POST-HIERARCHY ATLAS · 002

Who Will Decide? How a Group of Brainstormers Forms

From application registration to temporary collective intelligence

? Dotted terms can be clicked to open a short explanation.

The first article in the Post-Hierarchy Atlas followed a cultural project from an initial idea through collective analysis, funding and dissemination. But such a process can work only if a sufficiently competent group forms around a specific application. It cannot be a committee appointed in advance, yet it cannot be an unfiltered crowd either. How does such a group come into being?

A conventional system begins with the question: who is authorized to evaluate? It compiles a list of experts, appoints a committee and gives that body a decision mandate.

A post-hierarchical system starts elsewhere: what configuration of competence does this particular application require, and how can that configuration be assembled without turning temporary expertise into permanent authority?

In this article, a full participant in collective decision-making is called a brainstormer.

Post-hierarchy does not need to know in advance who the “best experts” are. It needs a system that can assemble the right configuration of people and specializations for each new problem from the history of reliable contribution.

The whole process — from application to cluster closure

1. An application is registeredIt becomes the object of a specific temporary cluster.
2. AI makes a preliminary estimate of likely specializationsBased on the application’s core subject and important peripheral issues, AI creates only an indicative list and uses it to guide invitations.
3. Interested participants registerThey may be invited by the system or join because they believe their knowledge can contribute to the application.
4. Some enter directly as brainstormersIf a participant already has sufficient weight in a specialization that AI considers relevant, the history of reliable contribution may be enough for direct entry.
5. Others go through the applicant filterAI-generated questions test whether a person understands the application’s subject matter and its important peripheral dimensions well enough to participate meaningfully.
6. The applicant becomes a candidateAfter reaching a sufficient result, the person evaluates real ideas in shadow mode: the system records the evaluations but does not yet include them in the collective result.
7. The cluster starts when a critical cumulative mass of specializations is presentOne or several decision projects may begin in parallel if each has enough reliable cognitive capacity.
8. New participants can join and new projects can emerge during the workThe cluster structure is not frozen at launch.
9. As a project approaches completion, its actual specializations are refinedParticipants propose which specializations best describe the work that was actually done and collectively evaluate their importance.
10. The cluster closes with clear resolutionsProjects are completed, unresolved questions may be transferred elsewhere, next steps for the application are defined, and the temporary group disperses without revealing its full membership.

1. One application — one temporary cluster

A cluster is not a standing community of “experts in photography,” “cultural economics,” “technology,” or any other fixed field. Its object is one concrete application.

An application may appear simple at first, yet analysis can reveal the need to examine technological feasibility, audiences, budget, artistic logic, risk, legal constraints, or issues that the author of the application did not initially formulate at all.

For that reason, several separate decision projects — brainstorms may operate within the same cluster, either sequentially or at the same time.

The architecture of the cluster is therefore not a finished plan drawn in advance. It develops together with the analysis.

2. Before work begins, AI can provide only orientation

Once an application is registered, AI first analyzes its thematic core and important peripheral dimensions. A cultural project, for example, may require not only artistic understanding but also knowledge of technological feasibility, implementation cost, audiences, intellectual property, logistics or risk.

On that basis AI produces preliminary indicators of likely specializations.

These indicators help the system search for people. It can invite participants who have already accumulated relevant specialization weights in previous clusters.

But an AI forecast is not a final classification. The real cognitive content of a project may turn out to differ from what could be inferred from the application text.

At the beginning, AI helps locate people who are likely to be useful. The collective process later reveals which specializations were actually important.

3. People do not register as “experts in general”

A participant does not carry a single universal expert score. Instead, they have many different specialization weights.

In one earlier cluster, a person may have strengthened a specialization in documentary photography; in another, visual anthropology; in a third, project risk analysis. A new application cluster can draw on whichever parts of that history are relevant.

People enter a new cluster in two main ways. The system may invite them because of preliminary specialization signals, or they may register on their own because they believe they have useful knowledge.

If their existing weight in a relevant specialization is sufficient, they may enter directly as brainstormers. If the system does not yet have enough evidence, they begin as applicants.

4. The applicant test asks whether the person knows what this application requires

The purpose of applicant testing is not to discover as many things as possible that a person does not know. The central question is simpler:

Does the applicant understand the problem field of this application well enough to participate fully in its cluster?

AI generates questions from the core subject of the application and from relevant peripheral areas. These may include implementation cost, technological feasibility, risk, organizational constraints, audiences or other aspects without which the discussion would remain superficial.

The questions need not be conventional multiple-choice tests with exactly one obviously correct answer. Possible responses may include:

• “None of the answers is correct”;
• “All of the answers are correct”;
• “I don’t know”;
• several answers whose degree of correctness must be evaluated.

AI may also pose a question whose premise is itself problematic or false. This tests more than factual memory: it tests whether the participant understands the structure of the question, detects false assumptions, distinguishes fully correct claims from partially correct ones and knows when an answer is not justified.

The correctness of particular answers can also be rated on a scale from −5 to +5. Here the scale has a different function than in idea evaluation: it expresses how false or how correct the applicant considers a particular answer to be.

Recognizing one’s own uncertainty is useful, but it is not the center of the test. The main issue remains: does the applicant’s mental model cover what must be understood for meaningful participation in the analysis of this application?

5. A candidate already works in the real process, but does not yet affect the result

An applicant who reaches a sufficient test result becomes a candidate.

The candidate now evaluates real ideas submitted by other participants. However, these evaluations are shadow evaluations: the system records them, but they do not yet affect the collective result.

Only after a sufficient number of shared cases does the candidate’s evaluation profile become visible.

That profile does not have to match the group average or copy another person. But it should show a reliable mental-model relationship with at least one brainstormer who already has strong weight in a relevant specialization.

Once reliability is sufficient, the candidate becomes a brainstormer. From that point on, their evaluations enter the collective result and they may submit their own ideas.

6. Ideas are evaluated from −5 to +5. But 0 is the worst score

The scale from −5 to +5 is used to evaluate ideas, not an entire project or application.

Positive scores indicate agreement with the idea; negative scores indicate disagreement. Extreme scores of −5 and +5 remain possible, but should not be used without a strong reason. In real decision environments, ideas are rarely absolutely hopeless or absolutely brilliant.

Yet the mathematical midpoint is not the semantic midpoint.

0 is the worst score an idea can receive. It does not express neutrality. It means that the idea should not continue consuming collective attention as a normally rankable alternative.

A zero score is used at least in the following cases:

• the idea is a duplicate — to avoid wasting participants’ time on repeated content;
• one submission contains several claims that should be evaluated differently — because they cannot be meaningfully compressed into a single score;
• the idea is banal — it adds no meaningful value to the decision process;
• the idea is off-topic for the specific project;
• the idea is written so unclearly that its meaning cannot be evaluated reliably;
• the text reveals the identity of its author — which would compromise anonymous evaluation and open the door to status or personal bias.

7. A cluster does not start merely because many people registered

After registration, the system must do more than gather participants. It must establish whether enough cognitive capacity has actually accumulated.

One or several initial brainstorms may begin once they reach a critical cumulative mass of relevant specialization weights.

This is not a headcount. Five people whose specialization profiles closely match the problem may create a stronger starting group than twenty people whose knowledge is only loosely related.

The launch of the cluster is also not the closing of entry. New participants may still join while the process is underway if their specializations become relevant or if they themselves recognize that they can contribute.

8. The real specializations of a project usually become clear near the end

At the start of a project, AI can only make a provisional estimate of which specializations will matter. What the project actually examined often becomes visible only after many ideas, evaluations and new framings of the problem.

For that reason, the specializations of a specific brainstorming project should almost never be fixed conclusively in advance.

During the project, any participant may propose a specialization if they believe it accurately describes the field actually being investigated. Such proposals are usually most meaningful near the end, when the group can see more clearly what kind of cognitive work has really been done.

Other participants evaluate the proposed specialization. The higher its collective evaluation, the more important that specialization is considered to have been for the project.

A specialization is not an administrative label attached to a problem before analysis. It is one of the results of the analysis itself.

9. Specialization weight grows with contribution — but it must saturate

As a project approaches completion, the system can connect two things: which specializations actually mattered and how much each participant contributed to the final result.

A participant’s weight in the relevant specialization then increases in proportion to their real contribution, also taking into account how important that specialization proved to be for the project.

But this growth cannot remain linear. If every successful contribution added the same amount of weight forever, long-term participants would accumulate an overwhelming historical advantage. Their evaluations would dominate not necessarily because they are best suited to the new problem, but simply because they entered the system earlier.

Exponential growth would be even more dangerous. It would rapidly create a small class of nearly unbeatable “super-experts.” The system might still look formally non-hierarchical while functionally becoming a new oligarchy.

For that reason, specialization weight must grow with saturation: early reliable contributions can increase weight fairly quickly, but as the participant becomes stronger, each additional success adds less.

Saturation is necessary because it:

• prevents early participation from becoming a permanent privilege;
• allows strong newcomers a realistic chance to catch up;
• prevents even a highly reliable individual from outweighing the whole group;
• keeps collective wisdom genuinely collective instead of turning it into the weighted judgment of a few veterans;
• rewards continuously demonstrated competence rather than historical reputation alone.

Prolonged inactivity may also gradually reduce specialization weight. The higher the weight a person wants to retain, the more continued reliable participation is required to maintain it.

Experience should matter. But in a post-hierarchical system, no amount of experience should become a permanent right to decide for everyone else.

10. The author may be one of the cluster’s most valuable sources of knowledge

The author of an application often knows things that outside analysts do not: why a particular format was chosen, which alternatives have already been rejected, where hidden risks lie, or which apparently minor detail actually determines the logic of the entire proposal.

If the author is excluded simply because they have an interest in the outcome, the collective system loses some of its most valuable information.

In an anonymous and accountable environment, the author can do more than evaluate other people’s ideas. More importantly, they can propose new brainstorming projects that expose an aspect of the application others have not yet noticed, and they can anonymously submit ideas in projects already under way.

Other participants do not know that a particular idea came from the author. The author’s name, status or interest therefore gives the proposal no privilege. If the wording itself reveals the author, the idea should receive a zero score and be removed from ordinary ranking.

The value of author participation is not a stronger vote. It is the ability to place into the shared field — anonymously — something the rest of the group may not yet know needs to be considered.

New projects can emerge, run in parallel, or leave the cluster

During the decision process, participants may propose new projects. AI may also detect that the pattern of existing ideas and evaluations has produced a separate question worth examining independently.

If enough cumulative specialization mass is available, a new brainstorm can begin within the same cluster while other projects are still running.

Sometimes, however, it becomes clear that the question would be handled better by a substantially different group. Participants or AI can propose moving that project outside the cluster. If the proposal is accepted, the brainstorm on that issue closes in the current cluster and the question is transferred elsewhere. A new cluster may be created and a different group assembled around it.

This means a problem is not trapped inside the administrative container in which it first appeared. It can move to wherever a better configuration of specializations can be formed.

When does the cluster end?

The purpose of a cluster is not to exist. Its purpose is to analyze one application and produce clear resolutions about what should happen next.

By the end, it should be clear which solutions are accepted, what must be revised, what should continue, what may require funding, what should be transferred elsewhere, what should stop, or what additional conditions must be met.

When existing projects have been completed or transferred and no new meaningful projects related to the application are emerging, the cluster closes.

The group disperses. Its members do not need to discover who else belonged to the same temporary group. Accountable anonymity remains after the process: the group should not become a private club of experts, a social network, or a new institutional caste.

What remains is not a list of members, but a memory of participation: updated specialization weights, collective-intelligence currency, financial compensation and the results of the collective process.

flowchart TD
    A["1. Application is registered"] --> B["2. A temporary application cluster is created"]
    B --> C["3. AI estimates likely specialization needs"]
    C --> D["4. Invitations + open registration"]
    D --> E{"Does the participant already have sufficient weight in a relevant specialization?"}
    E -- "Yes" --> F["Direct entry as brainstormer"]
    E -- "No / insufficient evidence" --> G["Applicant: AI test on the application core and relevant periphery"]
    G --> H{"Sufficient knowledge result?"}
    H -- "No" --> X["Not yet admitted to full participation"]
    H -- "Yes" --> I["Candidate: shadow evaluation of real ideas"]
    I --> J{"Evaluation profile reliable enough?"}
    J -- "Not yet" --> I
    J -- "Yes" --> F
    F --> K{"Critical cumulative specialization mass for the project?"}
    K -- "No" --> D
    K -- "Yes" --> L["One or several brainstorms begin"]
    L --> M["Ideas are generated and evaluated; new people may join"]
    M --> N["New projects may be created"]
    N --> O{"Would another group handle the issue better?"}
    O -- "Yes" --> P["Project transferred outside the cluster / new cluster created"]
    O -- "No" --> Q["Project continues in this cluster"]
    Q --> R["Near completion, actual project specializations are proposed and evaluated"]
    R --> S["Importance of specializations is refined"]
    S --> T["Participants' specialization weights grow with real contribution and saturation"]
    T --> U{"Are meaningful new projects related to the application still emerging?"}
    P --> U
    U -- "Yes" --> K
    U -- "No" --> V["Clear resolutions for the application's next steps"]
    V --> W["Cluster closes"]
    W --> Y["Group disperses without revealing full membership"]
    Y --> Z["Only the memory of contribution remains in updated weights and other records"]