatory direct gates but failed at DNS resolution (Could not resolve host: github.com), so Maven, Spotless, documentation search, pre-commit/pre-push and focused protocol tests were not run in this run. No pass is claimed for unrun gates.
• Documentation impact: none published. The next coherent change should add a criterion-level foundation-to-current-capability/gap traceability matrix, preferably in docs/SURFACE_INVENTORY.md with executable reconciliation where practical, without creating an MCP-only simulator or changing public contracts unnecessarily.
• Limitation / blocker: direct validation remains unavailable because the local environment cannot resolve GitHub. Per publication gates, no commit or draft PR was published. This is an infrastructure blocker, not evidence of a scientific or architectural failure.
• Next dependency: once direct gates are executable, implement and validate the criterion-level Fix MCP security lockout and propagate caller identity #2874/Add MCP model handles, response limits and execution bounds #2875/Add reusable MCP capability and process execution #3152 traceability matrix, then use it to convert generic trust fallbacks into evidence-backed capability or confirmed-gap records before moving to the four acceptance scales and measured runtime/memory/payload/convergence/balance/report baselines.

[EvenSol]

Campaign run ledger — 2026-08-23 14:58 CEST

• Exact base / head: branch created from 07124d06ba3896e2da6dd33dac19cd6c8e44e2e2; draft PR [AI-Assisted] feat(mcp): expose Phase 0 merged-foundation traceability #3185 exact head 3bb751ed670535dd1bba36ef4037fcda157c4dbe. Current master advanced independently to 3e7dc6c64ad16992de652f95707b0194c374870c through documentation PR docs: make reference index source-safe #3183 after branch creation; per campaign policy the MCP branch was not updated/rebased/merged from master solely for base drift.
• Autonomous implementation PRs: 1/10 open user-authored implementation PR observed after publication ([AI-Assisted] feat(mcp): expose Phase 0 merged-foundation traceability #3185); 1 active for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153.
• Selected roadmap item / stop boundary: Phase 0 merged-foundation reconciliation for Fix MCP security lockout and propagate caller identity #2874/Add MCP model handles, response limits and execution bounds #2875/Add reusable MCP capability and process execution #3152. Evidence/discovery metadata only; stops before generic trust-fallback conversion, four-scale acceptance fixtures, full traceability/maturity matrices, measured runtime/memory/payload/convergence/balance/report baselines, and all Phase 1 work.
• Files / PR: draft [AI-Assisted] feat(mcp): expose Phase 0 merged-foundation traceability #3185; src/main/java/neqsim/mcp/runners/McpEvidenceInventory.java, src/test/java/neqsim/mcp/runners/McpEvidenceInventoryFoundationTests.java, neqsim-mcp-server/docs/FOUNDATION_TRACEABILITY.md.
• Implementation/evidence: existing getCapabilities.phase0EvidenceInventory now exposes a mergedFoundations object with exact merged PR/commit identity, durable capability contract, representative current source/test paths, explicit boundary, and remaining Phase 0 dependency. No new MCP tool, simulator, thermodynamic/process algorithm, or public request/envelope contract was added.
• Validation: VALIDATION PENDING CI. GitHub source and merged PR evidence were inspected. A wrong initial TwoFluidPipeResponse source path was detected against current source and corrected before publication. The focused regression is named *Tests.java so it is picked up by Surefire without changing the existing frozen 67 *Test.java source inventory used by the protocol evidence contract. Local Maven/JDK/Spotless/pre-commit/Javadoc execution was not claimed. Exact-head hosted workflows started and are currently in progress: build/test/Javadoc, pre-commit, Spotless, CodeQL, and PaperLab.
• Reviews: no reviews or unresolved review threads at latest inspection.
• Documentation impact: added neqsim-mcp-server/docs/FOUNDATION_TRACEABILITY.md; MCP_CONTRACT.md remains unchanged because the tool/request/envelope contract is unchanged.
• Limitations / blocker: PR head is not READY TO MERGE while exact-head CI is pending. Overall Phase 0 remains explicitly incomplete; merged-foundation reconciliation alone is not scientific validation or campaign completion.
• Next dependency: after [AI-Assisted] feat(mcp): expose Phase 0 merged-foundation traceability #3185 merges and current-master evidence is rechecked, convert the 51 generic benchmark-trust fallbacks into evidence-backed capability records or explicit confirmed gaps, then define the four acceptance scales and traceability/maturity baselines.

[EvenSol]

Campaign run ledger — 2026-08-23 16:06 CEST

• Exact base / head: current master and branch base e06ba2f410b187075c1ddb88e4475d51e6c1622c (merged [AI-Assisted] feat(mcp): expose Phase 0 merged-foundation traceability #3185) → draft PR [AI-Assisted] feat(mcp): classify Phase 0 trust coverage gaps #3188 exact head 55b2a6787e0c532ec425b3a0b59c0c94f9c17f54. The branch was created directly from that master head; no update-branch, master merge, rebase, or cherry-pick was used.
• Autonomous implementation PRs: 2/10 open user-authored implementation PRs observed after publication ([AI-Assisted] feat(mcp): classify Phase 0 trust coverage gaps #3188 for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153 and unrelated electrolyte [AI-Assisted] fix(chemistry): separate Pitzer reaction standard state #3187); 1 active for [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153.
• Selected roadmap item / stop boundary: Phase 0 trust-coverage reconciliation. Convert the previously implicit 51 generic getBenchmarkTrust fallbacks into explicit confirmed-gap records while preserving the 20 tool-specific trust pages. Stops before adding new benchmark claims, changing BenchmarkTrust compatibility maturity, four-scale acceptance fixtures, full traceability/maturity matrices, measured runtime/memory/payload/convergence/balance/report baselines, and all Phase 1 work.
• Files / PR: draft [AI-Assisted] feat(mcp): classify Phase 0 trust coverage gaps #3188; src/main/java/neqsim/mcp/runners/McpEvidenceInventory.java, src/test/java/neqsim/mcp/runners/McpEvidenceInventoryFoundationTests.java, neqsim-mcp-server/docs/SURFACE_INVENTORY.md, neqsim-mcp-server/docs/FOUNDATION_TRACEABILITY.md.
• Implementation / engineering evidence: phase0EvidenceInventory.knownLimitations.coverageRecords now contains one deterministic record for every 71 published MCP tools. Twenty are EXPLICIT_TRUST; 51 are CONFIRMED_GAP. Each record carries the current implementation class, compatibility maturity, validation/verified-case/limitation/unsupported counts, and an explicit gap reason. coverageComplete=true means classification coverage only; scientificValidationComplete=false and the existing complete=false remain explicit. No thermodynamic, process, pipeline, DEXPI, dynamic, flash, optimization, live-plant, tool-name, request-schema, or response-envelope behavior changed.
• Tests: existing McpEvidenceInventoryFoundationTests was extended to pin 71/71 coverage, the 20/51 split, explicit runFlash trust, confirmed-gap getCapabilities, implementation traceability, and the distinction between coverage completeness and scientific validation. The regression remains in the existing *Tests.java file, so the frozen 67 *Test.java and 94 protocol-scenario source counts are unchanged.
• Validation: VALIDATION PENDING CI. GitHub source, current master, merged [AI-Assisted] feat(mcp): expose Phase 0 merged-foundation traceability #3185 evidence, exact branch commits, final four-file diff, and mergeability were inspected through the connected GitHub mechanism. No local Maven/JDK/Spotless/pre-commit/Javadoc run is claimed. Exact-head hosted workflows have started: PaperLab, pre-commit, Spotless, and CodeQL are in progress; build/test/Javadoc is queued. No unrun or unfinished check is claimed as passed.
• Reviews / mergeability: GitHub reports [AI-Assisted] feat(mcp): classify Phase 0 trust coverage gaps #3188 mergeable; no reviews or unresolved review threads at latest inspection. The draft is not READY TO MERGE while exact-head CI is incomplete.
• Documentation impact: SURFACE_INVENTORY.md now documents all 71 trust-coverage records and the coverage-vs-validation distinction; FOUNDATION_TRACEABILITY.md advances its Phase 0 handoff. MCP_CONTRACT.md remains unchanged because the public tool/request/envelope contract is unchanged.
• Limitations / blocker: 51 tools still lack tool-specific benchmark/accuracy/limitation evidence; they are now explicit gaps rather than implied validation. Hosted exact-head CI is still running.
• Next dependency: after [AI-Assisted] feat(mcp): classify Phase 0 trust coverage gaps #3188 is merged and verified on current master, establish the four public synthetic acceptance scales and full campaign traceability/maturity matrices, while closing individual confirmed trust gaps only where current source and public evidence support specific claims; then freeze runtime, memory, payload, convergence, balance-closure, and report-usefulness baselines.

[EvenSol]

Scheduled campaign ledger — 2026-08-23 18:15 CEST

• Exact base / head: current master and branch base eed348be246f941e9b25c0a307163b7ffb2a0464 (merged [AI-Assisted] feat(mcp): classify Phase 0 trust coverage gaps #3188) → draft PR [AI-Assisted] feat(mcp): add Phase 0 four-scale acceptance fixtures #3192 head 99631f0156ee36bdf93db83d20069913690e5815. The branch is ahead-only from that exact base; no master update/rebase/merge/cherry-pick was performed.
• Autonomous implementation PRs: repository/user open-PR inventory was empty before publication; [AI-Assisted] feat(mcp): add Phase 0 four-scale acceptance fixtures #3192 is now the sole active [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153 implementation PR and remains below the 10-PR global cap.
• Selected roadmap item / stop boundary: Phase 0 public synthetic acceptance fixtures at four scales. Scope stops at stable MCP inputs, production-runner execution contracts, and discoverable fixture metadata. It does not implement plant-wide optimization/constraint fidelity ([Campaign] Industrial-scale ProcessSystem and ProcessModel optimization #3154), generic process execution performance/scheduling ([Campaign] ProcessSystem and ProcessModel performance roadmap #2939), dynamics ([Campaign] Professional reservoir-to-market dynamic process simulation #2911), flash/stability algorithms ([Campaign] TP-flash stability, multiphase equilibrium and performance roadmap #2937), or DEXPI/P&ID ingestion ([Campaign] Professional DEXPI and P&ID workflow #2899).
• Implementation / files: draft [AI-Assisted] feat(mcp): add Phase 0 four-scale acceptance fixtures #3192 adds McpAcceptanceFixtureCatalog, executable McpAcceptanceFixtureCatalogTests, docs/ACCEPTANCE_FIXTURES.md, and additive integration into McpEvidenceInventory plus its existing foundation/trust regression. Five files changed. The catalog exposes four deterministic public-synthetic inputs through getCapabilities.phase0EvidenceInventory.acceptanceFixtures: (1) runFlash single calculation; (2) 10-unit/1-recycle ProcessSystem; (3) three-area/8-unit ProcessModel; and (4) generated 154-unit/1-recycle ProcessSystem for MCP transport/execution scale. Existing ExampleCatalog inputs and canonical JSON-builder equipment are reused; no proprietary converted-plant data is embedded.
• Tests / engineering validation: VALIDATION PENDING CI. Local Maven/JDK/Spotless/pre-commit execution was not used as a publication prerequisite. The new regressions require all four fixtures to execute through production FlashRunner/ProcessRunner; the large fixture must preflight valid, contain exactly 154 units, execute successfully with recycle, return a 154-unit canonical processDefinition, and report converged provenance. Hosted exact-head workflows currently in progress: build/test/Javadoc, Spotless format patch, pre-commit, CodeQL, and PaperLab. No unrun or pending check is claimed as passed. PR is currently mergeable; no reviews or unresolved review threads were present at inspection.
• Documentation impact: new neqsim-mcp-server/docs/ACCEPTANCE_FIXTURES.md records fixture rationale, canonical routes, acceptance questions, public/synthetic provenance, owner-roadmap boundaries, and the next measured-baseline tranche. Public MCP tool/request/envelope contracts are unchanged; fixture metadata is additive inside the existing Phase 0 evidence object, whose global complete flag remains false.
• Limitations / blocker: fixture definitions are now frozen in the draft, but their exact-head execution is still awaiting CI. The 154-unit fixture is intentionally an MCP scale/payload case, not a representative-plant, optimization, process-performance, or scientific-validation benchmark. Phase 0 trust gaps, criterion traceability/maturity matrices, and measured runtime/memory/payload/convergence/balance/report-usefulness baselines remain open.
• Next dependency: after [AI-Assisted] feat(mcp): add Phase 0 four-scale acceptance fixtures #3192 merges and its exact-head fixture execution is verified, run the frozen four-scale measurement harness and record runtime/environment, memory/allocation where available, response size/guarding, tool-call/selective retrieval, convergence/warnings, mass/component/energy closure as applicable, deterministic repeatability, and report usefulness; then complete the campaign criterion traceability and discipline-level capability/maturity matrices.

[EvenSol]

Campaign run ledger — 2026-08-23 20:03:10 CEST (exact-head update 2026-08-23 20:11:39 CEST)

• Exact base / head: 59ec0afe3efeb1ced0a6aeec9bcb7c852b1d83b6 → draft PR [AI-Assisted] feat(mcp): measure Phase 0 acceptance baselines #3196 exact head 57b446137b999278e68f75c3b858a7c451e7c4e5 (initial implementation head 7b872309bf80c13d5f523c7b256f5f332f77d7a0). The branch is two commits ahead and zero behind its creation base; no update-branch, master merge, rebase, cherry-pick, or force push was used.
• Autonomous implementation PRs: 1/10 positively identified autonomous implementation PR ([AI-Assisted] feat(mcp): measure Phase 0 acceptance baselines #3196); it is the sole active [Campaign] State-of-the-art NeqSim MCP process simulation and troubleshooting #3153 implementation PR.
• Selected roadmap item / stop boundary: Phase 0 exact-run measurement harness for the four public synthetic acceptance fixtures. Scope stops at environment-qualified runtime/heap snapshots, raw and guarded payload, tool-call count, convergence/warnings, deterministic repeat, structural report evidence, and explicit numerical balance-evidence paths/gaps. It does not change process performance ([Campaign] ProcessSystem and ProcessModel performance roadmap #2939), optimization fidelity ([Campaign] Industrial-scale ProcessSystem and ProcessModel optimization #3154/[Campaign] Reservoir-to-market production optimization roadmap #2941), dynamics ([Campaign] Professional reservoir-to-market dynamic process simulation #2911), flash algorithms ([Campaign] TP-flash stability, multiphase equilibrium and performance roadmap #2937), DEXPI/P&ID ingestion ([Campaign] Professional DEXPI and P&ID workflow #2899), public tool names, or live-plant behavior.
• Files / PR: draft [AI-Assisted] feat(mcp): measure Phase 0 acceptance baselines #3196 changes 11 MCP implementation/test/protocol/documentation files; adds McpAcceptanceBaselineRunner, McpAcceptanceBaselineRunnerTests, and docs/ACCEPTANCE_BASELINES.md; exposes only the bounded non-executing contract through phase0EvidenceInventory; reconciles seven current MCP guides. The exact-head follow-up removes an inactive Log4j retention claim and accurately assigns numeric persistence to a governed invoking workflow; runtime behavior and the public MCP contract are unchanged.
• Local tests and engineering validation: initial Spotless apply/check, documentation search (659 Markdown + 1 HTML), protocol Python compile, focused Java suite 16/16, Java 8 compile, and diff check passed. The harness executed all four fixture scales twice: 4/4 successful, 4/4 explicitly converged, 4/4 deterministic stable outcomes. After the documentation repair, Spotless apply/check, documentation search, Python syntax, focused baseline/inventory tests 5/5, and diff check passed. Payload guarding is asserted within the configured bound. Balance closure is never inferred: the report emits named response evidence or GAP_NO_NUMERIC_CLOSURE_IN_MCP_RESPONSE.
• Hosted exact-head CI: PaperLab Replication Gate: completed/success; Pre-commit checks: completed/success; Spotless format patch: completed/success; Run build, test and javadoc: in_progress; CodeQL Security Analysis: in_progress. On exact head 57b446137b999278e68f75c3b858a7c451e7c4e5, agent/skill cross-references, change detection, and Java 21 Javadoc have passed; Ubuntu/Windows Java 21 tests, slow-test shards, and CodeQL are still running. No pending check is claimed as passed. VALIDATION PENDING CI.
• Documentation impact: added the baseline guide and synchronized acceptance-fixture and surface-inventory documentation. README.md, API_REFERENCE.md, and MCP_CONTRACT.md remain unchanged because no public tool/input/envelope contract changed. No companion agent/skill update is required.
• Limitations / blocker: runtime and heap measurements are environment-specific observations, not portable thresholds or scientific/design qualification. Local pre-commit was unavailable and local Javadoc was blocked by invalid JAVA_HOME; new-head hosted equivalents remain authoritative. The final 11-file diff has no conflict markers or detected secret patterns; the PR is draft and mergeable with no reviews, Copilot comments, or unresolved threads, but is not READY TO MERGE while exact-head CI is incomplete. No roadmap checkbox is marked complete before merge.
• Next dependency: after [AI-Assisted] feat(mcp): measure Phase 0 acceptance baselines #3196 merges and exact-head evidence is verified on master, close only defensible numerical balance/report gaps exposed by the harness, then complete the campaign criterion traceability and discipline-level capability/maturity matrices.