Ends are hard to articulate
Bounded articulation names the upstream constraint: people and organizations do not always know how to state, share, revise, and authorize what they mean well enough for a system to act on it.
The sequence
Before a system can act, it has to know what kind of authority it is carrying.
Identity lives in commitments
A useful twin has to remember more than facts. It needs a record of the conditional commitments a person can stand behind: what they will do, refuse, protect, and revisit under pressure.
Codes carry commitments
Categories, roles, metrics, prompts, routines, and protocols let commitments travel. They become dangerous when the carrier starts acting like the commitment itself.
Should stays upstream
Machines are useful downstream. They can search, retrieve, compare, draft, route, and execute. They cannot decide which end deserves authority just because it is easy to optimize.
Direction should compound
Right Direction is the practical test: each interaction should sharpen memory, constraints, and the live objective instead of restarting from scratch or letting detection substitute for judgment.
This is the bounded-articulation problem: not only limited processing after an objective exists, but the harder work of making the objective itself stateable enough to guide action. Simon's bounded rationality belongs downstream, once a value premise exists.
Ends and means
The human settles direction. The machine processes within it.
Human settlement
Standards, tradeoffs, priorities, and acceptable risk remain owned by the person or organization using the system. Exuno can surface candidates for settlement, but it cannot author the end on your behalf.
Machine processing
Once direction is settled, machines can organize sources, propose knowledge, detect repeated patterns, run protocols, and execute bounded tasks with proof.
Codes carry commitments: categories, roles, prompts, metrics, routines, interfaces, and protocols. The code may move a commitment downstream. It may not replace the act of settlement that gave the commitment authority.
Grounding
The lineage is broad because the product problem is broad.
A single source cannot carry the whole problem. Once an end is articulated, it has to travel through memory, routines, categories, protocols, and delegated execution without being replaced by the machinery that carries it.
Theory stays below the surface because users should not have to read a syllabus to trust a system. Internally, the sources become product roles: premise separation, source-grounded memory, category friction, protocol lifecycle, governance, and anomaly-driven re-settlement.
Premises and direction
Simon matters because bounded rationality begins once a value premise exists. Thompson and Hayek show why coordination signals still presuppose direction. Rasmussen gives Exuno the vertical grammar for keeping purpose, knowledge, I/O, protocol, and execution apart.
Memory, routines, and boundaries
March/Cyert explain decision premises and search. Buckland, Walsh & Ungson, and Ackerman explain why memory needs source, frame, and retrieval context. Institutional theory and Feldman/Pentland explain why routines must graduate, perform, fail, and de-grade.
Identity, boundaries, and delegation
Stone, Tilly, and Zuckerman explain why action is interpreted through categories and boundaries. Identity/action and commitment theory explain why visible behavior is ambiguous without a record of what the actor has authorized.
Strategy, objective theory, and governed execution
Transaction cost and agency theory explain why delegation needs envelopes, monitoring, and receipts. Objective strategy, Right Direction, and APD/SAGE explain why new dimensions can be surfaced by evidence, while settlement stays human.
Practical result
The argument becomes product guarantees.
Source-grounded memory
Sources stay attached to the claims they support, so memory remains inspectable instead of becoming a confident blur.
Reviewable knowledge
Machine synthesis can propose. Durable knowledge waits for provenance, review, and a visible truth state.
Protocols that can change
Stable patterns can graduate into executable protocols. Counter-evidence can de-grade them back into review.
Constrained executors
Executors act through authorized premises, bounded envelopes, and receipts that say what happened.
What this prevents
Execution never pretends to be judgment.
A generic AI layer can answer quickly while losing the reason an answer counts. Exuno keeps source, review, settlement, protocol, and execution separate enough to inspect and connected enough to work. That is the practical philosophy: purpose remains human; processing becomes machine-owned.