BOIS
Business-Operational Intelligence System
Three ways to read this page
For players: BOIS helps MaOS keep facts, risks, unknowns, actions, and consequences visible.
For developers: BOIS is a rule-governed architecture for AI products that must preserve evidence, boundaries, and responsibility.
For players: BOIS helps MaOS keep facts, risks, unknowns, actions, and consequences visible.
For developers: BOIS is a rule-governed architecture for AI products that must preserve evidence, boundaries, and responsibility.
For researchers: BOIS connects domain, value interface, substrate, SIMA, BORIS, and accountable AI under incomplete knowledge.
How BOIS appears inside MaOS
Facts
What is currently confirmed inside the game world.
Assumptions
What may be true, but has not been verified yet.
Unknowns
What remains hidden, unstable, or not yet measurable.
Actions
What the player chooses to do with limited time and resources.
Consequences
What changes after each decision.
Verification
What becomes clearer after observation, testing, or failure.
What BOIS Can Be Used For
BOIS is not limited to MaOS. MaOS is one playable demonstration of a broader decision architecture for AI-assisted work under incomplete knowledge.
BOIS was created as a practical architecture for situations where fluent AI output is not enough.
Its purpose is to keep facts, assumptions, unknowns, risks, values, actions, consequences, verification, and stop-signals separated and visible. This makes BOIS useful wherever AI must support decisions without pretending to know more than it actually knows.
BOIS was created as a practical architecture for situations where fluent AI output is not enough.
Its purpose is to keep facts, assumptions, unknowns, risks, values, actions, consequences, verification, and stop-signals separated and visible. This makes BOIS useful wherever AI must support decisions without pretending to know more than it actually knows.
AI Workflows Under Incomplete Knowledge
BOIS can structure AI-assisted workflows where the situation is uncertain, information is partial, and premature confidence would create risk.
Decision Audits
BOIS can be used to review decisions after the fact: what was known, what was assumed, what was ignored, what risks were accepted, and where verification failed.
Structured Advisory Systems
BOIS can turn an AI assistant into a more disciplined advisor: one that separates evidence from interpretation, asks for missing inputs, tracks constraints, and knows when to stop.
Business Operating Protocols
BOIS can support business operations where decisions must be repeatable, accountable, and grounded in real constraints: resources, people, cash flow, risk, procedure, and timing.
Game Engines and AI-Installed Game Systems
BOIS can be used as a governance layer for AI-run games: maintaining state, limiting arbitrary invention, tracking consequences, and preserving meaningful player agency.
Personal Decision Environments
BOIS can help individuals organize complex personal decisions by separating what is true, what is assumed, what is unknown, what matters, and what should not be acted on yet.
MaOS is the first public game-shaped demonstration of this architecture.
The game shows BOIS in motion: not as a theoretical diagram, but as a pressure system where incomplete knowledge, limited resources, human fatigue, infrastructure risk, and consequences must be handled without collapsing into random improvisation.
The game shows BOIS in motion: not as a theoretical diagram, but as a pressure system where incomplete knowledge, limited resources, human fatigue, infrastructure risk, and consequences must be handled without collapsing into random improvisation.
BOIS: Not a Prompt, but a Governing Machine for AI Products
A conventional LLM [large language model] works as a generator of probable continuation: it receives a request, visible context, system instructions, and, where available, tools or memory. The model then predicts the next token step by step and forms an answer. This approach is fast, flexible, and convenient, but it has a systemic limitation: goals, boundaries, acceptable risk, authority, and final judgment remain outside the model - with the operator.
BOIS [Business-Operational Intelligence System] changes the architecture of interaction with AI. Instead of asking the model to “produce a useful answer,” BOIS routes the request through a governed structure: domain, values, substrate, evidence, stop signals, procedures, state, and feedback.
In this architecture, an AI product does not merely answer. It checks which domain the request belongs to, who owns the decision, what counts as evidence, where the boundary of confidence lies, which risks require stopping, and which next step is actually admissible.
How a Conventional LLM Works
A conventional LLM receives a request and context, converts them into tokens, and generates a probable answer. The operator formulates the task, clarifies the context, evaluates the result, and, when necessary, corrects the direction of the dialogue.
This mode works well for quick explanations, drafts, ideas, summaries, and flexible interaction. But in complex decisions, it may hide assumptions, mix fact and inference, weakly mark uncertainty, and create an impression of completion where a verification protocol is actually needed.
BOIS [Business-Operational Intelligence System] changes the architecture of interaction with AI. Instead of asking the model to “produce a useful answer,” BOIS routes the request through a governed structure: domain, values, substrate, evidence, stop signals, procedures, state, and feedback.
In this architecture, an AI product does not merely answer. It checks which domain the request belongs to, who owns the decision, what counts as evidence, where the boundary of confidence lies, which risks require stopping, and which next step is actually admissible.
How a Conventional LLM Works
A conventional LLM receives a request and context, converts them into tokens, and generates a probable answer. The operator formulates the task, clarifies the context, evaluates the result, and, when necessary, corrects the direction of the dialogue.
This mode works well for quick explanations, drafts, ideas, summaries, and flexible interaction. But in complex decisions, it may hide assumptions, mix fact and inference, weakly mark uncertainty, and create an impression of completion where a verification protocol is actually needed.
How BOIS Works
BOIS adds a governing layer to the AI product. The request first passes through D/V/S:
SIMA reconstructs which machine is already operating: which operations repeat, where errors occur, what is treated as evidence, who makes decisions, and which hidden rules govern the system’s behavior.
BORIS turns this analysis into a working domain physiology: roles, procedures, stop signals, state schemas, rules, tests, record owners, and mechanisms for returning experience into system repair.
BOIS adds a governing layer to the AI product. The request first passes through D/V/S:
- Domain — the area in which the task operates;
- Value interface — the goals, constraints, acceptable losses, and risks set by the operator;
- Substrate — the real environment in which the action is performed: a person, team, company, file, tool, AI system, or business process.
SIMA reconstructs which machine is already operating: which operations repeat, where errors occur, what is treated as evidence, who makes decisions, and which hidden rules govern the system’s behavior.
BORIS turns this analysis into a working domain physiology: roles, procedures, stop signals, state schemas, rules, tests, record owners, and mechanisms for returning experience into system repair.
The Core Difference
A conventional LLM optimizes answer generation from request and context.
A BOIS-oriented AI product governs the generation itself: it defines applicability boundaries, separates fact from hypothesis, keeps the unknown visible, preserves operator authority, and prevents textual fluency from replacing verified closure.
The result is not merely an “AI answer,” but a verifiable transition: from the unknown to a valuable question, from the question to a protocol, from the protocol to a next step, and from experience to an updated working physiology.
BOIS is needed where AI must not only sound convincing, but operate inside an accountable system: with boundaries, risks, evidence, authority, and repairability.
A conventional LLM optimizes answer generation from request and context.
A BOIS-oriented AI product governs the generation itself: it defines applicability boundaries, separates fact from hypothesis, keeps the unknown visible, preserves operator authority, and prevents textual fluency from replacing verified closure.
The result is not merely an “AI answer,” but a verifiable transition: from the unknown to a valuable question, from the question to a protocol, from the protocol to a next step, and from experience to an updated working physiology.
BOIS is needed where AI must not only sound convincing, but operate inside an accountable system: with boundaries, risks, evidence, authority, and repairability.
Research
The project is connected to the preprint “Philosophical machines under incomplete knowledge: BOIS, SIMA, BORIS, substrate physiology, and accountable AI.”
