Enough can be less.
“East is available” can group east-only and both-open together. Both permit an eastward move. Full identification of the layout is unnecessary for this choice; one undifferentiated instruction is insufficient.
Essai 01 · Moving the whale · 2 October 2026
Can we move
the whale?
A route exists. The people making the move see different parts of the situation. What must they know to act together?
Try the first move
The small project
One whale.
Three perspectives.
A museum is rearranging a gallery before reopening. Its oversized whale exhibit must pass through an east or west passage. The survey team has checked the layout. The handling team chooses the passage. Later, the curator decides whether to reopen.
Everyone wants the same outcome, tells the truth and understands the agreed signals. Authority to move is already granted. Still, an available route or a finished move does not automatically become information the next person can use.
The kind of question this foray asks
What must a local view preserve?
Two situations can look identical to a person yet call for incompatible actions. That is where information work has a job to do.
First result: three possible layouts need only two distinguishable instructions. Knowing enough to act can be less demanding than knowing exactly which world you are in.
The working experiment
Change the information.
Keep the whale.
Choose what information work is in place. The checker groups all admitted cases by what each person observes, then looks for a decision that works throughout each group.
Start with a comparison:
01 / Choose a passage
What can the handler choose?
Exactly three layouts are admitted: east only, west only, or both passages usable. A handler must choose one usable route. Waiting forever does not satisfy this decision task.
See the local decision table
02 / Decide whether to reopen
Finished—or known to be finished?
At a later checkpoint, the move may be complete or still underway. The curator must reopen in the complete cases and hold in the underway cases. Holding every time avoids an unsafe reopening, but fails this more demanding brief.
See the curator’s decision table
03 / Look through different eyes
Same moment. Different possibilities.
The curator
The handlers
Each card is an admitted possibility; only the outlined card is selected as actual. The curator’s view never receives the modeller’s chosen-case selector.
Work derived from the declared choices
What is the smallest sufficient arrangement?
The checker tries every subset of these six available interventions. It returns all equally cheapest arrangements that meet the selected brief in every admitted case.
Smallest means least total effort in this menu, with these prices and assumptions. The survey, permission and physical move are supplied parts of the scenario.
Inspect the complete finite check
All five partitions of the three layout possibilities are checked separately. For each observation group, the intersection of acceptable actions must be nonempty.
Read the executable model · Read the testsWhat this first essai found
Information is part of the work.
A report changes a view.
A received truthful completion report lets the curator distinguish complete from underway. If a report can still be missing, silence leaves those possibilities entangled. A physically finished move can remain unrecognised.
A receipt has a different job.
With uncertain delivery, a received acknowledgement can tell the handlers that the curator knows. It does not repair a report that never arrived. With delivery already guaranteed by the checkpoint, that extra receipt is unnecessary for this model’s knowledge goal.
What did the mathematics add?
A careful table can solve this small example. The contribution here is an explicit, reusable check: keep a person’s indistinguishable cases together, intersect their acceptable actions, and return either a local rule or the conflicting cases. Enumerating repairs makes the cost of a stronger coordination brief visible.
This is a first construction for the wider enquiry—not a discovery of a new logic, a general project planner, or evidence that the tool improves real projects.
The bridge back to the foray
From sources
to a visible construction.
End: understand when project participants can jointly act from their different information, and derive useful information work where possible.
Ways: epistemic planning provides states, observations and locally informed choices. Distributed cognition asks how the arrangement of people and artefacts changes that work.
This essai: an exact finite check at two supplied decision points, with an exhaustive search through a small information-work menu. Read the public foray prompt.
Bolander · epistemic states
Possibilities linked by what a participant cannot distinguish. We implement that as equality of explicit local observations. A Gentle Introduction to Epistemic Planning, §6.
Engesser and colleagues · local choice
A participant cannot choose differently between cases their own information leaves identical. We borrow that discipline; we do not implement their full planning and perspective-shifting algorithm. Cooperative Epistemic Multi-Agent Planning, §3.3.
Moses · conditions for acting
Necessary conditions for an action can impose necessary knowledge conditions. Our finite intersection argument is checked directly; the paper’s broader theorems are not transferred wholesale to asynchronous human work. The Knowledge of Preconditions Principle, §§2–3.
Halpern & Moses · communication assumptions
What can be known depends on the protocol, timing and delivery assumptions. Our model states these explicitly and distinguishes a received message from knowledge at its sender. Knowledge and Common Knowledge in a Distributed Environment.
Hutchins · people and representations
A shared artefact can change the cognitive work required. This motivates looking beyond “send more data”; no claim about real human performance is inferred from the finite checker. How a Cockpit Remembers Its Speeds.
The boundary of the result
The layouts, observation rules, truthful reports, response checkpoints and effort costs are authored. The model chooses safe route decisions and appropriate reopening decisions; it does not simulate handling equipment, prove that the move eventually finishes, model fatigue or misunderstanding, or guarantee real-world safety. The completion channel may miss a checkpoint; route cues are assumed available beforehand.
The first gain is modest but concrete: information arrangements now produce checked local decision rules and specific counterexamples. Whether that representation pays its way in larger project work remains an open question.