• Compression and recompression indices, preconsolidation pressure^?The effective stress a clay has carried before. Below it the soil is stiff and recompresses; above it, it compresses far more., void ratio, unit weight and permeability per layer.
• Permeability that follows effective stress through the permeability change index^?C_k: how much the permeability falls as the void ratio falls. Holding permeability constant makes long-term settlement arrive too early., rather than staying fixed.
• Undrained modulus^?E_u: the stiffness that governs the settlement happening as the fill goes on, before any water drains out. for the immediate settlement as the embankment is placed.
• Water table depth, with the in-situ effective stress profile computed and plotted.

How this is checked
The layered stack is measured against the per-layer closed form: 0.530 % mean error, and 2.179 % at worst on one point of the steep early rise. See the benchmark table →

• Real clays do not yield at a single pressure: a natural deposit holds a spread of particle arrangements, bond strengths and stress histories, and they do not all give way at once.
• PVDCal blends the compression index across the yielding zone instead of stepping it, which is both the more honest description and the better-behaved one numerically.
• Settlement is integrated in small effective-stress increments, each evaluated at the geometric mean^?σ′mid = √(σ′i · σ′f). On a logarithmic compression law the geometric mean, not the arithmetic one, is the natural midpoint of a stress step. of its own end points.
• The width of the transition is yours to set, and the compression path is drawn so you can see exactly where the layer yields.

How this is checked
No closed form exists for a blended yield transition, so it is checked at its limit instead: collapse the transition to zero width and the classical step response comes back. What a benchmark can and cannot prove →

• Band (I-shape) dimensions converted to the equivalent diameter^?d_w: the diameter of the circular drain that drains at the same rate as the band. The unit cell is axisymmetric, so a band section cannot be carried through it directly — every radial solution for PVDs works this way., or a circular drain entered directly.
• Square or triangular patterns, with the radius of influence^?r_e: the radius of the soil cylinder each drain serves. Spacing follows from it: S = r_e / 0.564 for a square pattern. and the drains per hectare each implies.
• Smear^?The soil disturbed by pushing the mandrel in. Its permeability is reduced, which slows drainage near the drain — often the single most important detail in a PVD design. and transition zones with their own permeability ratios.
• Well resistance checked as D = q_w/(k_hl_m²) against the discharge capacity you enter, with a flag when D falls below 5.

How this is checked
Radial consolidation with smear is measured against Hansbo: 0.299 % mean error, 1.030 % at worst. See the benchmark table →

• Mark any layer as having no drain: the treated depth drains radially to the drain, while the untreated layer below keeps draining vertically to whatever boundary it has.
• Below the drain tip the mesh switches to a full disk to the axis, so flow around and under the tip is resolved instead of assumed.
• Splitting a layer at the drain tip is offered as a decision, not done silently, and the split inherits the parent layer's parameters.
• The case is verified at its limits: a drain through the full depth reproduces the classical radial solution, and no drain at all reproduces vertical consolidation.

How this is checked
Checked at both of its limits: a drain through the full depth reproduces the radial solution, and no drain at all reproduces vertical consolidation — 0.016 % mean error across the pair. See the benchmark table →

• Multi-stage ramp loading with rest periods and surcharge removal.
• Vacuum preloading^?A sealed membrane and pumps lower the pore pressure in the drains. It adds to the embankment load without adding fill, and helps stability. as a suction boundary on the drain, including loss with depth.
• Unload and reload along a recompression path, so removing surcharge is not treated as virgin compression.
• Creep^?Creep: the slow compression that continues after excess pore pressure has gone, set here by the ratio of the secondary compression index to the compression index. carried through the run, so long programmes are not flattered.

How this is checked
Ramp loading is measured against Olson–Barron (0.392 % mean error) and the vacuum boundary against the published analytical solution (0.530 %). See the benchmark table →

• Latin-Hypercube sampling^?A sampling scheme that spreads samples evenly over every input range, so a few hundred runs cover the space that random sampling would need thousands for. across the ranges you set.
• P10, P50 and P90^?P10 is the spacing exceeded by 90 % of the sampled profiles — the conservative end. P50 is the median; P90 the optimistic end. spacing for both patterns, with the drains per hectare each implies.
• A tornado of what moved the spacing: each declared range on its own, in millimetres — the same chart in the app and the report.
• Runs spread over parallel workers, with results identical to a single-threaded run.

How this is checked
The bands were scored against ground whose answer was known: the design range contained the true final settlement in 77–79 % of several hundred synthetic sites, with the acceptance criteria registered before the runs. How that was measured →

• Consolidation and settlement against time, settlement-based or pore-pressure-based.
• Effective stress with depth, and the compression path per layer.
• Settlement plates and pore-pressure probes at any depth you choose.
• The numerical mesh^?The finite-volume grid the solver actually used: rings in radius, slices in depth, refined through the smear and transition zones. itself, with a zoom lens over the cells near the drain.
• Excel export from any chart, and image or animation export for reports.

How this is checked
The same numbers whatever the machine: a parallel run and a single-threaded run of the same case differ by exactly zero on settlements, pressures and plotted curves. How the workers are kept identical →

• Play it through the programme, or drag the time slider to any day and read the state of the ground at that moment.
• Switch between degree of consolidation and excess pore pressure in kPa, with the smear and transition boundaries marked on the radius axis.
• Slice it: a profile with depth at a chosen radius, and a profile with radius at a chosen depth, both moving with the animation.
• Export an animated GIF for a report or a presentation, or a snapshot at one instant — the loading schedule and settlement play alongside, so cause and effect sit on the same screen.

How this is checked
It is a view of the same solved field the benchmarks are scored on, not a second calculation — the mesh it is drawn on is the mesh the answer came from. See the benchmark table →

• Settlement plates and piezometers used together, in their own units.
• Sweep Search^?The guide's name for the method: a planned sweep of three knobs inside walls set by your laboratory values, then a score-weighted band. Nothing is sampled at random, so the same inputs always give the same answer. over permeability, compressibility and rate, within the walls you set — minutes in a browser, where a blind random ensemble of the same family takes hours to days.
• Fit statistics, the parameters recovered, and a flag when a value sits on a bound.
• A forecast range for the settlement still to come, with the reliability of that range measured on synthetic grounds whose true answer is known.

How this is checked
Tested on the Ballina trial embankment: 1.5 % from a 400-day record and 0.8 % from 700 days — and 9.4 % out, with its range missing, from 200 days, which is reported beside the successes. See the Ballina case →