# Knowing enough to act together

**PW-225 · higher-autonomy, agent-selected enquiry · established 2 October 2026**

## End

Understand when a physically feasible project can be carried out by participants acting from different and changing views of the situation. Develop representations that expose the difference between a successful plan visible to an all-seeing modeller and a way of working people can select and follow.

The ambitious possibility is to derive useful information work: observations, inspections, exchanges, confirmations or changes in how work and information are distributed. Some communication may prove unnecessary. The eventual aim is useful intuition pumps for thoughtful project people, with genuine constructive results where possible.

A revealing impossibility, counterexample or reason to abandon a method can be a successful essai: a trial, not a guaranteed finished answer.

## Ways

Initially combine epistemic planning with distributed cognition. The first makes available distinctions and local choices explicit; the second keeps attention on people, artefacts and arrangements doing work together.

Preserve the differences between truth, observation, knowledge, belief, authority, choice and completion. A report being sent, delivered, understood or acknowledged are separate events unless an explicit toy assumption combines them. Knowledge is factive in the initial models; false beliefs require different semantics. Never assume everyone must know everything.

On each essai, reopen the sources needed for its claims and record the actual definitions and limitations used. Compare the proposed construction with a simpler account. Methods may be revised or rejected for a reason; a citation index is not a substitute for reading.

| Source | Role and boundary |
|---|---|
| [Bolander (2017), A Gentle Introduction to Epistemic Planning](https://arxiv.org/abs/1703.02192) | States, events, updates and observation-based policies; also computational and representational limits. |
| [Engesser et al. (2017), Cooperative Epistemic Multi-Agent Planning for Implicit Coordination](https://arxiv.org/abs/1703.02196) | Action ownership, perspective shifts and local policies. Arbitrary independent replanning need not converge. Formal participants need not be AI agents. |
| [Moses (2016), The Knowledge of Preconditions Principle](https://arxiv.org/abs/1606.07525) | Necessary conditions for knowledge under stated action/system assumptions; not a sufficiency theorem for project completion. |
| [Halpern & Moses (1990), Knowledge and Common Knowledge in a Distributed Environment](https://groups.csail.mit.edu/tds/papers/Halpern/JACM90.pdf) | Communication and timing assumptions determine coordination claims. Common knowledge is not a universal teamwork requirement. |
| [Bolander, Burigana & Montali, Depth-Bounded Epistemic Planning (2025 v2)](https://arxiv.org/abs/2406.01139v2) | Optional bounded search method for sequential plans; not automatically a distributed-policy algorithm. |
| [Belardinelli & Bolander (2023), Attention!, corrected v2](https://arxiv.org/abs/2303.13494v2) | Proposition-specific attention and false belief challenge automatic observation. Not an empirical predictor of human attention. |
| [Hutchins (1995), How a Cockpit Remembers Its Speeds](https://pages.ucsd.edu/~c8johnson/COGS102B/Hutchins95.pdf) | People and representations jointly perform cognitive work; situated evidence rather than a planning proof. |
| Dorian Taylor (2014, updated 2019), Toward a Theory of Design as Computation | Cognitive artefacts and material arrangements as a conceptual bridge; not a theorem. |
| [Wynn & Clarkson (2021), ASM2.0](https://doi.org/10.1007/s00163-020-00354-5) | Practical process-awareness and process-library limits, particularly §§6.2.3–6.3; not epistemic results. |

## Means

The provisional intuition pump is an oversized museum whale moved through one of two passages, with survey team, handlers and curator holding different information. It is a controlled fictional scene, not the eventual use case or a constraint on future enquiries.

Use small executable models, exhaustive finite checks, worked counterexamples and primary papers. State the possible situations, local observations, action owners, communication assumptions and success criterion. Keep safe choice separate from eventual completion, and a bounded failed search separate from a general impossibility.

Make the formal objects visible. Show the actual situation, the alternatives each participant still considers possible, an observation that changes them and the resulting choices. Preserve a source → concept → construction → result → limitation route in the normal reader entrance.

The agent may choose the next consequential bounded question during authorised research sessions. Do not treat this initial scenario, source set or first result as a fixed roadmap. Explain revisions and correct mistaken interpretations plainly. Prefer one illuminating result to a large demonstrator with shallow semantics.

Keep research findings, implementation checks and demonstrated human usefulness distinct. [Essai 01](ESSAI.md) is the first bounded construction; it does not settle the wider End.
