Decisions that pass tend to fall into a few types:
- A pricing change, a new tier, or paywalling something that was free
- A product or feature launch
- A policy or terms change, including moderation and community rules
- A brand, positioning or naming change
- A move into a new market or segment
- Removing or shutting down something people use
A decision that fails a gate can often be reframed into one that passes, and some can’t. Manic Edge says which on the first call, before there is a fee.
Why it runs more than once
AI agents are probabilistic, so two runs of the same simulation play out differently. That variation is useful. A reaction that shows up in every independent run is something the population reliably does when this change lands. A reaction that shows up once is noise until another run repeats it.
For that reason every simulation runs at least twice from the same starting point, with the decision as the only change between the simulated market and the real one. Changing two things at once would make it impossible to say which one caused what moved.
How to read the output
Every claim in a Manic Edge readout carries one of two tags. [SEED] means the claim is observed in the company’s own data: customers have already said or done this. [EDGE] means the claim emerged in simulation, and the readout says how many runs it recurred in. A fidelity audit closes the readout and lists any claim that cannot be traced to either.
The balance between the two tags depends on the seed. With two or more data sources, most of a readout can rest on observed material. With one source, more of it rests on what emerged, and the readout says so. A seed built only from a public community leans toward the people who post, and the readout says that too.
Reactions are ranked by how often they recurred and by how much they would change the decision. The readout leads with what to do about each one, with the validation material behind it.
Adding competitors
Some decisions land on competitors as well as customers. A War Game adds them to the simulation. Manic Edge profiles each named competitor from the public record, with a close look at how they responded the last time someone made a similar move, and simulates their strategy room reacting to the decision. The output is a set of counter-move scenarios ranked by how often they recurred across runs. The ranking is the deliverable, and it describes what came up in simulation. It makes no claim about what a named competitor will actually do. For how to plan around the ranking, see planning for competitor counter-moves.
What it can and cannot tell you
A simulation is good at showing how a defined group reacts to one defined change: which objections recur, which segments split, what people say to each other, and where a rollout plan has a gap. It is useful before a decision is final, while the answer can still change something.
It has limits, and the readout states them. It makes no forecast of revenue, churn, conversion or any other financial result. It reflects the data it was seeded from, so a thin or one-sided seed produces a thin or one-sided population. It models a market of customers, so decisions that land mainly on a company’s own staff need a different approach. Manic Edge delivers every readout as professional opinion and analysis, written to inform a decision before it ships.
What a readout changes
In March 2026, a regional transit authority announced a fare increase of about 5% across Metro Vancouver, taking effect on July 1. The stated reason was its ten-year investment plan, with a smaller increase expected the following year. With any price rise, the worry is a loud public backlash: petitions, protests, a story that runs for weeks.
Across every run, the simulated riders mostly absorbed the increase and carried on. Some argued it was better than service cuts. The people who felt it most cut back quietly or found ways to pay less, and the riders on student and senior passes barely engaged. The grievance that did come through, in every run, came from the riders it cost most: people on low incomes and people paying for three zones to get to work. That matches the real reaction, where the organised response was a small campaign for a low-income transit pass, built around what a three-zone pass costs over a year.
For a team about to make this call, that changes where to spend the effort. A plan for handling broad public anger would be aimed at something that didn’t happen. The decisions that matter are what the increase means for low-income riders on long commutes, and whether to have an answer ready for them before the announcement. One of the three runs got as far as riders organising around a low-income pass, which is the campaign that followed in reality. That appeared once, so the readout would have reported it as a possibility worth preparing for rather than a finding.
How the method is checked
Manic Edge checks the method by running it on past decisions whose outcomes are already public. We pick a decision with a well-documented public reaction and build the population only from material that existed before it was announced. We also disguise the names of the company, its rivals and its services, so the model can’t simply recall what happened.
Before the first run, we write down the reactions we expect to recur and lock the scoring rules, so the results can’t be graded generously after the fact. The simulation then runs at least three times, independently, and each run is scored against those rules. A reaction counts only if it recurs across runs. The answer key stays out of the simulation until grading is done.
Each backtest report records the misses alongside the matches and traces each miss to its cause. The report also records anything that could have contaminated the result, such as a real name surfacing in a run. Numbers the simulation produces along the way, like sign-up counts or percentages, are struck out and never used in grading.
A backtest checks whether the method reproduces the shape of a real reaction: who reacts, what they object to and what they do about it. It’s how we find weak spots in the method before a client relies on it.
Where to start
The free Decision Check takes about ninety seconds. It asks eight questions about the decision, the people it lands on and the data available, and says whether the decision can be simulated as it stands. It runs in the browser and stores nothing.
Manic Edge runs simulations in three tiers. A Community Run simulates customers. A War Game adds competitors. A Full Loop runs move, counter-move and market reaction over several rounds. Each has a fixed fee agreed in writing before work starts, and the details are on the market simulations page.
Take the Decision Check About market simulations
