Synapse Insights is publicly available on GitHub:https://github.com/masao-watanabe-ai/Synapse-Insights
When people hear “organizational analytics,” many think of KPIs, post volumes, meeting attendance, task progress, or engagement metrics.
These indicators are certainly important. But they alone do not reveal how an organization is actually functioning.
A high volume of posts does not necessarily mean that decisions are moving forward. High activity levels may coexist with a fragile form of collaboration in which coordination and decision-making are concentrated in a small number of people. Those who hold critical knowledge may speak relatively little and therefore remain invisible in surface-level rankings.
What truly needs to be observed is not the quantity of activity, but how knowledge, trust, decision-making, and collaboration flow through the organization.
This is the idea from which Synapse Insights began. With v2, that idea evolves further.
What Synapse Insights v1 Aimed to Achieve
From the beginning, Synapse Insights did not treat interactions inside an organization as mere logs.
Replies, questions, proposals, reactions, agreements, and disagreements were treated as relationships with meaning. These relationships were then understood as a graph that changes over time.
Its original framework centered on three concepts:
- Semantic Interaction: Reading conversations and reactions as meaningful interactions, such as questions, proposals, agreements, dependencies, and knowledge sharing.
- Temporal Graph: Observing how relationships, influence, friction, and knowledge propagation change over time.
- Organizational Intelligence: Understanding the intelligence structure of the organization as a whole, rather than simply measuring individual activity.
This framework makes it possible to begin seeing who consults whom, where knowledge is concentrated, and which issues are causing decision-making to stall.
However, conversations and relationships alone are not enough to truly understand organizational intelligence.
Conversations only represent the organization’s actual state when they are connected to real work activities, knowledge assets, the evidence underlying trust, and the outcomes of decisions.
What Changes in v2
Synapse Insights v2 is not merely an Organizational Intelligence Analytics platform.
It evolves into an Organizational Intelligence Runtime that integrates and observes conversations, activities, knowledge, trust, and decisions.
Here, “Runtime” does not mean analyzing organizational conditions only once. It refers to a continuous foundation that ingests daily interactions and activities, interprets their meaning, captures change over time, and connects important situations to human review when appropriate.
The core of v2 consists of the following eight layers.
- Interaction Layer
Ingests messages, replies, mentions, reactions, meeting records, and AI agent events. - Activity Layer
Ingests real operational activities and their progress, including tasks, reviews, approvals, incident response, handovers, and meetings. - Knowledge Flow Layer
Connects documents, specifications, meeting minutes, FAQs, decisions, lessons learned, and candidates for tacit knowledge extracted from conversations. It captures how knowledge is created, used, and delayed. - Trust Infrastructure Layer
Treats trust not as a single score, but as a set of evidence concerning Capability, Knowledge, Intent, Compliance, Explainability, Coordination, and Impact. - Decision Trace Layer
Connects Signal, Proposal, Permit, Override, Execution, and Outcome, making it possible to trace why a decision was made, who was involved, and what result it produced. - Organizational State Twin Layer
Represents the workload, responsiveness, friction, and resilience of organizations, teams, and projects through observable operational signals. - Temporal Intelligence Graph
Integrates the above data into a time-series graph with nodes, semantic edges, timestamps, confidence levels, evidence, and access conditions. - Insight Runtime
Provides visualization, detection, prediction, explanation, alerts, and requests for human review.
Do Not Separate Conversations from Activities
A key feature of v2 is that it does not treat conversations as the only object of analysis.
For example, the fact that questions to a particular member are increasing does not, by itself, reveal whether this represents healthy knowledge sharing or a dangerous concentration of dependency.
However, if reviews, approvals, incident responses, and handovers are also concentrated on that person—and response delays and unresolved issues are increasing—the organization may be becoming excessively dependent on a single individual.
Conversely, a team may have relatively few posts but still function healthily if its knowledge assets are well maintained, decisions are made quickly, and knowledge transfer to other teams is progressing.
By connecting conversations with activities, Synapse Insights can observe not simply what occurs frequently, but what is actually functioning as organizational intelligence.
Connecting Knowledge Flow to Organizational Memory
Organizations contain both explicit knowledge and tacit knowledge that exists only in conversations and experience.
The former remains in documents, specifications, meeting minutes, FAQs, and repositories. The latter exists in the answers provided by experts only when difficult problems arise, in the reasoning behind reviews, and in decision criteria shaped by past failures.
The Knowledge Flow Layer in v2 does not merely check whether documents exist.
It observes:
- which knowledge was referenced in which issues and decisions;
- who provided knowledge and who reused it;
- where knowledge handoffs stopped;
- which knowledge can only be reached through specific experts; and
- whether the knowledge needed for onboarding is actually flowing through the organization.
This makes it possible to identify Hidden Experts and future Knowledge Bottlenecks that are difficult to detect through post volume alone.
The purpose of knowledge management is not to increase the number of documents. It is to create a state in which knowledge reaches the places where it is needed and can be reused in decisions and action.
Why Connect It to Trust Infrastructure?
Within an organization, influence and trust are not the same thing.
A person who speaks forcefully, makes many decisions, or receives many requests for advice may be trusted in some areas but not in others. It is necessary to distinguish the domains in which a person or agent is trusted and the evidence on which that trust is based.
Synapse Insights v2 does not treat Trust as a simple popularity ranking or favorability score.
For a person or an AI agent, it separately considers evidence such as:
- whether the required capabilities are present;
- whether the relevant domain knowledge is present;
- whether actions align with delegated purposes and constraints;
- whether rules, boundaries, and compliance requirements can be respected;
- whether decisions can be explained and verified afterward;
- whether collaboration with others is appropriate; and
- whether the impact of actions can be traced.
This structure connects directly to the concept of Trust Infrastructure.
As a result, Synapse Insights does not merely show “who is popular.” It becomes an observational foundation for considering who can be entrusted with what, and within which scope.
The Organizational State Twin Is Not Individual Surveillance
A concept newly clarified in v2 is the Organizational State Twin.
It is not a mechanism for inferring an individual’s emotions, health condition, personality, or aptitude. Such use can easily lead to inappropriate conclusions from incomplete data and can damage trust within the organization.
The Organizational State Twin deals only with the operational condition of an organization or team.
For example, it can observe organizational-level signals such as:
- whether requests and reviews are becoming concentrated in a small number of people;
- whether approval or response delays are increasing;
- whether repeated explanations of the same issues or unresolved discussions are increasing;
- whether knowledge and workload can be redistributed after an incident; and
- whether the density of collaboration has suddenly declined.
The crucial point is not to treat a signal itself as a conclusion.
For example, slow responses may suggest an increase in workload, but they may also result from vacation, time-zone differences, role expectations, or missing data. Therefore, v2 presents the observation period, supporting events, confidence level, and limitations of interpretation, while assuming human review.
The Organizational State Twin is not a model for evaluating organizations. It is a model for designing conditions in which organizations can collaborate sustainably and recover from change.
Decision Trace Connects Analysis to Action
If an analytics system displays a “trend that requires attention” without showing why it reached that conclusion, users cannot trust it.
Likewise, if someone intervenes after receiving an alert but the rationale for that intervention and its eventual outcome are not recorded, the organization cannot learn.
For this reason, v2 connects to the Decision Trace Model.
For example, when signs of dependency concentration are detected, the following flow can be traced:
Signal → Interpretation → Proposal → Human Review → Permit or Override → Execution → Outcome
This turns an insight into an evidence-based proposal rather than an automatic command.
For interventions with significant impact, the Decision Runtime can apply Permit, Boundary, Human Gate, and Override mechanisms. This is necessary to avoid separating organizational observation from governance and to clarify the responsibilities associated with each.
From Human Organizations to AI Agent Organizations
Synapse Insights v2 is not limited to organizations made up only of people.
As AI agents gather information, delegate to other agents, call tools, propose decisions, and sometimes execute actions, AI becomes not merely a feature but part of the organization itself.
The structure that must be observed therefore expands.
If a conventional organizational model is centered on:
User / Message / Reaction
then an AI Agent Organization also includes:
Agent / Tool / Decision / Signal
What matters is not only how many times each agent operates.
It is necessary to understand:
- how delegation occurs between agents;
- where tool usage and decision-making are becoming concentrated;
- how failures propagate through the system;
- whether the evidence for explainability and compliance is sufficient; and
- at which points humans should intervene.
Synapse Insights v2 observes human organizations and AI Agent Organizations through the same semantic graph and temporal axis. This makes it possible to understand a Multi-Agent System not simply as “multiple models running,” but as an organization through which knowledge, trust, decisions, and responsibility flow.
The v2 Roadmap
v2 does not assume that everything will be implemented at once. It will mature in stages.
First, conversations and operational activities will be integrated, establishing basic Semantic Edges and organizational overviews.
Next, Knowledge Flow and Trust Evidence will be added, enabling observation of Hidden Experts, dependency concentration, and knowledge bottlenecks.
After that, Decision Trace and the Organizational State Twin will be connected to provide evidence-based alerts and pathways for human review.
The system will then expand into an Agent Organization Monitor by integrating AI agents, tools, delegation, and failure propagation.
Ultimately, Synapse Insights will move toward detecting early signals from temporal change, comparing scenarios, and analyzing organizational resilience.
However, prediction is not intended for automatic personnel decisions or automated control. Its purpose is to help organizations understand their own state and choose better designs for collaboration.
Seeing Organizations as Flows of Intelligence
Synapse Insights v2 does not aim to treat organizations as collections of headcount, conversation volume, and KPIs.
It aims to observe organizations as flows of intelligence in which knowledge, trust, decisions, activities, and responsibility interact and change over time.
To achieve this, it integrates the Semantic Organizational Graph, Temporal Intelligence Graph, Knowledge Flow, Trust Infrastructure, Organizational State Twin, and Decision Trace.
This is not simply an extension of analytics.
It is a Runtime for observing how organizations think, collaborate, learn, and adapt.
The public repository for Synapse Insights is available here:
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