Formulate
State the possible conditions, evidence, actions, consequences, and preferences you actually know.
Probability · evidence · decisions
Answers you can examine,
not just accept.
Benjamin Knowledge Models gives your AI a governed mathematical service for probability, evidence, and decisions under uncertainty. It validates typed inputs, refuses missing judgments, and shows the assumptions, warnings, and complete path to every result.
Click to sign in, create a revocable key, and start with a free allowance that resets daily. No payment method required.
- Versioned mathematical contracts
- Independently checked cases
- Deterministic and stateless
QUESTION
Which observation deserves scarce telescope time?
01State what is uncertain4 states
02Compare candidate actions3 actions
03Value observations3 options
04Compare informationnet VOI
Every result remains conditional on the supplied model.
Trace complete
Governedtyped inputs, explicit limits, and visible refusals
6reviewed and independently checked cookbook cases
MCP-nativeyour connected agent discovers, assesses, and executes
Free Exploreraccount login, revocable API keys, usage ledger, and resetting allowance
The useful division of labor
Your AI holds the conversation.
BKM holds the calculation.
An AI client can gather context and assemble a proposal. BKM checks whether that proposal is ready, performs only supported mathematics, and returns a typed result with its working shown.
Repeatable where it matters.
In our recorded repeated-run comparison, ChatGPT returned divergent answers when given the same BKM-type question and the same input. BKM does not ask you to trust a fresh narrative calculation: the same canonical proposal, capability meaning version, and numerical policy replay to the same result and trace. The comparison concerns ChatGPT without a BKM connection; BKM augments the agent rather than replacing it.
Assess
BKM identifies missing or inconsistent material and gives the next question to ask. It supplies no unstated number.
Execute
Ready analyses run in explicit decimal arithmetic and return structured results, warnings, and a replayable trace.
Judge
BKM does not admit evidence, establish that a model fits reality, or authorize an action. Those responsibilities remain visible.
Broad capability, disciplined use
Bring the problem.
BKM determines what applies.
BKM is not a menu of formulas. A connected agent describes the question and the available facts; BKM selects only mathematics covered by its versioned contracts, identifies what is missing, and refuses to manufacture an answer when the formulation is not ready.
Turn uncertainty into an explicit model.
Separate observations, assumptions, alternatives, and consequences so the basis of the answer remains visible.
Test whether the question is ready.
Surface missing values, inconsistent structure, model limitations, and the next fact needed before calculation.
Apply reviewed mathematics deterministically.
Use typed, versioned contracts and numerical policies instead of improvising arithmetic in a conversation.
Return a result that can be challenged.
Show assumptions, warnings, intermediate values, provenance, and a replayable calculation trace.
The detailed contract is delivered to authenticated agents at the point of use. The open website demonstrates behavior and evidence without publishing an inventory of internal calculation families.
Interactive, bounded, and transparent
Explore a complete decision problem.
Start with the fictional exoplanet case—or choose any reviewed cookbook. Adjust only its synthetic information-acquisition costs, assess readiness, then execute the same typed proposal when the shared playground is operator-enabled.
Preparing the playground…
Inspect the typed BKM response
The playground accepts no uploads or arbitrary requests and stores no proposal or result. Operators can disable execution independently; session, client, concurrency, and global limits protect the shared demonstration when enabled.
Grounded examples
Cookbooks, not magic tricks.
Every case keeps its typed proposal, checked BKM output, independent numerical verification, and provenance boundary together in the repository.
Install where you already work
Give your AI a calculator
it can explain.
The public BKM plugin teaches a compatible agent how to use Sidekick and connects it to the governed MCP service. The package contains no credential; invitation, OAuth, entitlement, and quota remain enforced by Sidekick.
Fastest route
Paste this into your agent
Your agent can identify its own supported installation route and stop when a browser login or approval needs you.
Install and verify BKM Sidekick for me from the authoritative public plugin distribution at https://github.com/RoyBBenjamin/bkm-plugin. First identify this exact client, version, operating system, and native plugin or marketplace mechanism. Do not guess commands or inherit a support claim from another client. The plugin contains no credential. Never ask me to paste an API key, password, authorization code, access token, or client secret. Use Sidekick OAuth and pause when I must complete browser login, consent, restart, or workspace approval. After installation, report these separately: plugin installed, Sidekick authenticated, tools discovered, contract read, and calculation verified or not run. Before any quota-bearing assessment or calculation, explain the test and ask once for confirmation. If anything fails, preserve the bounded error and stop rather than retrying in a loop or silently changing configuration.
codex plugin marketplace add RoyBBenjamin/bkm-plugin
codex plugin add bkm@bkm/plugin marketplace add RoyBBenjamin/bkm-plugin
/plugin install bkm@bkmcopilot plugin marketplace add RoyBBenjamin/bkm-plugin
copilot plugin install bkm@bkmInstall BKM Sidekick from the Plugins directory, then complete the Sidekick OAuth sign-in when prompted.
Codex CLI and ChatGPT Work have independently completed the recorded Sidekick authentication and calculation sequence. Installation in another product is reported separately until that exact surface is exercised.
Choose the route that fits your clientUse OAuth where supported, or manage a revocable Sidekick API key without placing it in source, prompts, or screenshots.
Open the plugin
Full installation guide
Manage Sidekick access
Authority boundary
“A calculation can be exact while its inputs or model remain uncertain.”
BKM does
Assess structural readiness, calculate supported analyses, and expose every assumption, warning, and mathematical step.
BKM does not
Decide whether evidence is true, establish model adequacy, retain your proposal, or authorize an action.