Connect to the Mechanical session that holds your LS-DYNA system. Args: port: gRPC port of the Mechanical session (changes every session)

Disconnect, leaving Mechanical running. Args: shutdown (True sends Mechanical a forced exit request whoever started it, so unsaved work is lost; a session Workbench owns may restart or linger, so normally close it from Workbench instead), confirm (shutdown=True is refused unless confirm is exactly "SHUT DOWN MECHANICAL")

List every analysis in the Mechanical model, flagging which are LS-DYNA (solver target LSDYNA@LSDYNA, or RestartLSDYNA@LSDYNA for an LS-DYNA Restart system).

List the public properties of an object in the LS-DYNA analysis tree. Use it to discover real property names before calling set_analysis_settings. Args: path: dotted path below the analysis, e.g. "" (the analysis itself), "AnalysisSettings", "Children[0]"; analysis_index; max_items

Read the Analysis Settings. Each row has the property name ("Step Controls/Endtime"), the Details-panel caption ("End Time"), the raw value, its type, and whether it is read-only/visible. Enum settings (solver precision, processing type...) come back as integers; -1 is the Program Controlled default. Values are in the model's solver units (End Time here is seconds). Args: analysis_index; group (name prefix, e.g. "Output Controls"); visible_only

Set Analysis Settings; each value is read back and compared. Each key is attempted independently and reported. A key matches the full property name ("Step Controls/Endtime"), the caption ("End Time") or the last name segment, ignoring case and punctuation; if a caption is ambiguous (e.g. "Time Step Size" exists under Step Controls and EM Step Controls) the reply lists the full names to use instead. Values are set raw, in the model's solver units (End Time is seconds); numeric text such as "5e-4" is parsed (text with a unit in brackets is refused, never silently converted), true/false become 1/0, and plain integer settings take their integer value (read the current one with get_analysis_settings; -1 is Program Controlled). Non-finite values and values outside a sensible range (End Time in (0, 1e6) s, cycle counts > 0, time-step factors in (0, 1]) are rejected before anything is sent. Each value is read back and compared (NaN and Infinity never compare equal): a setting that keeps its old value is reported as 'not applied'. Selection (drop-down) settings such as Hourglass Type, Output Format, Stress File Type or Automatic Mass Scaling were seen to accept only -1 through this interface; set those with keyword snippets (add_keyword_commands, set_dynain_output) or in the GUI. Args: settings_json: JSON object, e.g. {"End Time": 0.0005, "Output Controls/Stress": 1}; analysis_index

Add a Keyword Snippet (Commands object) holding raw LS-DYNA keyword text, for cards Mechanical has no native object for (e.g. *CONTROL_ENERGY, *DEFINE_CURVE). Leading/trailing blank lines are stripped: trailing blanks are read as empty mandatory cards and break the solve. INCLUDE cards in the text may only name files inside the allowed folders (the solver itself would otherwise open them, e.g. a network path). Args: keywords (must start with a '' card such as *CONTROL_ENERGY; do NOT add *KEYWORD, the snippet goes inside the deck Mechanical writes); analysis_index; scope_path: where to insert it, as a dotted path from the Model, e.g. "Geometry.Children[0].Children[0]" for a body (*MAT, *EOS, *HOURGLASS) or "Connections" (contact types the system lacks). Empty = under the analysis (any keyword).

List the cards and fields of an LS-DYNA keyword (field name, type, column, width, default), read from the installed PyDYNA package (pip install ansys-dyna-core; it holds Ansys-generated definitions of about 3,100 keywords). The name is matched exactly, hyphens included (MAT_SAMP-1 and MAT_SAMP_1 are different keywords). default is what build_keyword writes for a field you leave out (PyDYNA writes the manual default explicitly, e.g. SPALL 2.0). builder_supported is false for keywords with repeating-row (table or point) cards such as DEFINE_CURVE, and for keywords with file or path fields: use add_keyword_commands for those. Args: name (e.g. MAT_JOHNSON_COOK, with or without the leading *)

Write one keyword card as LS-DYNA text, laid out by PyDYNA's own definition of the keyword (nothing here is typed from memory). Unknown field names and wrong value types are refused. Fields you leave out are written with the manual default (PyDYNA writes it explicitly, for example SPALL 2.0): the reply's defaults_written lists them. After writing, the text is read back and every value you gave must come out unchanged, else ok is false and the field is named (value too wide for its column, a float that would lose digits, a field of an inactive conditional card). A field name that occurs in several cards must be addressed as "card3.name" (card numbers as in describe_keyword). title adds the _TITLE option (printable ASCII only, no control characters). Cannot do: keywords with repeating-row (table or point) cards such as DEFINE_CURVE or *PART with a table, and keywords with file or path fields; use add_keyword_commands with explicit text for those. The same safety checks as add_keyword_commands apply. With add_to_analysis=true the card is added as a Keyword Snippet (scope_path as in add_keyword_commands); otherwise the text is only returned. Args: name (e.g. MAT_JOHNSON_COOK), values (field name -> value, see describe_keyword), add_to_analysis, analysis_index, scope_path, title

Write the LS-DYNA keyword deck Mechanical would solve to output_path, without solving (Analysis.WriteInputFile). If this release lacks it, falls back to a generate-input-only solve and copies input.k. With Split Input File = Yes, the deck includes control.k, mesh.k etc. that stay in the solver working directory. output_path must be a .k/.key/.dyn file inside an allowed folder and must not exist unless overwrite=True. Args: output_path, analysis_index, overwrite

Solve the LS-DYNA analysis. Mechanical's local solve only runs blocking (it refuses to run in the background), so with wait=False (the default) this server runs the blocking solve on a background thread and returns at once with the solver working_dir: watch the run with check_run_health / tail_solver_file on that folder, and get_solve_status says 'Solving' until it ends. While a solve runs, other Mechanical tools report 'busy'; the file and job tools stay responsive. With wait=True the call itself blocks until the solve ends. Args: analysis_index, wait

Solution status, Mechanical messages and the solver working directory. While a solve started by solve_analysis is running, answers 'Solving' without touching Mechanical (which is busy), then returns that solve's result once. Args: analysis_index

DISABLED unless the server is started with LSDYNA_MCP_ENABLE_SCRIPT=1. Run a custom IronPython script inside Mechanical (full ACT API) with stock IronPython semantics. Args: script

Add an LS-DYNA analysis (restart=True: an LS-DYNA Restart analysis) to the Mechanical model via Model.AddLSDynaAnalysis / AddLSDynaRestartAnalysis. The Workbench system itself has template ids LSDYNA and RestartLSDYNA (see create_lsdyna_system_snippet). The new analysis is detected as LS-DYNA. It cannot be deleted through the scripting API (null-reference error), so remove it in the GUI if unwanted. Args: restart

Workbench journal lines that create an LS-DYNA (or LS-DYNA Restart) system, for execute_workbench_script in the Workbench MCP. Template ids come from the installed LSDYNA ACT extension. Args: restart

Add a load / boundary / initial condition to the LS-DYNA analysis and scope it. kind (case and spaces ignored): Velocity, InitialVelocity, Acceleration, EarthGravity, FixedSupport, Displacement, RemoteDisplacement, Force, RemoteForce, Moment, Pressure, HydrostaticPressure, LinePressure, RotationalVelocity, RotationalAcceleration, DetonationPoint, ImpedanceBoundary, BoltPretension, ElementBirthAndDeath, NodalForce, NodalDisplacement, FrictionlessSupport, SimplySupported, FixedRotation. properties_json sets properties by name, e.g. {"DefineBy": "Components", "XComponent": "5000 [mm s^-1]"}; each is reported, and unknown names say so (use inspect_object on the new object's path to see what exists). Properties are applied in the order given (put "DefineBy" first). If a property or the scope fails the new object is removed again, so a retry cannot create duplicates (keep_partial=true keeps it). Without a named_selection the object is created unscoped. A Velocity/Displacement overrides Fixed Support on the same body (Users Guide 8). Args: kind, named_selection, properties_json, analysis_index, keep_partial

Add a result object under Solution. kind: TotalDeformation, DirectionalDeformation, EquivalentStress, EquivalentPlasticStrain, MaximumPrincipalStress, MinimumPrincipalStress, ShearStress, NormalStress, TotalVelocity, DirectionalVelocity, TotalAcceleration, ForceReaction, Position, UserDefinedResult, EnergyProbe, DeformationProbe, VelocityProbe, ReactionProbe, StressProbe, StrainProbe. For d3max/d3part see set_analysis_settings (Capture Maximum Stress Over Time, Selective Output). Properties are applied in order; a failed property or scope removes the new object again unless keep_partial=true. Args: kind, named_selection, properties_json, analysis_index, keep_partial

Add a Binout Tracker (SolutionInformation.AddLSDYNAGeneralTracker). branch is a binout branch: abstat, bndout, dbfsi, defgeo, deforc, elout, glstat, jntforc, matsum, nodout, rbdout, rcforc, rwforc, secforc, sleout, spcforc, trhist, randles, isoPot. Add trackers BEFORE solving. Branch, then sub-branch, then component are set in that order; a failure removes the tracker again unless keep_partial=true (list_binout_tracker_options shows the valid names). Args: branch, component (e.g. kinetic_energy), subbranch, named_selection, analysis_index, keep_partial

Which binout branches and components a Binout Tracker accepts in this release (queried from the solver extension, not hard-coded). With no branch: the branches (e.g. glstat, matsum, rcforc, spcforc...). With a branch: that branch's components. A throwaway tracker is created and deleted. Args: branch, analysis_index

Members of the Solution Information object (solver output, time step / energy histories) to find what this release exposes. Args: analysis_index

List LS-DYNA output files (d3plot, d3hsp, glstat, binout...) in a solver directory. Args: directory (the working_dir from get_solve_status)

Last lines of a solver text file plus termination status. The default name (empty, or the bare d3hsp / messag) picks the NEWEST d3hsp[nn] / messag[nn], i.e. the file the latest run or restart is writing; the reply's file says which one was read. Any other name (d3hsp01, mes0000, lsrun.stdout.txt...) is read as given. Text is redacted (see README, Privacy). Args: directory, name, lines

Judge a run from glstat (or binout) and d3hsp. Checks termination, energy balance, hourglass energy and negative sliding energy at their worst point in the run and, for quasi-static runs, kinetic energy. The reply has a verdict: 'running' (a job of this server is still active in the folder), 'no_results' (nothing to judge), 'truncated' (normal termination before the requested end time), 'failed' (error termination), 'suspect' (energy problems, or the energy file looks left over from an earlier run), 'incomplete' (no termination banner yet: still running or killed), 'unverified' (finished but no energy data) or 'ok'. ok: true only means the files were read, not that the run is good. Staleness is judged by file times: run each job in a clean folder. Args: directory, hourglass_limit (max hourglass/internal, default 0.10), balance_tol (allowed deviation of the energy ratio from 1), quasi_static (also flag kinetic energy above 5% of total, a rule of thumb)

Energy history from glstat, thinned to at most max_states evenly spaced states. Args: directory, max_states

Summarise a .k/.key file: keyword counts and missing essentials (termination, d3plot output, timestep control). Follows *INCLUDE, so a Split Input File deck (input.k pulling in database.k, control.k, mesh.k, component.k, contact.k) is scanned as a whole. Args: path, follow_includes

Read-only audit of an LS-DYNA keyword deck before it is solved (plain text parsing, nothing is executed). Reports the cards present by family, duplicate ids (MID, PID, SECID, EOSID, HGID, LCID/TBID), references to ids that do not exist (PART to SECID/MID/EOSID/HGID, curves), first-card format problems, whether MAT/PART cards look Workbench-written or snippet-written, a guess of the unit system, the end time against the output intervals, and a table of known Workbench defaults (for example *MAT_JOHNSON_COOK SPALL 0, which LS-DYNA treats as 2) each with the place in the LS-DYNA keyword manual it was read at, or marked unverified. Follows *INCLUDE only inside the allowed folders. Returns findings (severity info/warning/error, card, field, message) and a compact summary. Args: deck_path, include_children (follow *INCLUDE files)

Read-only field-by-field comparison of two keyword decks, for example a standalone reference against the deck Workbench wrote. Cards are matched by keyword and id (MID, PID...), not order; card name case, comma versus fixed format, number formatting (1.0e3 equals 1000) and *PARAMETER values are normalised. Fields of the common cards carry the manual's field names (MAT_JOHNSON_COOK, MAT_ELASTIC, MAT_PLASTIC_KINEMATIC, EOS_GRUNEISEN, SECTION_SOLID/SPH/SHELL, CONTROL_TERMINATION/ SPH/HOURGLASS/TIMESTEP/CONTACT/ENERGY, DATABASE_BINARY_D3PLOT, INITIAL_VELOCITY, HOURGLASS); other cards use "field N". Status per row: equal, differs, only-in-reference, only-in-candidate, default-vs-explicit (one side blank and the other the manual default); known_workbench_default marks a candidate value that is a known Workbench default (see audit_deck). At most 300 rows (differences first); truncated says if more existed. Args: reference_path, candidate_path, cards (keyword prefixes such as ["MAT_JOHNSON_COOK", "CONTROL_"]; default all), candidate_unit_scale ({"length": 1000.0, "mass": 0.001, "time": 1.0}: how many candidate units one reference unit is, per base quantity; converts density, stress, velocity, energy and time of the reference before comparing; absent = compare raw values)

Count errors, warnings and implicit step retries in d3hsp, messag or mes0* files: '*** Error nnnnn', 'FATAL ERROR', 'Nonlinear solver failed to find equilibrium', 'automatic time step size DECREASE, RETRY step'. The default name (empty, or the bare d3hsp / messag) picks the newest d3hsp[nn] / messag[nn]; file in the reply says which was read. Args: directory, name

Steer a running LS-DYNA job by writing a sense-switch file (d3kil) in its run directory. sw1 stops and writes d3dump (continue later with a simple restart); sw3 writes restart files and keeps running; sw4 writes a plot state; swb/swc/swd write dynain (swd also terminates); sw2 prints status; swa flushes buffers. For implicit runs: conv (force the step to converge, troubleshooting only), ttrm (retry the step with a smaller increment), rtrm (stop at the end of the step), lpri / nlpr / iter (toggle solver output). LS-DYNA reads and deletes d3kil while running; a switch sent to a run that already finished is refused. A reply of ok means the file was written, not that the solver has acted on it yet. Args: directory (the solver working_dir), switch

Validate a Simple / Small / Full restart before solving: the documented rules (e.g. a small restart is impossible for a serial single-precision solution; a small restart needs a later end time) and, if directory (the earlier run's folder) is given, whether the restart files it needs are really there (d3dump for simple/small, d3full for full). Times of 0 are treated as unknown. Args: restart_type, solver_precision (single|double), processing_type (SMP|MPP), number_of_cores, previous_end_time, new_end_time, directory

Restart files a run left in its folder: d3dump* (simple/small restart), d3full* (full restart) and dynain / dynain.lsda (initial state for a follow-on analysis), with sizes and times. Args: directory

List the LS-DYNA executables Workbench knows about (lsdyna_solvers.xml: version, SMP/MPP, MPI type, precision, path). Default location is %APPDATA%/Ansys//ACTLSDYNA/lsdyna_solvers.xml. An xml_path must be inside an allowed folder. Args: xml_path (optional)

Browse and read binout (LSDA) results, where Workbench keeps glstat, matsum, rcforc, spcforc, sleout, nodout... Call with no branch to list branches, then branch to list sub-branches/variables, then branch+subbranch+variable for the data (thinned to max_points). Needs: pip install "ls-dyna-mcp[binout]". Args: directory, branch, subbranch, variable, max_points

Read an ASCII time-history file (matsum, rcforc, sleout, spcforc, rwforc, nodout). Returns the record count, a min/max/last summary per part/interface/node/wall from ALL records, and records thinned to max_records (at most 500). NaN/Inf values come back as null and are counted in the summary as nonfinite_values. Use this when the run wrote ASCII files instead of (or besides) a binout. Args: directory, kind (one of the six), entity (optional id to keep), name (file name if not the default, e.g. matsum), max_records

Field results from the d3plot: where and when a quantity peaks. With no field it lists the fields stored. With a field (node_displacement, node_velocity, element_solid_stress, element_solid_effective_plastic_strain, element_shell_stress...) it returns, for up to max_states evenly spread states (or the single state), the min / max / mean, the id of the node or element holding the max, and a count of non-finite values (a blown-up run). Vectors give their magnitude and stresses von Mises (worst integration point). Needs lasso-python. Args: directory, field, state (-1 = spread), max_states (at most 20)

Implicit (statics / dynamics) solver progress from messag: per time step the iterations to converge, stiffness reformations, the last residual norms and whether the step converged, plus totals, the count of recoverable DIVERGENCE warnings and any lines about non-convergence or step-size cuts. Use it to judge a Workbench Transient/Static LS-DYNA run that is slow or stuck. Args: directory, name (default: the newest messag[nn], else mes0000; or the captured screen log lsrun.stdout.txt), max_steps (the last N steps are returned in steps)

Make the run write a dynain (final deformed shape and stress state) in the format you can use, by adding a *INTERFACE_SPRINGBACK_LSDYNA keyword snippet. Checked on the real solver: FTYPE 0 writes an ASCII dynain (which prepare_dynain can turn into an initial-state deck), 1 writes dynain.bin, 2 both, 3 dynain.lsda (the Workbench default, not convertible here). The Workbench setting 'Stress File for flexible parts' cannot choose ASCII through scripting, so leave it off (0) or the deck gets two springback cards. Args: psid (the part set to write; must exist unless part_ids_json is given), fmt (ascii | binary | both | lsda), part_ids_json (JSON list of part ids: a *SET_PART_LIST with this psid is added), analysis_index

Search the LS-DYNA user guide (HTML help) that Ansys installs, for example to find the options of a setting such as Hourglass Type. Only that installed help is read, nothing is written. A page matches when its text contains every word of the query (case-insensitive); hits are ranked by the number of matches and by words in the page title. Each hit has the file name, the page title and a short excerpt. The help gives option names and descriptions, not the integer values behind a setting; after setting a value, call get_analysis_settings to see what is stored. The help is an optional Ansys install component. Args: query, max_results (1-20), help_dir (optional folder with the .html pages; must be inside an allowed folder; default: the help of the installed release, or LSDYNA_MCP_HELP_DIR)

What loads and results can be scoped to: the named selections (name, kind, entity count), the parts and bodies of the geometry (name and geometry id, to use with the location options of add_load_or_condition), and the objects already in the analysis. Args: analysis_index, max_items

The Add* methods the analysis (loads, supports, conditions), its Solution (results, probes) or its Solution Information (Binout trackers: AddLSDYNAGeneralTracker...) offers in this release, so a kind can be found without guessing. host "solution" also lists the Solution Information methods (in solution_information_methods), because trackers live there; host "solution_information" lists only those. By default the acoustic, frequency-response, PSD/response-spectrum, electromagnetic and optimisation items, which do not apply to LS-DYNA, are left out (all_physics=true lists them). Args: host ("analysis", "solution" or "solution_information"), filter (case-insensitive text), analysis_index, all_physics

Delete one object this server (or the user) added to the analysis: a load, support, result, tracker or keyword snippet. Only these paths are accepted: "Children[n]" (loads, supports, snippets), "Solution.Children[n]" (results) and "Solution.SolutionInformation.Children[n]" (trackers). The object's current name must equal expected_name, so a stale index (indexes shift when objects are added or removed) deletes nothing. Analyses, Analysis Settings, Solution, Solution Information, Initial Conditions and anything outside the analysis (named selections, mesh, contacts) are refused. Look first with list_scoping_targets or inspect_object. Args: path, expected_name, analysis_index

Run LS-DYNA on a .k file outside Workbench (what LS-Run does). Only the solvers listed in Workbench's lsdyna_solvers.xml can be run, and the input must be inside an allowed folder. Jobs queue (max concurrent: env LSDYNA_MCP_MAX_JOBS, default 1; one job per folder); output goes to lsrun.stdout.txt in the input's folder. SMP: lsdyna_*.exe i=... ncpu=...; a NEGATIVE ncpu gives reproducible results at up to 15% more CPU time. MPP: mpiexec -np N (IntelMPI uses -localonly). Needs the Ansys LS-DYNA license (LSTC_LICENSE=ANSYS) and, beyond one core, HPC licenses. Args: input_file; ncpu (non-zero, at most 512 in magnitude); process (SMP|MPP); precision (single|double); mpi (IntelMPI|MSMPI); memory (words, e.g. 200m; empty = solver default); depends_on (job id to wait on for normal termination, -1 = last submitted); version; extra_args_json (JSON list of key=value strings; only case, memory2, jobid are accepted, e.g. ["case=1"]); dry_run (build the command, don't run); restart_file (name of a restart file such as d3dump01 or d3full01 in the input's folder: the run is started with r=; check_restart_setup says whether a restart is possible and what the input must contain)

All jobs submitted in this server session with status (pending, running, done, failed, stopped, cancelled) and termination (normal/error/unknown from d3hsp).

One job plus the tail of its lsrun.stdout.txt. Args: job_id, tail_lines

Cancel a pending job. For a running one, send sw1 (it stops and writes d3dump, so it can be restarted) or, with force=True, terminate the process. Args: job_id, force

DISABLED unless the server is started with LSDYNA_MCP_ENABLE_LSPREPOST=1. Run an LS-PrePost command file (post-processing automation) in working_dir (must be an allowed folder). headless=True uses 'runc=' (no graphics: only non-graphical commands work); False uses 'c= -nographics'. The call runs on a worker thread with a hard timeout (at most 900 s) that kills the whole process tree. Files created are listed. Args: commands (cfile text, at most 20000 characters), working_dir, headless, timeout_s

EXPERIMENTAL and DISABLED unless LSDYNA_MCP_ENABLE_LSPREPOST=1. Have LS-PrePost load a d3plot and write its lspost.msg summary (s=). In testing LS-PrePost 4.13 crashed (0xC0000005) loading a d3plot without a GUI, and the reply then says so. Prefer read_binout / check_run_health. Args: d3plot_path, timeout_s

Which *MAT/*EOS cards Workbench can write, and how each field maps to Engineering Data (from the installed LSDYNASolverRepresentation.xml). query is matched against the card name, e.g. "JOHNSON_COOK" or "*MAT_ELASTIC". An xml_path must be inside an allowed folder. Args: query, xml_path (optional)

Explicit-to-implicit sequential solution, step 2: copy the solver's ASCII dynain file to dynain.k so the External Model system can import it. Get an ASCII dynain with set_dynain_output(psid, fmt="ascii") in the explicit run, and leave Workbench's own Output Controls > Stress File for flexible parts OFF: that setting cannot choose ASCII through scripting (the Workbench default is the binary dynain.lsda, which this tool cannot use; multi-system transfers use the .lsda file directly, no conversion needed), and turning it on as well gives the deck two springback cards. Rigid-body displacements are not saved. An existing dynain.k is kept unless overwrite=True; a dynain larger than LSDYNA_MCP_MAX_DYNAIN_MB (default 2048) is refused. Args: directory (the explicit run's working_dir), overwrite

Write the deck Workbench would solve (the same Analysis.WriteInputFile path as write_solver_input, no solve) into an allowed folder, audit it (audit_deck) and, with reference_path, compare it with the reference deck (compare_decks). safe_to_solve is false on any error-severity finding (from the audit or the comparison, for example a Workbench default such as *MAT_JOHNSON_COOK SPALL 0 against the reference's 3) and when the reference has cards the written deck lacks (end time, initial velocity, contacts, materials lost by an External Model import). True only means nothing blocking was found: read reasons and diff. The file goes next to the reference (else into the first allowed folder) under output_name (a plain .k name; default wb_written_.k); an existing file is never replaced. Args: analysis_index, reference_path (the standalone deck to compare with), output_name

Progress of a running or finished solve, read-only (never touches or kills a process). From messag / mes0nnnn (MPP), restart messag01 / mes010000, d3hsp and lsrun.stdout.txt: latest time, cycle and time step, percent of the end time, the solver's own estimate lines when it printed them (estimated total / to-complete cpu and clock time, percent complete), our own rate and ETA from successive calls (observations are kept in the per-user state folder, not in the run folder, so call it again a minute or more later), hang hints (no solver process, newest output file older than stale_minutes, d3plot not advancing, simulation time unchanged between calls) and the solver's error block (Error nnnnn, termination due to energy, forrtl) from the tail. state: running, finished, failed, stalled or unknown. The ETA assumes a constant rate. Args: working_dir, stale_minutes

Early-time gate: compare a scalar history of a short run with a reference CSV (time,value) before committing to the long run. quantity: 'glstat:kinetic_energy' (glstat, or the binout glstat) or '::' for matsum, rcforc, sleout, spcforc, rwforc, nodout (for example 'matsum:1:kinen', 'nodout:5:x_vel'). The reference is interpolated at the simulation's output times inside [t_min, t_max] and multiplied by reference_scale (unit conversion; times must already be in the model's unit). Returns points compared, max and RMS relative error (%), the worst point, the first time the error exceeds tolerance_pct and a verdict within_tolerance / exceeds. A mismatch means diagnose the model (audit_deck, compare_decks): never tune parameters to match an experiment. No solver is run. Args: working_dir, reference_csv, quantity, t_min, t_max, tolerance_pct, reference_scale

Check before a solve, read-only (nothing is deleted or started). Reports free RAM and free disk, a memory estimate (the solver's own 'Memory required' report in a previous d3hsp of the folder when there is one; otherwise a rough heuristic from the node and element counts of the deck, uncertain by a factor of about 3 either way), the LS-DYNA result files in the folder that a new run would overwrite (names only), ncpu against the physical and logical cores, and the unit-system hint and error findings of audit_deck. verdict: ok, warn or stop, with reasons. Give working_dir (the newest .k in it is used) and/or deck_path. Args: working_dir, deck_path, ncpu (0 = not given), memory (words as for the solver, e.g. 200m)