Counterfirm
Next year gets planned out of opinions in a room. The numbers in the deck are somebody's guess with a confident voice on top, and when the year goes another way nobody can say which assumption was the wrong one, or where the thing would have broken.
The constraintNo API key and no network call in the engine. The same seed gives the same answer every time, and every figure traces back to the assumption and the file it came out of.
Founder, model architect, designer and engineer. The extraction, the graph, the agents, the engine, the ledger, the brief, the interface and the public site.
- 400 simulations in a run
- 15 kinds of agent
- 57 levers to pull
- 0.1 seconds for a year
- 01Five steps: read the files, pull out the facts, build the graph, generate the agents, run the year
- 02Fifteen kinds of agent, twelve of them on every twin
- 03Customers, competitors, suppliers, salespeople, executives
- 04Employees, investors, regulators and partners, each with objectives
- 05An assumption ledger, every number traced to the file it came from
- 06Nine stages a month, from the levers landing to the watchers checking
- 07400 runs a standard scenario: 200 of it, 200 of the baseline
- 0820 metrics a month, at the median and the tenth and the ninetieth
- 09A brief in six blocks: answer, timing, cause, risk, watch, assumptions
- 10Sweeps across 57 levers, tripwires, and the breaking point
A digital twin of your company. Test the decision before you make it.
Five steps, from a folder of files to an answer you can defend. Upload what you already have, a profit and loss and a customer list is enough, and it reads the files, pulls out the facts, builds a graph of the company and generates the agents. Fifteen kinds of them, twelve on every twin whether or not your files mention them: customers, competitors, suppliers, salespeople, executives, employees, investors, regulators and partners, each with its own objectives. Every assumption it used goes into a ledger you can open and argue with.
Then it runs the year. Nine stages a month, in the same order every month: the levers land, management reacts, competitors resolve the moves they queued, suppliers drift, customers churn and new logos arrive, sales capacity ramps, the books close, morale and attrition move, and the watchers check their thresholds. A standard run is 400 simulations, two hundred of the scenario and two hundred of the baseline, and it takes about a tenth of a second. A sweep runs the same question across 57 levers to find where the thing actually breaks.
What comes back is a range rather than a number, and a brief in six blocks: the answer, when it happens, what caused it, what could go wrong, what to watch for, and every assumption underneath. No API key, no network call, and the same seed gives the same answer every time. Free to start, with one twin, three files and ten runs a month.