Business Gym
Fork your operations into isolated instances, run what-if scenarios, decide with reproducible evidence. Built on the Blobfish world factory — every number is reproducible: seed + content digest.
The 3D isn't decoration — it's the working surface of a pilot. Three verbs: watch, build, decide.
Every pilot replays its simulated day on your facility floor — flat plan or full 3D, same deterministic clock. Work items travel the aisles, queues pile up and change color, errors flash, completions pulse at the exit. Click any station to open what's underneath: the mined process step, the world tool's real source, live database rows.
Console → open an instance → World tab → 3D
Open your consoleDrag machinery from a 24-piece kit — AGV tuggers, smart conveyors, self-service kiosks, robotic arms — or mint a custom brick with its own size, effect, and price tag. Snap it to the grid, wire it to a process step, and export the whole floor as OpenUSD: it opens in NVIDIA Omniverse or Isaac Sim with physics colliders and the economics riding on every prim.
Instance → Build in 3D tab
Start from a templateRun the assembly against the identical seeded demand as the baseline. The Impact Report prices the difference: payback months on the capex, operating cost per year, whether the bottleneck actually moves or just shifts to the next station — and the machine appears back on the World floor.
Run → KPIs & Impact Report tab
Run a one-click demoOpen a generated business world, then drag machinery and operational assets from the reusable 3D kit onto its floor. Snap each component to a process step and set the speed, error, and cost assumptions you want to test.
Define what-if scenarios — staffing changes, step overrides, automation, demand surges — and run them against an isolated instance. Baseline and variant share identical seeded arrival streams (common random numbers), so the difference you see is the change you made.
The Impact Report shows every KPI as baseline vs variant with mean ± standard error, absolute and % deltas, and a direction-of-good verdict — plus per-seed values and a reproducibility block (seeds, content digests, engine version).
These are not customer logos or marketing claims — they are live results from this engine on public templates. Every figure is deterministic: the seeds and scenario digest are printed next to it, and the matching one-click demo reproduces it exactly.
Warehouse Automation (AGV pick-and-pack)
+5.14% composite KPI change
seeds 101, 102, 103 × 7d · digest f57160ab2c…
Reproduce this in one clickQuick-Service Restaurant (store operations)
+3.97% composite KPI change
seeds 101, 102, 103 × 7d · digest e6e0f444cb…
Reproduce this in one clickAutomotive Components (weld & inspection line)
+2.81% composite KPI change
seeds 101, 102, 103 × 7d · digest a263cf01e7…
Reproduce this in one click~2 min template → first pilot, self-serve
Pick a template or bring one of your generated worlds; the first replay is one click later.
0 consultants or site visits required
Consulting-built digital twins typically take a site-modeling engagement measured in months before the first virtual pilot.
A–D backtest grade on every instance
The engine grades itself against held-out history before you trust a forecast — ask any alternative for the same number.
A ready-made business world (or one of your generated worlds) you can instantiate.
An isolated fork of a template's database — your experiments never touch the original.
One scenario run: baseline vs variant simulated over the same seeded demand.
Per-KPI baseline vs variant statistics with verdicts and a reproducibility block.
A simulation cannot prove a future. What it can be is calibrated against your past, transparent about what it assumes, explicit about how much each assumption matters, and accountable for what it predicted. Every Impact Report carries all four layers:
The engine is fitted on the early window of your own history and graded A–D on the held-out rest — including whether it reproduces how your operation slows down on busy days, the exact queueing mechanism every lift flows through.
Every parameter says where it came from: mined from your data, heuristic default, domain archetype, or assumed (vendor claims). Nothing assumed is ever dressed up as measured.
For each assumed number, the report computes the break-even where the recommendation flips and hands you a field protocol: the one measurement a short real-world trial must produce to confirm — or kill — the forecast before capital is committed.
When you implement a pilot, record what actually happened: the frozen forecast is graded against your measurements and the instance builds a public track record — a ledger the model cannot hide from.
Anyone can hand you a KPI forecast. Business Gym is the one that grades itself first: every instance fits the engine on an earlier window of its world's history, simulates the held-out later window, and shows you the simulated-vs-actual error per KPI — an A–D grade right on the Impact Report. Upload your own historical export and the backtest runs against your reality instead. When the history can't support a split, it says so — no grade is invented. Ask any forecast vendor for the same number.
Free first pilot instance
Per instance / month — priced with design partners
Custom — modeled with you