Status of this Edition
Open the Development Atlas for the product scope, current paths and remaining work. Previous status updates ↗
This source survey describes implemented routes and known limits
at revision 8ce3ccaa49, surveyed on 14 November 2026.
The manual’s Today marks link here. The case records identify their own execution paths and source pins.
Historical records retain their earlier evidence, including revision 462f66975d.
Three pressure cases retain their numerical comparisons from 12 October 2026 at revision a72f994d89.
They have not been rerun for this edition.
These comparisons establish only their stated cases and paths.
Testing of results against a commercial validator solver has begun; this edition publishes no results from it.
The original review examined revision
bc6d3459b
(24 September 2026), then
e8562c068
(25 September 2026). Those changes admitted real project data and mixed rigid/nonlinear
support directions. The current account uses source comparison, rather than assuming
every original finding still applies.
Established section links remain valid. Each row distinguishes an implemented route, a remaining limit and historical behavior. A merge, test count or CI result does not establish general engineering validation. Read results with their route, model version and current limits visible.
The model and the route
A supported calculation, a readable result and an available output are separate capabilities.
Model document 0.9.0 is the current format. Models 0.1.0 through 0.8.0
are refused with a re-author diagnostic; the program does not migrate them. A pressure-free model omits
pressure_contract. An exact-contract model declares 3.0.0/exact_pressure_v3 and its shear model.
Pressure contract 2.0.0/exact_straight_pressure_v2 and its result identities are retired.
The model retains slots for tags, notes, assumptions and open issues, lines, constraints, equipment interfaces, rule-pack references and structured entity provenance. Lines, constraints and equipment interfaces are recorded, not evaluated. Undeclared fields, including inside nested records, are refused by name rather than ignored. A document without project units is refused at open.
| Route | Implemented calculation and desktop availability |
|---|---|
| Pressure-free | Publishes preview-physics-2. Straight members use Euler–Bernoulli by default, or opt-in Timoshenko with identity preview-physics-timoshenko-1. Realized arcs remain Euler–Bernoulli.
Nonlinear supports have a four-pass iteration limit and a separate numerical confirmation after convergence.
Imposed movement includes supported rigid restraints, spring bases and nonlinear-support ground. Outputs depend on the surface and result standing; a local private report and report package can be saved for a current result. |
| Exact contract | Publishes pressure-2. Straight members select Timoshenko or Euler–Bernoulli;
new exact-contract authoring defaults to Timoshenko. Realized arcs remain Euler–Bernoulli.
The route includes separate section bases, bend stress quantities and bounded expansion-joint laws with optional elastic tie rods.
Eligible desktop results can export result JSON and stress-neutral JSON and enter user rule checks. Eligible results also admit local-private reports and report packages. |
| Captured exact-source recovery | Can publish pressure-source-1 for eligible straight-member cases,
now including Timoshenko and nonzero mill tolerance or corrosion allowance. Pressure regions must be explicitly empty.
Bends, connectors and nonlinear supports remain outside the subset, which also has size and arithmetic-work limits.
The same output surfaces are open only when the result reader admits the case; source results with load/reference states remain ineligible. |
| Dormant precision method | The separate extended-precision successor path remains in engine/test code and is outside the v1.0 target. The desktop instead shows a numerical-health indicator built from existing diagnostics (below). Captured-source recovery above remains a distinct active mechanism. |
Source basis: model format, desktop output policy pressure and tee scope, embedded-agent scope, desktop performance measurements, and source-recovery limits.
Working in the desktop
Run-sheet entry, numbering reservations, reviewable batches, bulk load-case fills and component quantity units are joined by condition-set controls, nested combination editing, a static-seismic difference/SRSS template, tee modifier-source entry and explicit support-direction entry. Changes are checked by the operation engine and queued for Review/Apply.
A maintained authoring test enters the complete reference-system fixture through desktop surfaces and session functions, then saves and reopens it. This is functional browser-path evidence: it performs no solve and supplies no human timing or native usability qualification. The embedded agent’s full build/solve/explain journey remains separate and unfinished.
Desktop results now stay in native memory. The interface receives summaries and requests detail pages; the native reader, run recording and project save use the held result bytes. This removes the blanket 64 MiB display refusal for that path, while some consumers still require a full copy limited to 64 MiB. Autosave settings, recovery offers and unsaved-change prompts remain connected to project handling. Read the walkthrough, authoring-test scope and native result design.
The v1.0 target and remaining work
The first publication-ready target remains v1.0, with macOS on Apple Silicon first. This is development code, with no delivery date or qualified release announced. Windows development does not establish a qualified Windows release.
This snapshot adds report composition, per-report user release, native MBF import and subset round-trip checks, optional model backups, and explicit case selection for pressure elongation. These implemented paths retain the limits below.
Numerical qualification is a central open item: a constructed counterexample disproves the general inverse-norm assumption used by the current integrity check. The solver’s checks passed verdict is not an established error bound, and release criterion 18 is unmet. Large-model performance, complete reference-system workflows and the owner’s product verdict also remain open.
Modal mass, eigen-solution and reader work remain internal. No desktop, headless or report surface admits a user’s modal analysis yet. Timoshenko straight members, including the exact-route authoring default, are refused by modal preparation. Response-spectrum analysis, hydrotest, exact-route hanger sizing, applying a sizing selection and ordinary spring preload remain unfinished. See modal admission and the release criteria.
The embedded agent can read supported product information, propose changes, compute supported jobs and request human acts. Every proposed change and requested human act still needs the person’s action. The full native reference-system journey remains unqualified; Reader research is not connected.
A. Mechanics and result limits
Use the declared model route and its numerical standing when interpreting results.
| Limit | What happens, and what to do meanwhile |
|---|---|
| Pressure | The retired Model 0.9.0 accepts Constant-effort supports work on plain and load/reference-state documents. Gap, one-way, lift-off and friction supports now work on plain exact documents through the shared active-set loop. They remain refused beside an objective connector or on a load/reference-state document. Gap states retain unqualified recovery; convergence alone does not qualify a result. Each pressure region has equal bore, terminal ownership, closure-transfer paths and declared E/ν, with derived G. Shared junction terminals must transfer their cap load to the wall; compensated terminals are refused. Unequal pressures or an incident pipe outside the regions produce closure warnings. Pressure is uniform signed internal differential within a region, with zero external pressure increment. Negative differential pressure is admitted without external-pressure stability or collapse assessment. Pressure thrust remains applied. Closed-end pressure elongation is now optional and off unless the model settings explicitly name the operating cases that include it. The desktop provides a switch and case checkboxes; results disclose included, omitted or no-effect settings. This does not model pressure-induced bend opening. Mechanical, selected pressure-strain and cap-transfer terms remain distinct. Hydrostatic head, steady-flow momentum, transients, Bourdon opening, pressure-dependent flexibility/SIFs and ovalization remain excluded. Combinations retain compatibility gates. Equivalent-static generators admit plain exact documents but refuse exact load/reference-state documents. Eligible results retain the output boundary. Exact-route admission and pressure contract define the supported subset. |
| Bends drawn in the program | A realized The earlier review recorded anchor forces about 58 % high and elbow stress about 23 % low for that chord treatment in an L-bend. Those figures do not qualify the current curved mode. The pressure route includes selected pressure free strain and member-owned end-cap transfer. Realized bends use thin curved-beam end flexibility: transverse shear, neutral-axis shift, axial–bending coupling and ovalization are omitted. The engineering application range of that element has not been established; there is no automatic refusal for that limitation. User flexibility factors do not by themselves establish adequacy. Wall axial and effective force remain distinct. A region changes direction only at realized bends; mitres, straight-to-straight kinks and geometry-only pressure bends are refused. Current results add signed normal-stress extremes, transverse-shear measures and bounded maxima along the arc. An entered in-plane/out-of-plane SIF pair supplies a separate directional bending measure and its arc maximum. These are mechanical quantities, not code stresses. Pressurized arcs also publish a toroidal membrane hoop idealization and radial surface stresses when bend radius exceeds the section outer radius. The shear measure uses a local straight-annulus field without a curvature correction. Neither idealization is a general shell solution. A maximum can still be withheld when its enclosure check fails; solution error is separate. Meanwhile: read the model mode, section basis and stated approximation before interpreting bend results. |
| Unstable models | The earlier review found apparently solved mechanisms with rotations of 109 radians or more. Current structural paths apply numerical integrity and formation gates. Successful factorization alone does not establish physical stability or accuracy of every recovered quantity. Meanwhile: look at the largest displacement and rotation of every run before reading anything else. Read the result’s numerical admission and coverage limits. |
| Numerical recovery and its missing guarantee | The solver reforms the intended system at 128-bit precision, applies displacement corrections and checks recovered member and support actions. Its error estimate assumes the prepared matrix’s inverse norm is no more than 1,024 times its Hager estimate. That assumption is not established over the admitted domain and has failed on constructed models accepted through the ordinary model-input and solve path. In the recorded counterexample, a support-state interval missed the exact value and a one-way support was reported inactive and established when the exact solution has it active. Linear variants also passed their checks with internal displacement intervals that missed the exact solution. The published one-way witnesses use deciding force terms 36–39 decimal orders below the main loads. They demonstrate a failure of the claimed guarantee, not a measured error level for ordinary engineering models. The application’s checks passed label reports the solver’s verdict. It does not establish the error bounds or confirm support states on which that verdict depends. Release criterion 18 remains unmet. Reports now state this missing basis. Gap models separately remain unresolved and for inspection only; convergence alone establishes neither numerical qualification nor an admissible support state. Meanwhile: read case standing, model assumptions and independent comparisons together. The kernel’s evidence account and independent counterexample describe the limitation. No fresh solver execution is claimed by this website survey. |
| Support loads | Ordinary Historic |
| SIF review row | Geometry-only bend markers and tee/branch records retain the separate user equal-factor intensified bending measure, i·hypot(My,Mz)/Z, on supported adjacent straight ends. Realized bends now accept an explicit positive in-plane/out-of-plane SIF pair. It forms a directional intensified bending measure in the arc tangent frame and a bounded maximum along the arc. The flexibility factor, axial stress and torsional stress are not added to this measure, and it is never combined into a code stress. The retired summary multiplied stress by SIF and, by default, flexibility factor; the historical test gave 699 MPa where 47 MPa was expected. That old row is not the current measure. User-entered factors still need their own engineering sources. |
| Summary stress | Current results publish supported straight and curved circular normal-stress extrema, with torsional shear kept separate. Arc maxima now use an enclosure over the declared arc problem, not just the five sampled stations. Incomplete member/case coverage still withholds the affected headline. The exact route publishes nominal and reduced section results separately and selects no governing section basis. The new station extremes and shear measures do not supply principal, von Mises or Tresca stresses or a code stress range. The retired summary added |σby| + |σbz|. Circular bending instead uses √(My2 + Mz2)/Z. The old sum could be up to 41 % high. Work from current signed resultants, section bases and coverage evidence. |
| Expansion joints | Legacy finite-span joint definitions and their four-rate element are refused and require re-authoring. An objective connector replaces one straight span with an authored symmetric 6×6 work matrix, frame, attachments and installed reference. Its law uses small displacements and rotations. Model 0.9.0 admits explicitly untied hardware or explicit axial rods with a bilateral linear elastic law. Each rod has its own stiffness, attachment offsets, reference extension and source. Rod forces and the joint’s elastic and pressure actions are reported separately. The unpressurized law remains available. The Ideal inextensible, tension-only and gapped rods, nuts, stops and collective constraints are not supported. Neither are finite rotation, geometric stiffness or pressure- and temperature-dependent tangent laws. The replaced span supplies no member result, self-weight or inherited load. Meanwhile: use the supported connector and rod laws explicitly. An annotation-only joint supplies pipe mechanics and is refused on the pressure route. |
| Constant-effort hangers | The entered constant force acts in every solved case on both routes, including thermal-only cases. It supplies no stiffness or restraint row. Results identify its applied force; a historical zero reaction does not mean no force was applied. A mechanics combination must count that fixed force once. Absolute-sum and SRSS operands must contain no net fixed-load contribution; derived differences can meet that gate. A named refusal explains incompatible combinations. General hanger state history remains unsupported. Meanwhile: if a separately justified approximation uses an applied hanger force only in the weight case, do not also retain an active constant-force device. |
| Spring hangers and sizing | Pressure-free hanger sizing now runs in the desktop and headless runner. Supply a private hanger library in format 1.1.0, candidate supports, operating cases and your selection criteria. The desktop runs the sizing job in a capped worker and displays a read-only table, details, reader standing and two CSV tables. No supplier catalogue or suitability decision is supplied. Sizing does not apply its computed selection to the model, create a recorded run or add a report section. Exact-route sizing and held-operating-status sizing remain unfinished. A sizing result’s verification is distinct from a product release or engineering acceptance. Ordinary variable springs still use their entered rate without the installed cold-load preload. Entered installed, cold and hot loads warn rather than act as preload; inactive or locked participation is refused. Pressure-free spring-base motion is implemented. The sizing design distinguishes the computed selection from the unfinished application and preload paths. |
| Combinations and solved states | Algebraic combinations, result-state subtraction, envelopes, absolute sums, SRSS and combinations of combinations now have solve paths on both routes and desktop authoring. The editor supports typed operands, factors and nested combinations, with cycle/reference checks and explicit Review/Apply. A static-seismic template creates differences and an SRSS over them; this is not response-spectrum analysis. Mixed-modulus, fixed-load and nonlinear gates still apply. General superposition of nonlinear cases is withheld. A bounded form combining one solved case at factor 1 with increments having zero fixed-load count is admitted as a derived result, not a new equilibrium solution. Differences, envelopes and unsigned combinations preserve operand standing. The pressure-free route supports an explicit Meanwhile: solve physical nonlinear loadings together as cases. Read the combination’s basis, gates and standing before using its rows. Combination implementation and operating-case comparisons give the detailed rules. |
| Section bases and allowances | On the exact route, nominal wall supplies stiffness and pipe-metal mass. Wall reduced by entered mill tolerance and corrosion allowance supplies the reduced stress section and pressure bore. Contents mass uses that bore. Stress rows are published on both nominal and reduced section bases, even when they coincide. The pressure-free route retains its single wall-minus-mill-tolerance basis for stiffness, weight and stress. Meanwhile: identify the model route and each row’s section basis. The former blanket advice to omit mill tolerance does not describe the exact route. Source: section bases and shear model. |
B. Models that cannot yet be built or solved
These limits block some models or require a separately validated method.
| Limit | What happens, and what to do meanwhile |
|---|---|
| Nonlinear iteration limit | The product active-set iteration stops after four passes. The earlier rack examples with a dozen or more rests, or five friction shoes, exceeded this limit. A higher-limit investigation converged those recorded models; it does not establish general convergence or change the product limit. Meanwhile: keep nonlinear supports to the few that matter. Treat non-convergence as a failed solve, not an approximate solved state. |
| Imposed movements and cold spring | Model 0.9.0 supports explicit installation, material, thermal and fit reference states for independently solved straight circular members. Rigid support translations and rotations are absolute prescribed boundary values. The pressure-free route admits prescribed rigid-support motion alongside nonlinear supports, motion of nonlinear-support ground, and linear spring base motion. Gap, lift-off and friction laws use displacement relative to the moved ground. A rigid direction on a mixed nonlinear support can also move. Constant-effort supports remain refused as movement targets, as do node and element targets. Thermal and fit strain compose once as
Device preload/reference laws, inactive or locked supports, per-case mass states and installation/contact history remain unsupported. A fit reference strain is not a general cut-and-close joint model. Meanwhile: apply unsupported movements in the validating program. |
| Rigid elements and component weights | Valves, flanges, reducers, rigid and specialty components are solved as their mapped straight member with its stiffness multiplied by a user-entered scale, on both routes. There is no separate rigid element. On the exact route, body stress rows are those of the mapped reference section, not a body stress. When self-weight is generated for the mapped member, the entered component weight is applied at the entered centre of gravity as a force and its moment, replacing that member’s distributed weight. Without a valid weight and centre of gravity, a warning is given and the member keeps its pipe weight. The software supplies no scale, weight or centre of gravity. A reducer’s declared eccentric offset is checked against the node geometry, which governs; entering the offset does not place nodes. Unresolved or inconsistent run geometry is refused. Pressure-region reducers require the bounded pressure-step arrangement described above; other arrangements remain refused. Meanwhile: enter scale, weight and centre of gravity from component data with their sources. A model that also carries the weight as a separate point force counts it twice once self-weight is regenerated. |
| Restraint directions | Ordinary restraints use explicitly listed global axes; a type does not infer pipe-relative directions. A bounded pressure-free linear path now admits an authored direction vector on translational springs and rigid line stops. Springs can share a node; at most one skewed rigid stop is admitted per node, subject to the constraint checks. Skewed directions remain refused on exact-pressure documents and in a model with any nonlinear support. Motion of a skewed support, multiple skewed stops at one node, two-sided gaps and general links between nodes remain outside this subset. The creation form accepts a direction; editing or removing a stored direction is not implemented. The vertical support still uses global Z. Only supported spring families consume finite stiffness. Unused stiffness blocks the solve. Skew-restraint scope and direction entry describe the limits. |
| Friction | Ordinary friction is a separate support on one global axis. Its normal force is explicit or comes from an admitted reaction source. The active-set solve now consumes one-way and lift-off normal sources, but their numerical confirmation remains narrower. A recorded one-way-normal case cycles despite an admissible equilibrium; non-convergence does not prove that no equilibrium exists. This does not provide general friction on a lifting rest or a coupled Coulomb circle. Independent sliding axes can produce the earlier square-law behavior. |
| Nozzles and branches | General nozzle flexibility and nozzle-axis reporting remain unavailable. Both pressure-free and exact routes admit rigid tee records and supported user-SIF bending rows. A tee does not supply a general flexible branch element or an independently established branch section modulus. Declared component modifiers do not create a code-qualified fitting model. |
| Hydrotest | Hydrotest pressure is refused by the mechanics solver. Hydrostatic head and a coupled contents pressure/weight analysis remain unsupported. Model 0.9.0 carries test-condition inputs, but hydrotest and the locked-spring workflow remain unfinished. Supported pressure-free operating-condition resolution is described below. Meanwhile: check hydrotest in the validating program. |
| Units | Entered quantities need explicit compatible units. Supported inputs include bar, kN,
kN·m, GPa, N/mm and °F. Unit conversion does not supply a pressure reference:
A new project chooses SI, US customary or a custom unit for each of seven quantities. Units can be changed later, but stored values are not converted, and the length unit is fixed once nodes exist. Each stored quantity keeps its own unit; node positions use the project length unit, and new entries start in the project unit. A missing project unit is refused by name rather than assumed. A display-unit setting does not alter the physical meaning of stored quantities. Pressure-reference assumptions belong in the explicit pressure contract. |
C. Bounded and unfinished capabilities
These capabilities remain bounded, unavailable, or separate from engineering acceptance.
| Limit | What happens, and what to do meanwhile |
|---|---|
| Solve memory and large results | Desktop solves run in capped local workers with progress and cancellation. The default shared pool is 75% of physical memory; the default per-solve budget is the smaller of 12 GiB and that pool. Admission, command leases and worker growth depend on available headroom. A worker may wait or stop when the pool cannot supply its next grant; request pricing is not a measured bound on the entire application. Native-held results supply summaries, viewport data, paged tables and table CSV without the former blanket webview-copy limit. Reports, comparisons that need the current full result, rule checks and several exports still require a full copy limited to 64 MiB. Save Result JSON is a derivative export, not a download of the held envelope. File > Save Handoff has a separate path and does not require that full copy. A separate native-reader limit remains: when the priced parse exceeds 12 GiB, the completed held result cannot be read, displayed, recorded or used for output. Native custody alone does not make every completed solve usable. SQLite, reader working memory, page cache and the complete engine result envelope also retain limits. The 2,000-node performance criterion is unmet. The nonlinear operating reference rack is refused from 1,775 nodes by the dense-memory guard. Development-host timing observations do not establish the required whole-desktop times on the 16 GiB Apple-silicon base machine. There is no qualified large-model size or speed promise. See performance evidence and the remaining memory limits. |
| Code stresses | Mechanical load categories and user-authored rules do not provide a code stress implementation. Displacement ranges, allowables and code acceptance require the declared engineering method. The program supplies bounded force, moment and stress components (Part II K). Private rule packs in format 0.5.0 declare stress categories, required load cases and load-case mappings; each check runs once per mapped case or combination. A pack in another format version is refused, and a check with missing inputs is reported as incomplete. Formula evaluation supports intervals and refuses nonfinite intermediate arithmetic. A passing formula gate is not standards approval. Make the code check in the program you validate in. |
| Operating conditions | The ordinary solve selects a material basis per case and withholds mixed-modulus combinations. Model 0.9.0 records each member’s actual, selected and installation temperatures, expansion definitions and selected E/ν, with derived G. Operating-condition sets now resolve temperature change, added weight and wind exposure on the pressure-free route. Desktop controls assign sets to pipes, choose a case’s use and show the engine’s generated records before solving. Values are not defaulted. Pressure use and contents density remain refused, and exact-pressure documents refuse condition use. Hydrotest and locked springs remain unfinished. This does not create code stress categories or a cold-modulus displacement range. Condition admission and desktop condition use define the current path. |
| Saved-state comparison | Local projects can be saved and reopened. Named, immutable model states are saved in the local store, and each solve is recorded as a run against its state, with the model’s rule packs resolved at solve. The desktop compares two states, two runs, or, within one run, a chosen load case with combination rows. Reports, exports and the local report package carry the chosen within-run comparison. Comparisons are view-only. A combination with a quantity of undeclared dimension is refused by name. CSV exports of the displayed comparison tables are implemented. A mapping editor, tolerance classification, graphical differences and a complete comparison-envelope export remain unfinished, and the saved handoff carries no comparison. Result-state subtraction is a comparison of solved quantities, not a new equilibrium solve. |
| Model batch file export and import | File > Export MBF and File > Import MBF now provide native review, save and import for a declared subset. Export writes an SI pilot model and companion identity file after review and an explicit local-private choice. Import reads bounded SI and English forms, requires a material-provenance declaration, shows losses and assumptions, and adopts an admitted model through the normal open/unsaved-changes guard. Imported models start unsaved. A bound import verifies the companion file against the MBF bytes and the regenerated writer output. It restores the identities the pair preserves; designated-source checks can establish the stated identity set, not equivalence of the entire model. Edited or externally rewritten files have an unbound interpretation and stricter source-dependent refusals. The writer’s subset still excludes tees, rigid-family components, nonlinear supports, hangers and other unsupported families. Combinations are omitted and disclosed; some newer combination forms block export. Invented subset round-trip checks are implemented. Complete reference-system coverage and the required round trip through the actual desktop build remain unestablished. Target-version and record-family questions remain open. A saved or imported pair establishes no receiving-program acceptance or comparison result. Model transfer reads no solve result and has separate privacy, loss, size and memory gates. An unknown export-save outcome still blocks another export until reload or restart. See interchange scope. |
| Libraries and tables | Library import validates declared provenance and bounded payloads. An imported or selected record does not establish solve readiness. Hanger libraries require explicit identities, quantity provenance and supported units; missing engineering values remain absent. A reviewed record can be applied to a model entity, which keeps its source record; the store warns when an applied record is unresolved or has changed. No bundled catalog, sizing, ranking, source scraping or vendor parser is supplied by that import boundary. Standards, material and fitting data need their own valid sources and selection evidence. |
| Wind and seismic | Ordinary equivalent-static generators retain bounded behavior. Wind acts on entered exposed spans along the declared global axis. Straight pipes permit disjoint fractional extents. Overlapping extents or mixed whole-span and fractional declarations on one pipe are refused. Realized curved-bend spans require whole-span loading; partial extents are refused. Wind supplies no height profile or automatic projection perpendicular to the pipe. Seismic load acts in one sense per case on the declared axes. A generator does not provide a complete wind or seismic analysis. Equivalent-static wind and seismic generators now also admit plain exact-pressure documents. Exact load/reference-state documents still refuse these generators. Supported combinations retain their own compatibility gates. Pressure-free condition sets can now resolve temperature, added weight and wind exposure. Contents mass and hydrostatic analysis remain unfinished. |
| Shear deformation | Exact-contract straight members select
|
| Result points | Arc end and station rows use their declared tangent frames and signs. The current result format adds signed normal-stress extremes and transverse-shear measures at stations, optional directional SIF measures, and separately bounded arc maxima. Exact-route rows identify nominal and reduced section bases. A bounded maximum applies to the arc problem formed from published binary64 end actions and uniform load intensities. Its enclosure does not include solution error, which has separate checks. A missing bound or incomplete member/case coverage still withholds the affected maximum or headline. Five sampled values alone do not establish a governing maximum. Earlier case records can retain older row sets and chord-frame end actions; read their own source pins and metadata. |
1.Implemented paths and evidence.
The original review found specific useful foundations. Its curved-bend comparisons agreed with a 64-segment straight model to five significant figures and independent L-bend solutions to eight. Those figures describe their recorded configurations and paths.
The current source contains coordinate transformations, exact restraint elimination, sparse structural solves, consistent span/arc loads and active-set support logic. Each capability has route and input limits. None establishes broad validation, complete numerical accuracy or shipping readiness for all piping systems.
Rust supplies the product result-reader decisions across native, WebAssembly and Python consumers. Python verifier copies remain in tests. Native solves retain the producer’s exact result bytes; exported envelopes identify copied members by canonical-value hashes.
The desktop can start a project with chosen units, author, solve, inspect results, save a local project and reopen it. It saves the model document and a verified handoff package without solving. Export remains subject to the result’s standing and route policy. Saves check the stored revision and write in one transaction. Another writer’s change requires a replacement prompt; a further change before confirmation is refused and prompts again. Recorded native smoke checks exercise this path; they do not issue a release.
The maintainer’s later product-direction decision excludes a standalone verified-bounds feature from the v1.0 target. The retained-precision successor code remains dormant. The desktop shows a numerical-health indicator using the solver’s existing residual, conditioning and per-case verdicts, in the results, the status bar and the Solve panel. It is diagnostic feedback, not a guarantee of error bounds or engineering validity.
Its design records the decision. Existing captured-source recovery remains active within its supported subset; removing the standalone precision station does not remove that recovery or the current integrity checks.
Captured exact-source recovery can publish pressure-source-1 for eligible straight-member cases,
including Timoshenko and nonzero mill tolerance or corrosion allowance. It requires explicitly empty pressure regions,
linear restraints or springs and supported nodal or load/reference-state inputs, within a 256-DOF ceiling and additional work limits.
Bends, connectors, component records, actual pressure inventories and nonlinear supports remain excluded.
A failed selected-source finalization now falls back to an ordinary publication on both supported source routes, with the reason recorded.
That fallback keeps its own numerical standing and gains no retained qualification.
See the source-family implementation
and fallback design.
Desktop result JSON, stress-neutral JSON and user rule checks admit eligible pressure-2
and pressure-source-1 results. The native reader must accept the source and its numerical standing; for source recovery,
the registered live invocation is required. Pressure-source results with load/reference states remain ineligible, while an ordinary
pressure-2 load/reference-state result can qualify. Stored or copied bytes alone do not supply that live source binding.
Local-private CSV export is now available for results tables, displacement extremes, stress/ratio tables, support loads, weight and centre of gravity, and run/state comparison tables. It carries units and provenance and includes the rows admitted by the current filters across pages. This is separate from the stress-neutral interchange CSV and the MBF/PCF text files, which retain their own restrictions.
The desktop now has a report section composer and Save Shareable Report Package. Composition selects and orders admitted subsections and their cases/views; required identity, numerical standing, warnings, provenance, limitations and audit content remain. A shareable package requires a fresh statement for that report that the user has the right to share their project data. This saves a file locally; it publishes nothing and records no engineering approval.
Models carrying a rule-pack reference or detectable private-library origin are refused a shareable package. User-rule outcomes, library records and state/run/comparison and handoff sections are not carried by that context. Local-private reports retain their separate scope. An exact-route readable, solved result can now be rendered with sensitive or unresolved standing stated; package saves and user releases still require eligible standing. Pressure-free desktop reporting retains its eligibility gate.
Reports and packages still require a full result copy, refused above 64 MiB. A named Diagnostics-check limitation can also refuse some library-finding combinations. Report generation beyond the committed witnesses is not established. The solver’s checks-passed verdict is qualified by the unestablished error-bound premise. See report scope and release limitations.
A source merge, a successful check or a saved report does not release the product. The numerical-bound repair, performance evidence, remaining mechanics and full workflows still need qualification against the v1.0 criteria. Testing against a commercial validator solver has begun; this edition names no product and publishes no comparison results.