Learn
Glossary
Twenty four terms the software uses on screen without stopping to define them. Definitions are what the word means inside this product, which is occasionally narrower than what it means in general.
The model
| Term | What it means here |
|---|---|
| Twin | One company, modelled. Its sources, facts, knowledge graph, agents, assumption ledger, scenarios and runs. Twins are separate from each other and nothing is shared between accounts. |
| Knowledge graph | The map of things and how they relate: customers, segments, named accounts, competitors, suppliers, departments, products. Nodes read from your files are drawn solid, nodes the app had to invent to fill a gap are drawn faded and listed as inferred. |
| Agent | An actor in the simulation with its own objectives, constraints, the slice of the world it can see, a personality, relationships and history. Nine classes: customers, competitors, suppliers, salespeople, executives, employees, investors, regulators and partners. Every twin gets all nine, whether or not your files mention them. |
| Objective | What an agent is trying to achieve, which decides when it acts. A competitor taking share when an opening appears, a named account renewing at a price it can defend, an executive protecting a margin number. |
| Personality trait | Six axes that change how an agent behaves rather than what it wants: appetite for risk, willingness to move against others, weight put on an existing relationship, tolerance for a bad month before acting, how much the near term dominates, and how openly the position is stated. Two competitors with the same objective and different aggression behave differently and differently on purpose. |
| Assumption | One number the simulation needs, held in the ledger with a value, a band, an origin and, where it came from a file, the quoted line. Elasticity, monthly churn, gross margin, contract length, competitor reaction speed and so on. |
| Origin | Where an assumption came from. Read from a document, derived from something that was, estimated by a language model, set by you, or an industry default. Origins are never blended into one confident figure, and defaults sort to the top of every list because they are the ones worth replacing. |
| Elasticity | How much demand moves when price moves. Minus 1.3 means a one percent price rise loses about 1.3 percent of volume at small moves. It is the single most load bearing number in most pricing questions, which is why the sensitivity almost always puts it first. |
| Churn | The share of customers leaving each month. The model starts from your stated churn by construction and only moves it when conditions deviate from today: a price gap against competitors, a quality gap, a satisfaction change. |
| Fidelity | How closely the twin resembles the company. Not a single score. It is the combination of how much of the ledger came from your record, how many of the five key source kinds you have uploaded, and whether the named accounts and departments in the graph are the real ones. |
| Readiness | The honest version of a progress bar, from nought to a hundred. Sixty percent of it is how much of the ledger is standing on something you uploaded, forty percent is how many of the five key source kinds you have. A low readiness does not stop you running anything, it tells you how much the answer is industry prior rather than your company. |
The run
| Term | What it means here |
|---|---|
| Lever | One thing you change, with a value and the month it lands. There are twenty four of them in seven groups. A scenario is an ordered list of levers and nothing else. |
| Scenario | A named question with its levers, horizon, replication count, seed and whether agents are allowed to adapt. Fifteen ship ready made and any of them can be edited or built from scratch. |
| Baseline | The same twin doing nothing, run on the same seeds as the scenario. Every figure in a brief is a comparison against it, so the difference reported is the decision and not the noise. |
| Counterfirmual | The difference between the scenario and the baseline. The word matters because it names the only thing a simulation can honestly claim: not what will happen, but what changes if you do this rather than nothing. |
| Replication | One complete run of the year with one set of parameter draws. A standard run is 200 replications of the scenario and 200 of the baseline, which is 400 simulations. |
| Monte Carlo | Running the same model many times with different draws from the uncertainty in the inputs, then reading the distribution of outcomes rather than any single run. That is all the term means here. |
| Seed | The number that determines every random draw in a run. The same seed gives the same answer, forever, which is what makes a result something you can return to rather than something you re-perform. |
| Sweep | One lever across a range, with a full Monte Carlo at every point. Twenty five settings at eighty replications is two thousand simulations and returns a response curve with a best point on it. A second lever turns the curve into a grid. |
| Sensitivity | Moving each load bearing assumption to the ends of its own band, one at a time, re-running, and ranking by how far the answer swings. The honest version of asking how sure you are. |
The output
| Term | What it means here |
|---|---|
| Band | The spread of outcomes across replications, reported at each month. Not a confidence interval and not a claim about the real world. It is the spread the model produces when the uncertainty in your own numbers is taken seriously. |
| p10 and p90 | The tenth and ninetieth percentiles across replications. One run in ten came in below the p10 and one in ten above the p90. The median sits between them and is the number most people quote, which is why the other two are printed next to it. |
| Event frequency | The share of replications in which an event happened at all. A competitor matching your move at 88 percent is close to a certainty inside the model. A named account walking at 30 percent is a real risk with a name on it, held as a risk rather than a forecast. |
| Tripwire | A specific signal, a month to check it, an expected value and the condition that would mean the model is wrong. The earliest point at which the real world can tell you to stop, which is always cheaper than the latest. |
If a term on a screen is not in this list and is not obvious, that is a documentation fault worth reporting.
The words are easier once you have seen them do something
The worked example has a ledger with every origin in it, a brief with bands and frequencies, and tripwires you can go and check.