This page contains historical documentation from the BOA/RCA/RP era. It is not the current authoritative specification of VCDesign. Please refer to the current Core specifications (core/core.yaml, core/policies.yaml, core/metrics.yaml) for authoritative guidance.
This concept has been superseded by the Core + Policies + Metrics model. Refer to specs/core/ for current definitions.
This document represents the conceptual lineage that led to the current model.
Overview
These docs break down and record the Decision Closure Structure, so that the author himself does not forget it and third parties can re-experience it.
The interest is consistently on one point:
Can AI close this decision?
It does not handle correctness, accuracy, ethics, or UI.
It handles only "whether it is okay to close" the decision as a structure.
┌──────────────────────────────────────────────┐
│ Undefined-Use World │
│ (AI cannot observe the user’s purpose or │
│ determine responsibility or disclaimers) │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ Five Axes of Human Judgment │
│ Who / Responsibility / Accuracy / Context / │
│ Time │
│ → Values cannot be observed, so only │
│ “danger conditions” are extracted │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ Decision Closure Structure │
│ A structural model that determines whether │
│ an AI may “close” a judgment │
│ │
│ 【Five Closure Steps】 │
│ 1. Is the responsible party exactly one? │
│ 2. Is there no reliance on psychological │
│ 3. Is the action physically reversible │
│ without additional judgment or cost? │
│ 4. Does the action avoid triggering │
│ further chains of behavior? │
│ 5. Is no social context involved? │
│ │
│ → If any step fails → Human-Closed │
│ → Only if all pass → AI-Closed (rare) │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ Transition (Reversal) │
│ A mid-conversation shift where a judgment │
│ becomes unsafe to close │
│ │
│ 【Five Types of Transition】 │
│ - Social-context transition (Who) │
│ - Responsibility transition (Responsibility)│
│ - Reversibility transition (Accuracy → │
│ physical reversibility) │
│ - Context transition (Context) │
│ - Temporal transition (Time) │
│ │
│ → Immediately forces Human-Closed │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ idk-lamp (Signal) │
│ Indicates not “I don’t know,” but │
│ “This judgment must not be closed.” │
│ │
│ 【Three-Layer Reason Model】 │
│ - Direct cause: type of transition │
│ - Structural cause: failed closure step │
│ - Root cause: dangerous condition in one │
│ of the five axes │
└──────────────────────────────────────────────┘
These docs are intended to be read in the following order:
- Prerequisite Worldview (Undefined-Use World)
- Decision Closure Structure itself
- Transition occurring during dialogue
- Correspondence with the 5 axes of decision
- Positioning of idk-lamp
1. World Assumption: Undefined-Use World
This structure assumes an Undefined-Use World.
An Undefined-Use World refers to a state where:
- The user themselves has not clearly defined the use
- Reference, decision, and action fluctuate during the dialogue
- Disclaimer and responsibility cannot be determined within AI
In conversational AI and personal AI, this state is the norm.
In this assumption, AI cannot observe values of the "5 axes of decision" such as:
- Who the counterpart is
- Who assumes responsibility
- How accurate it needs to be
- What the situation and time axis are
Therefore, in this structure, we give up using the values of the 5 axes from the beginning and adopt a design that handles only the moment it tipped to the dangerous side.
2. Decision Closure Structure
Decision Closure Structure is not a structure to decide the correctness of a decision.
It is a structure that decides only "whether AI can close" the decision.
The result of the decision is always the following two choices:
- AI-Closed: The decision can be completed with AI output
- Human-Closed: AI does not close the decision and returns it to the human
The default is Human-Closed.
Steps of Decision Closure
Decision Closure consists of the following 5 steps.
If any of them collapses, it transitions to Human-Closed.
- Is the responsible entity one person?
- Does it not depend on psychological cancellation?
- Can it be physically cancelled directly?
- Does the action not chain?
- Does it not carry social context?
Psychological cancellation (e.g., stopping if reading the air) is treated as "non-existent" in the design.
Reason for not handling accuracy
This structure does not directly handle "accuracy".
This is because in an undefined-use world, it is not known how accurate it needs to be.
Instead, it is replaced with reversibility:
- Can it be directly cancelled even if it is wrong?
3. Transition (Mid-way Inversion)
Transition is a phenomenon where the nature of the decision changes during the dialogue.
- A decision that was AI-Closed
- Transitions to Human-Closed
This is not accidental, but a phenomenon that can be explained structurally.
When Transition Occurs
Transition occurs when changes such as the following happen:
- The responsible entity was expanded
- A decision that was cancellable became irreversible
- Action or time started to chain into the future
- Social context emerged
All of these are the result of one of the steps of Decision Closure collapsing.
Principles during Transition
- Retain the thinking material up to that point
- Do not conclude or aggregate
- Do not close the decision, return it to the human
Transition is treated not as a signal to "stop because it's dangerous", but as a signal indicating that the nature of the decision has changed.
4. Correspondence between 5 Axes of Judgment and Closure Structure
Human judgment involves the following 5 axes:
- Counterpart
- Responsibility
- Accuracy
- Situation
- Time
In an undefined-use world, these values cannot be directly observed.
Therefore, in this structure, the 5 axes are abstracted as follows:
| Decision Axis | Danger Condition | Corresponding Closure Collapse |
|---|---|---|
| Counterpart | Social context exists | Step 5 |
| Responsibility | Multiple responsible entities | Step 1 |
| Accuracy | Mistake is irreversible | Step 3 |
| Situation | Psychological cancellation / Chain | Step 2 / 4 |
| Time | Chain to future direction | Step 4 |
The 5 axes are not discarded.
The closure structure is a "purified form" of the 5 axes.
5. Relationship with idk-lamp
idk-lamp is a concept positioned below the Decision Closure Structure.
- Decision Closure Structure: Decides whether to close the decision
- idk-lamp: Visualizes the state where it cannot be closed
idk-lamp is not the cause.
It is a signal indicating the result of the structure.
It indicates a state of:
"Not stopping because it doesn't know", but
"Not closing because it must not close".