The decision register
Every decision recorded from its evidence, signed, frozen, and scored once the outcome is in. One calibration figure says whether to trust the next band.
The registerA digital twin of your company
Test the decision before you make it
Upload your numbers. Counterfirm builds a working model of your company and runs your decision through it hundreds of times. You get a range instead of a guess, and a source for every figure in it.
Free to start. No card. Your files stay private to your account, and the simulation never sends them to a language model.
A spreadsheet gives you one version of next year. Counterfirm gives you the range, and tells you who moved it.
A price rise rarely lands the way the spreadsheet said. The spreadsheet has no competitor who matches you in month four. No salesperson who starts discounting when they are behind on quota. No customer whose contract ends the month your letter arrives. Counterfirm has all three, and it names the one that cost you the most.
Five steps
From a folder of files to an answer you can defend
01
Upload what you already have
Financial statements, a customer list, contracts, pricing, notes on competitors, whatever you have. CSV, Excel, Word, text and markdown are read directly. A PDF works if it has a text layer. If it does not, the app says so instead of guessing.
02
It builds a map of your company
Accounts, suppliers, departments and competitors become objects with sizes and connections. Every number that can be tied back to a line in one of your files is tied to it. Anything your files never mention is still created, and clearly marked as an estimate.
03
Everyone in it makes their own decisions
Fifteen kinds of agent. Each one has goals, limits, a view of the world and a speed it reacts at. Price shoppers leave early. Long standing accounts argue and stay. A reseller quietly stops selling you long before it asks for a better deal. They behave differently because they are different.
04
Then the year runs, hundreds of times
Every run picks slightly different numbers from the uncertainty in your own record. You get a middle case with a good case and a bad case around it, a list of what happened and how often, and a brief that names what it assumed.
05
And every number shows its working
Press Why on any figure. You get the calculation behind it, the assumptions it used, and where each one came from. If it was read from your file, you see the sentence. If it was an industry default, it says so and names the document that would replace it.
What happens to a file
The ecosystem
Fifteen kinds of agent, and twelve of them are always there
Customers, competitors, suppliers, salespeople, executives, employees, investors, regulators, partners, distributors, the labour market and analysts. Every twin gets all twelve. Lenders, landlords and carriers only appear if your files say they exist. Inventing a rent would count a cost your operating figures already include.
This matters. A model built only from what you wrote down will agree with you, and it will tell you a price rise costs nothing. So anything added to fill a gap is drawn faded on the map and listed as a default in the ledger. You can always see which is which.
- Customers split into segments by how they behave, plus your named accounts one at a time
- Competitors match part of your move, late, and only once
- Executives act when their own numbers cross a line, and that reaction is often the whole answer
- Regulators stay quiet until a large rise by a company with real share wakes them up
What is in the box
Enough to answer the question and then defend the answer
Scenarios
Fifteen ready made, or build your own from fifty seven levers. Price, headcount, a department, a supplier, a launch, a new market, an acquisition, a layoff, a shock. Every scenario runs against the same company doing nothing, using the same random seeds, so what you see is the decision and not the noise.
Sweeps
One lever across a range, fully simulated at every point. Twenty five settings at eighty runs each is two thousand simulations. It comes back as a curve with the best setting marked.
Side by side
Six scenarios, one baseline, one set of seeds, one table.
The assumption ledger
Every number, where it came from, and how wide the band is. Edit any of it.
Sensitivity
Which assumption is moving the answer. Measured, not asserted.
Tripwires
The earliest month the real world can tell you the model was wrong.
The knowledge graph
Who depends on whom, and how much of you sits with one account.
And a network trained on it
A small neural network learns your company from the engine's own answers. It then ranks everything on the map by how fragile it is and what it costs if it breaks. Open any score and you see the inputs behind it, each one traced back to a line in one of your files.
Ask
Any question about the company, worked through the whole record in a moment, with every step and every source shown.
A readiness score
Where the model is thin, for each kind of question you might ask, and which file to send next.
And a brief you could put in front of a board
The answer, how it played out month by month, which agent caused it, the bad case, the early warning signs, and every assumption it rests on. Defaults come first, because those are the ones worth replacing.
Ask anything
A question about the company, answered with the working shown
Type the question the way you would put it to your finance lead. Ask finds what it is about, a concept like concentration or runway or a customer by name, and works it through the ledger, the graph, every line of every file and the industry profile, in a fraction of a second.
The answer names its figures and marks each one with where it came from. Under it: the steps it took, the arithmetic written down, how sure it is and why, the file that would make it surer, and the next question worth asking. A question that is a decision goes to Analysis with one click, to be run.
- Thirty three concepts and every name on your graph
- Every figure one click from the row, the line and the file
- Confidence computed from where the numbers came from
- No language model needed, and none given anything but the working
Then govern it
What happens after the answer
The forecast against what happened
The band pinned, each month laid on it, drift called after two misses on one side, and the ledger row behind it named with the value the actuals imply.
MonitoringFourteen shocks, one resilience score
Each run against the same company doing nothing on the same seeds. The weakest point named, and every shock openable as a scenario.
Stress testsRoles, snapshots, every change
Four roles checked on every route. The twin frozen at every approval. A diff between any two moments. The page an auditor opens.
The recordThe model tells you before you ask
Eight rules asked every few hours: drift, a decision past review, a short runway, a resilience score in the red. Yes opens an alert once; no closes it.
AlertsA second signature on every number
A change to the ledger is written up with the reason and what it does to the next twelve months, and a reviewer signs it or refuses it. The proposer cannot sign twice.
Four eyesThe answer as a deck
Any analysis or run becomes a presentation in one click: the question, the answer, the numbers with their bands, what it stands on, the case against, the sources. Arrows move, Escape leaves. Nothing is computed for it.
The analysis layerKeys in, hooks out
Your finance system posts its export and its actuals with a key. The twin calls your systems, signed, when something happens.
IntegrationsThen the book
More than one company? One book, in the order the day should go
01
Every company you hold, added up
Run rate, cash, burn, people and customers, summed from the ledgers. Nothing on the book is a new number, so the book and the twin never disagree.
The portfolio02
Who needs a person today, ranked
Every reason a company has is weighted and added: drifting actuals, a decision past review, a runway shorter than a hiring cycle, a model standing on defaults. The sum is the rank.
Alerts03
One view of the world, in every model
The interest rate, wage drift, the growth of the market: kept once as the house view, applied to any ledger with a snapshot taken first and the origin written on the row.
Four eyesA sweep
What a pricing decision looks like
The part most tools skip
It tells you what it is guessing
Every number the simulation uses sits in one table, next to where it came from. Read from your document, with the line quoted. Worked out from something that was. Estimated. Set by you. Or an industry default.
Defaults are listed first. A guess sitting underneath a decision worth six figures is the most useful thing this software can point at. Replace one and the range narrows. Nothing is quietly averaged into a single confident number.
The engine
Fast enough to ask the next question
400
Simulations in a standard run
Two hundred runs of your decision and two hundred of doing nothing, on the same seeds
0.1s
Typical time for that run
Measured over twelve months with forty agents, on the machine this was built on
57
Levers to pull
Price, people, supply, customers, growth, cash, terms, market
15
Kinds of agent
Twelve of them on every twin
The questions
Start with the one you keep putting off
Each one ships as a scenario with its levers already set. Each has a page explaining what the model does with it and where it is weakest.
Raise price 20 percent
The extra margin lands in month one. Customers take about a year to answer. Competitors answer somewhere in between.
What the model does with itLose the largest customer
Concentration is just a number until the month it becomes an event. This is that month, including the reference they were giving your salespeople.
What the model does with itEliminate a department
The saving is easy to work out. What that department was quietly holding up shows up around month three.
What the model does with itSupplier raises prices 15 percent
First your margin. Then whether you pass it on. Then how customers answer that.
What the model does with itHire ten salespeople
Two quarters of cash going out before the first one hits quota, plus the ramp most plans forget.
What the model does with itAcquire a competitor
You buy the revenue and inherit the company. The second part is where the money goes.
What the model does with itCompared
What the four usual options give you
| A spreadsheet | Asking a model | A consultant | Counterfirm | |
|---|---|---|---|---|
| Gives a number quickly | yes | yes | no | yes |
| Shows a range rather than one path | no | no | sometimes | yes |
| Has competitors who react | no | in prose | yes | yes |
| Says which assumption is doing the work | no | no | sometimes | yes |
| Points at the line in the file a number came from | no | no | yes | yes |
| Same question, same answer tomorrow | yes | no | yes | yes |
| You can ask the next question yourself | yes | yes | no | yes |
| Carries an explicit model of your industry you can inspect | no | no | no | yes |
| Knows your industry better than you do | no | no | sometimes | yes |
| Can be wrong in a way that is hard to notice | yes | yes | yes | yes |
Two rows need explaining. The industry row does not claim the software has read more about your market than you have. It means it carries a written down model of how your kind of business works: how money moves through it, which levers matter, the three or four ways companies like yours get into trouble and when each one first shows, a seasonality shape, and typical ranges for nineteen figures. All of it is stated as an assumption and all of it is editable. Most people cannot write that down on demand, and you cannot inspect what a consultant keeps in their head. The last row applies to us too. A confident simulation is a convincing way to be wrong, which is the reason for the ledger, the ranges and the early warning signs.
Every number, taken apart
It will tell you exactly why it thinks so
The larger version
Where this is going
A whole organisation in one simulation. Employees, departments, customers, vendors, financials, contracts and workflows, all in one model you can question rather than a dashboard you can only read.
Most of the machinery for that is already built: the map, the agents, the multi agent simulation, the report. What is missing is depth in the operating model, not a different design. The status page lists exactly what is built and what is not.
Security and privacy
Built for the most sensitive files you have
A twin is your accounts, your customer list and your contracts in one place, so it is kept like it. Files go into a private store that belongs to your account, under random names, behind a rule that refuses every request from the web. Every request checks that the twin is yours before it reads a byte.
Nothing you upload trains anything, nothing is pooled with anybody else, and nothing leaves the application unless you switch it on yourself. Delete a file and it is gone, with everything made from it. On Enterprise, the whole thing can run on your own server so it never leaves your building at all.
- A private store per account, never a public folder
- Four roles, checked on every request
- Everything on the record, with who did it
- Nothing trains on your data, ever
- Encrypted between you and us
Coverage
Every answer says how much of it came from your own files, weighting each number by how far it can move the conclusion.
A default that cannot move the answer barely counts. A default under the largest lever counts for a lot, and the answer names the file that would replace it.
0 to 100%
on every answer
The questions people ask before uploading anything
Is this just a language model with a chart on it?
No. The simulation is arithmetic. Demand responds to price, competitors respond to you, capacity is people, people cost money, money is cash. Same seed, same answer, every time, and it runs with no API key at all.
A language model is used in three optional places: reading prose the rules could not parse, rewriting a finished brief into better English, and answering questions in Ask. No number from it reaches the simulation without going through the ledger first, labelled as an estimate.
What if I do not have good data?
Then it says so. It falls back on industry defaults, and the ledger shows how much of the answer rests on them. The readiness score on the overview is that number, and missing files are listed in the order worth fixing.
One profit and loss statement and a customer list is enough to be useful. With nothing at all, you get a tutorial.
How accurate is it?
It is only as accurate as its assumptions, so every one of them is listed with a range around it. It is good at shape, rough size, timing, and which actor causes the damage. It cannot tell you what your market will do next year.
There is a full page on this, and you can read it before paying for anything.
Where do my files go?
Into a private store that belongs to your account, outside anything the web can reach, over an encrypted connection. They are split into text chunks so numbers can be cited. Nothing you upload trains anything, and nobody else's account can see it. Delete the source and both the file and the chunks go with it.
Can I run this on my own server?
Yes, on Enterprise. The whole application can run as a private deployment on a server you control, so your files never leave your own infrastructure. Talk to us about it on the enterprise page.
Is there a free tier?
Yes. One twin, three files, ten runs a month. No card, no expiry. Everything that makes the output honest is included: the full ledger, every source, every range, every Why chain. What you do not get is scale or the neural network.
Bring one decision and one file
The free tier takes real files and gives real answers, so you can see what this says about your own company before paying anything. Or open the worked example: an invented contract manufacturer with one customer far too large and a price it has not moved in three years, with every question on this page already set up.