Design, size and stress-test operating models across teams, regions and time zones - before testing them on people. Three fused capabilities:

1. orgspec - the operating model as code (YAML): diffable, lintable scenarios
2. engine - an Operational Research core: a discrete-event simulation run as a Monte-Carlo experiment (seeded replications, warm-up discarded, percentiles only), cross-checked against queueing theory; the ONLY source of numbers
3. the bench - scenarios compared on one interactive dashboard: every variable a slider, P10-P50-P90 range bands, an optimized-path recommendation under constraints, one-click PDF export

An LLM copilot layer (agents/) designs, critiques and narrates - it never computes. Calculation QA (qa/) re-derives headline numbers by independent methods and holds veto power; the surrogate behind the sliders is validated against held-out engine runs and its max error is printed on the dashboard.

python3 scripts/run_experiment.py            # fictional example end to end
open dashboard.html                          # tune sliders, export PDF
cat BRIEF.md qa/QA_CERTIFICATE.md

Or bring your own model: copy orgspec/baseline.yaml, edit, add scenario diffs, then python3 scripts/run_experiment.py my.yaml my_scenario.yaml.

• The copilot narrates, the engine computes - every number traces to a run id
• Nothing ships without the QA certificate (PASS required)
• Zero proprietary data, zero named entities, zero PII - all examples fictional
• Distributions over averages; recommendations always state their accepted trade-off
• Hyphens only, plain English, answer first

See MASTER_PROMPT.md for the full system definition and agents/ for the bench roles. Run it conversationally via the /operating-model-simulator-now skill (Chief Simulator asks 7-10 discovery questions, then drives this repo).