Open architecture アーキテクチャ

How Telstar thinks about decisions.

Telstar is an evolving open architecture for turning organizational knowledge into evidence that machines can reason over deliberately. This page is the technical view — the decision tiers, the knowledge model, and an honest account of what exists today.

INSTRUMENT LAB // FOUR DECISION TIERS COORD // T0–T3 EVALUATION MATRIX
01 Architecture アーキテクチャ

Use the cheapest reliable layer that can resolve the question.

The architecture is being developed around escalation: resolve a question at the lowest tier that is genuinely sufficient, and escalate deliberately when the evidence does not support an answer. Cost matters, but so does declining to answer.

Organizational knowledge ナレッジ Documents · policies · systems
Polaris ポラリス Knowledge core
Structured evidence ストラクチャード Authority · scope · lifecycle
AI & agents エージェント Answers · actions
Telstar decision layer デシジョン
Escalates upward only
T0 Implemented
Rules & governance

Permissions, policy, lifecycle and conflict edges. Settles what a lookup or a hard rule can settle.

T1 Implemented
Ranking & sufficiency

Routing, relevance ranking, and a judgement about whether the retrieved evidence is actually enough.

T2 In development
Typed decisions

Specialized mechanisms for bounded decision classes with a closed set of outcomes.

T3 Planned
Reasoning

Intended for genuinely residual ambiguity, reached with better structured context when it is needed.

Never spend a model on what an edge lookup can settle.

Implemented = present in the current codebase. In development = actively being built. Planned = designed direction, not yet built.

Conceptual architecture, not a deployed system. The tiers above describe how the decision path is designed to work.

インプット Input デシジョン Decision アウトプット Output
02 Knowledge model ナレッジモデル

Knowledge that carries its own standing.

Telstar is designed around structured records rather than loose passages. A record is modelled to carry its authority, its conditions, whether it is current, and what it replaces — so that a decision can lean on it deliberately rather than hopefully.

Decision node KB-0000
Model

A claim about how the company works

node_id
KB-0000
node_type
policy · fact · process …
status
lifecycle state
source
authority class
scope
audience · permissions · validity
relationships
supports · conflicts · supersedes
volatility
how fast this claim goes stale
Field names from the current model ノード

The knowledge lifecycle

A node moves through guarded states rather than being trusted on arrival. Transitions are enforced, not advisory.

# lifecycle states RAW → CLASSIFIED → CANDIDATE ↘ CANONICAL ↘ HOLD ↘ CONTESTED ↘ REVIEW ↘ ARCHIVED

Authority is by class

Sources are ordered by the class of authority they carry, not by the form they arrive in. An official document and an official page belong to the same class; a help-centre article and a sales note do not.

official_docs
Handbook, policy, manual
product_database
System of record
help_center
Support articles
sales_notes
Unverified, lowest class

Implemented the node model, lifecycle, and authority classes

03 Observatory オブザバトリー

A control plane, not yet designed.

Observatory is the intended surface for inspecting and governing the system — where knowledge, evidence, and decisions move, and what the system has declined to answer.

Observatory Concept

Planned control plane

A planned control plane for understanding how knowledge, evidence, and decisions move through Telstar.

Design status Not built

No interface has been committed to.

We have not designed or prototyped a production interface for Observatory. It is represented here architecturally, because showing a finished-looking console would describe work that does not exist.

The same applies to the rest of the product surface. Polaris is implemented as a knowledge core and store; Orbits exist as a permission and scope model. Neither has a finished interface.

INSTRUMENT MATRIX // IMPLEMENTATION STATUS COORD // PROVENANCE VERIFICATION
04 Development status ステータス

We know what we are building. We are building it.

Telstar is early-stage and under active development. This page distinguishes what exists from where the architecture is headed, because the difference matters when deciding whether to work with us.

Implemented In the codebase
  • Deterministic decision core — permissions, policy, lifecycle transitions, and conflict edges resolved by rule.
  • Retrieval and sufficiency — routing, ranking, and an explicit judgement about whether evidence is enough.
  • The knowledge model — decision nodes, scope, authority classes, and a guarded lifecycle.
  • An evaluation harness — measuring whether a tier resolved a decision correctly, or should have declined.
In development & planned Ahead
  • Learned decision tiers — specialized mechanisms for bounded decision classes.
  • End-to-end decision routing — connecting the tiers into a single serving path.
  • Observatory — a control plane for inspecting the system.
  • Source connectors — bringing a company's systems into the knowledge core.
What we are building Knowledge Evidence Decision Reasoning

A knowledge system that can reason about authority, sufficiency, conflict, and escalation — before asking a frontier model to decide.