🎥 The YouTube version is also available:
Turning Long-Term Goals into Executable Decision Paths | Transitive Decision Planning | Chinoba
Books: Decision Trace Fabric on AWS: Built-in and Attached – Connecting Enterprise Decisions

In enterprise improvement initiatives, some work can be completed quickly, while other work requires a long period of time.
For the former, it may be enough to identify the issue, compare options, obtain approval, and execute.
But initiatives such as preventing the recurrence of quality defects, advancing equipment maintenance, transforming cross-functional business processes, or fully deploying generative AI involve a long distance between the current state and the final goal. Along the way, more information becomes necessary, stakeholders change, and assumptions are updated.
When AI is asked to create an end-to-end plan for such a long-term goal all at once, the result can easily become plausible but difficult to execute.
On the other hand, optimizing only the next immediate step can make the organization lose sight of the overall destination.
What is needed is not an attempt to connect the final goal directly to the current state.
It is the discovery of reachable intermediate decision states—states supported by evidence and approval—and the recursive linking of those states.
At Chinoba, this idea can be implemented as Transitive Decision Planning.
Why errors accumulate in long-term decisions
In long-term planning, an error in an earlier decision becomes an assumption for the next decision.
Consider the goal of introducing AI support into a manufacturing operation to improve quality.
At first, an organization may incorrectly assume that sufficient data is available. Based on that assumption, it designs search and analysis capabilities. It then defines field operations and KPIs based on that design.
If it later discovers that record granularity differs by process, that workers cannot see evidence supporting recommendations, or that responsibility for exceptional cases has not been defined, major rework will be required in later stages.
This occurs because the longer a plan is connected in one direction, the more small initial errors accumulate.
In Chinoba’s approach, a long-term goal is not treated as one enormous chain of reasoning. Instead, the problem is decomposed through reachable intermediate states.
What is an intermediate decision state?
An intermediate decision state is not merely a milestone.
It is a state in which not only what has been executed, but also the evidence, agreements, constraints, and unresolved issues needed to move to the next decision are organized.
For example, in an AI implementation project, intermediate decision states may include the following:
| State | What has been achieved | Conditions for proceeding |
|---|---|---|
| Problem definition completed | Target process, users, and pain points have been agreed | Out-of-scope areas and success criteria are clear |
| Evidence inventory completed | Original sources, versions, scope, and gaps are understood | Required Evidence Coverage is satisfied |
| Minimum prototype approved | Prototype Brief and human decision boundaries are defined | Evaluation metrics and stop conditions exist |
| Field validation completed | Results of actual use and exceptions have been recorded | Evidence exists for broader application |
| Operations integrated | Approved rules and responsibilities are embedded | Ongoing monitoring and update methods exist |
By passing through such states, the distance from the current position to the final goal can be divided into smaller, verifiable subproblems.
Good intermediate goals should be selected from proven states
It is risky for AI to freely invent intermediate goals in long-term planning.
Abstract goals such as “improve the data,” “align stakeholders,” or “evaluate a prototype” may sound reasonable. But if it remains unclear what must be in place to declare them achieved, what evidence is required, or who must approve them, they cannot serve as assumptions for the next stage.
At Chinoba, candidates for intermediate states should be selected, as much as possible, from proven states that already exist in Decision Traces:
- Business processes that were previously approved and executed
- Prototypes with recorded outcomes, including success or failure
- Approval states defined by current regulations or procedures
- Validation-complete states that proved effective in similar themes
- Evidence Sets and decision boundaries actually reviewed by people
This prevents imaginary, unmapped locations from becoming waypoints simply by assumption.
Of course, a new state may sometimes be necessary. In that case, it should be explicitly labeled as a “hypothetical state,” with its achievement conditions, validation method, and responsible owner defined before it is used.
The basic form of Transitive Decision Planning
Let the current state be S, the goal state be G, and an intermediate decision state be M.
Rather than planning directly from S to G, divide the problem into two questions:
- Can the organization safely reach intermediate state M from current state S?
- Is there evidence that the organization can progress from intermediate state M toward goal state G?
If an M that satisfies both questions can be found, the plan from S to G can be treated as a shorter, verifiable path rather than a single direct leap.
The same decomposition can then be applied recursively to each segment.
| Segment | Question |
|---|---|
| S → M | What must be verified, agreed, or prototyped to reach the intermediate state? |
| M → G | What about the intermediate state supports the possibility of reaching the next state? |
| Non-decomposable segment | Is information missing, is human judgment required, or is the goal too large? |
The advantage of this design is that when planning stops, the cause is easier to identify.
Is the current state unclear?
Are the achievement conditions for the intermediate state undefined?
Is there insufficient evidence supporting the connection to the goal?
Or are execution authority and human agreement missing?
Instead of treating the problem simply as “the AI could not produce a good plan,” the organization can return to the part of its Decision Infrastructure that needs strengthening.
Comparing multiple transit paths rather than choosing one optimal route
In practice, there is not always one clearly optimal intermediate state from the beginning.
For example, there may be at least three possible routes toward quality improvement:
- Improve data quality first
- Conduct a minimum viable prototype in a limited process
- Formalize field decision rules before anything else
Each may be valid, but they differ in required time, risk, stakeholders, and evidence maturity.
At Chinoba, rather than selecting and fixing one intermediate state too quickly, multiple candidates can be compared together with their Evidence Sets.
| Candidate intermediate state | Main advantage | Primary uncertainty | Required confirmation first |
|---|---|---|---|
| Data quality improved | Increases the reliability of later analysis and automation | Short-term impact may be difficult to see | Missing-data rate, recording burden, ownership |
| Limited prototype validated | Enables early validation of field value | May not generalize to other processes | Scope, evaluation metrics, stop conditions |
| Decision rules approved | Defines the boundary between AI and human judgment early | Validity with real data remains unproven | Expert agreement, exception conditions |
The purpose of comparison is not for AI to select one answer. It is to enable people to decide which intermediate state should be pursued first, with evidence and trade-offs made explicit.
Value estimation should include confidence, not only maximum potential
One danger in long-term planning is selecting only the path that appears most attractive.
An intermediate state may promise a large benefit, but if that benefit depends on a small amount of uncertain evidence, it should not be treated as a definitive plan.
At Chinoba, the value of a path should not be fixed as a single expected value. It should be shown together with:
- The strength and independence of evidence
- Past records of reaching similar states
- The degree of match between applicability conditions
- Unresolved issues and counter-evidence
- How far the organization can safely return if the path fails
- Points where human judgment and approval are required
This makes it easier to select not the most dramatic option, but the option that the organization can genuinely pursue from its current position.
Transforming Decision Traces into reusable routes for long-term goals
A Decision Trace is not only a record for auditing past decisions.
When intermediate decision states are retained, future AI systems and people can learn viable routes toward similar goals.
For example, when a new department introduces AI support, it may retrieve knowledge from previous cases such as:
- Which evidence gaps led to prototype suspension
- Which intermediate states, when established first, made field validation easier
- Which approval boundaries should have been defined early
- Which KPIs appropriately demonstrated value in actual use
These are not merely summaries of completed implementation cases.
They are reusable decision paths: records of which state transitioned to which state, and on what evidence.
Chinoba can accumulate these paths as Knowledge Flows, then search, compare, and reconstruct them for the next relevant theme.
AI that is strong with long-term goals is not AI that predicts the distant future in one attempt
What AI needs to address long-term business goals is not the ability to write a grand roadmap that sounds convincing.
It is the ability to accurately understand the current state, identify evidence-backed intermediate states, verify achievement conditions, and reconfigure the route when necessary.
Long-term decision-making is not about directly optimizing a distant goal.
It is about recursively connecting verifiable intermediate decision states with evidence.
Chinoba preserves conditions and provenance through Semantic Chunks, connects evidence through an Evidence Graph, and records state transitions through Decision Traces.
By adding Transitive Decision Planning, long-term change can be treated not as an abstract planning document, but as a reusable and verifiable decision path.

Chinoba
Intelligence as Relationship
Research Platform
founded by
Masao Watanabe
AI Systems Architecture
Decision Trace
Human–AI Coordination
Algorithmic Governance
Related Research
This topic is part of the Chinoba Knowledge Base.

コメント