Map your reasoning
Lay out hypotheses, experiments, evidence, and outcomes as nodes. Connect them with typed edges — supports, weakens, tests, falsifies, contradicts, depends-on, conditional-on.
Bayesian Probability Dashboard
Solution to your painful problem.
You bring facts, BayesBoard does math for you. No more gut decisions. No more missing out in money, business or life.
Guest mode runs entirely in your browser. Free accounts include 3 cloud-saved boards — see pricing.
One board
every fact in one view
Together
your team and your AI
Explained
math you can read
How it works
Your reasoning, as a living board.
Spreadsheets calculate. Whiteboards communicate. BayesBoard does both — and shows you exactly why your confidence sits where it does.
Update transparently
Set priors and likelihood judgments. The engine combines them in log-odds space and writes an explanation record for every node: prior, each edge’s contribution, posterior.
Stress-test it
Mute any piece of evidence and see the posterior delta instantly. If one analyst memo is doing all the work in your argument, you will know.
What ships today
Everything on this page is in the product now.
No roadmap slides. The editor, the engine, the explanation records, and the demo board are all live — including dark and light mode throughout.
Transparent Bayesian engine
Likelihood ratios combined in log-odds space. No black box — the math behind every number is inspectable.
Explained updates
Every node carries an explanation record for its current confidence — which evidence moved it, and by how much. Full revision history is on the roadmap.
Mute evidence
Switch any evidence off to run a one-click sensitivity check and see exactly how much it was moving the result.
JSON import & export
Boards are portable files you own. Export them, version them, bring them back.
Guest mode
The full editor runs in your browser at /editor — no account needed. Guest boards stay on your machine.
Cloud-saved boards
Create an email/password account to save boards to your account and pick them up from anywhere.
Structured subjective Bayes
It makes your assumptions explicit. It doesn't pretend to replace them.
The probabilities on a BayesBoard are yours. The tool's job is to keep them explicit, consistent, auditable, and explorable — not to tell you what to believe.
What BayesBoard does
- Makes your assumptions explicit — priors, likelihoods, and edge types are all written down.
- Keeps your updates consistent — the same math is applied the same way, every time.
- Makes your reasoning auditable — every posterior traces back to the inputs that produced it.
- Makes your beliefs explorable — mute evidence, adjust weights, and see what changes.
What it deliberately doesn't
- No causal inference — edges encode your judgments, not discovered causation.
- No literature mining or AI interpretation — you supply and weigh the evidence.
- No predictions — posteriors are your subjective probabilities, made consistent.
- No collaboration or board sharing yet — BayesBoard is single-player today.
BayesBoard is a decision-support tool. Its outputs reflect your inputs and weights — they are not professional advice.
Flagship demo
See it on a real decision: EU Expansion Decision.
The demo board maps a market-entry call — a core hypothesis, conflicting evidence, an experiment gate, and the outcome it all feeds. Open it in the editor, mute the expert forecast, and watch the posterior move.
Pricing
Free to start.
- Free — 3 boards, full engine, JSON import/export.
- Pro — $9/month or $90/year for more boards and storage.
Put your next big decision on a board.
Questions
Frequently asked, plainly answered.
What BayesBoard is for, what the Bayesian part actually does, and where it sits next to the tools you already use.
What people use it for
Yes. Put the bet as a hypothesis (for example, "Entering the EU market in 2027 pays back within two years"), give it a starting probability, and attach the market research, pilot results and expert views as evidence. The flagship demo board is exactly this kind of decision.
It works well for structuring them. Make "This candidate will succeed in the role" the hypothesis and add each interview, work sample and reference as evidence with its own weight. The board shows whether one strong interview is carrying the whole case. The judgments remain yours, and so does the responsibility for a fair process.
Yes. Map a claim, the experiments that test it and the results that come back, with typed edges such as supports, weakens, tests and falsifies. Each update keeps an explanation record, so you can show a reader how your confidence got where it is.
Model the launch as one or more hypotheses (demand exists, the price holds, the channel converts) and attach beta feedback, sign-up numbers and competitor moves as evidence. Experiment nodes let you mark what still needs testing before the go/no-go call.
No. Moving cities, changing careers or choosing a school are decisions with uncertain evidence too. Guest mode runs entirely in your browser, so a personal board never has to leave your machine.
What "Bayesian" means here
Your probability that a hypothesis is true before you look at the evidence on the board. It is a starting point you write down, not a fact the tool supplies. Two people can start from different priors, and the board makes that difference visible instead of hiding it.
A way to say how strongly one piece of evidence points: how many times more likely you would be to see it if the hypothesis were true than if it were false. A likelihood ratio of 3 means three times more likely if true. A ratio of 1 means the evidence tells you nothing.
BayesBoard converts your prior to log-odds, adds one term for each piece of evidence, and converts back to a probability. For example, a 30% prior with one supporting likelihood ratio of 3 becomes 56%, not 90%: likelihood ratios multiply odds, not probabilities. Every node keeps a record of each contribution.
In log-odds, independent evidence is combined by addition, and addition does not care about order. Whether the pilot results arrived before or after the expert forecast, the board lands on the same posterior. (The optional fixed-delta adjustments, which nudge a probability directly, are the one documented exception.)
How much the conclusion depends on a single input. Mute any piece of evidence and BayesBoard recomputes the board without it, showing the change in every posterior. If silencing one memo moves your answer from 80% to 40%, that memo is doing a lot of work, and deserves a second look.
Yes. Like most practical Bayesian tools, it combines evidence as if each piece were independent given the hypothesis. If two items really report the same underlying fact, count it once or lower their weight; otherwise the board will be more confident than it should be.
How it compares
A spreadsheet can hold the same numbers, but the reasoning lives in cell formulas that few people read. BayesBoard keeps the structure visible (what supports what), applies the same update rule every time, and records why each number is what it is. You can still export the board as JSON.
A weighted matrix scores options against criteria and adds the scores up. It answers "which option looks best on my criteria?" BayesBoard answers "how likely is this to be true, given the evidence?" and updates that answer as evidence arrives. The two work well together: use the board to estimate the uncertain inputs a matrix needs.
Whiteboards are excellent for laying ideas out, and BayesBoard has drawing tools for that too. The difference is that the boxes on a BayesBoard have probabilities and the arrows have meaning, so moving a piece of evidence changes the conclusion, and you can see by how much.
Full Bayesian-network tools model joint probability tables and are built for specialists. BayesBoard deliberately does less: one likelihood ratio per edge, combined under an independence assumption, with every step explained. That makes it faster to build and easier to read, at the cost of the dependencies a full network can capture.
Your gut is a prior. BayesBoard asks it to show its work. Keep the instinct as your starting probability, then let the evidence move it in the open, where you and others can see whether it moved for good reasons.