# SenTient

> Canonical HTML: https://initkoa.org/technology/sentient
> Markdown mirror: https://initkoa.org/technology/sentient/index.html.md
> Route: /technology/sentient
> Source: app/technology/sentient/page.mdx
> Generated: 2026-06-13T02:25:42.872Z

[Open the HTML page](https://initkoa.org/technology/sentient)

# SenTient

## Semantic Entity Intelligent Transformation

**Version:** Draft architecture note
**Status:** Conceptual / experimental — not production-ready

> SenTient is an architectural direction and working specification, not a released production system. References to pipelines, components, latency targets, benchmarks, or output profiles describe intended design behavior, not current operational readiness.

## 1. Executive Summary

**SenTient** is the proposed semantic resolution and transformation engine of the Koa ecosystem.

Its intended role is to bridge messy, ambiguous, multilingual, or semi-structured inputs with structured epistemic artifacts that can later be compiled, validated, federated, queried, and rendered.

SenTient does not decide universal truth.

SenTient is intended to help transform signals into structured candidates, preserve ambiguity when resolution is uncertain, and prepare material for Kristal-compatible epistemic compilation.

In the Kristal v5 model, SenTient may contribute to:

* entity reconciliation;
* relation extraction;
* predicate resolution;
* ambiguity preservation;
* evidence linking;
* confidence scoring;
* normalization;
* candidate ranking;
* projection into **Structured Epistemic State**;
* optional Claim-IR or Resolved Claim-IR profiles when an extractor pipeline uses them.

Claim-IR is not the universal required boundary in Kristal v5. It remains a useful extractor and resolver profile. The normative Kristal v5 input unit is the **Structured Epistemic State**.

## 2. Core Philosophy

SenTient is designed as a **Hybrid Orchestration System**.

It combines three technological lineages into a single funnel:

* **Speed:** fast lexical tagging and candidate discovery;
* **Semantics:** contextual NLP, embeddings, and disambiguation;
* **Structure:** durable state, reviewability, traceability, and export into structured epistemic forms.

The goal is not to force every input into a single answer.

The goal is to produce structured, inspectable, traceable candidate states that preserve enough metadata for downstream validation, authority recognition, federation, and reader policy.

This page describes the intended architecture. It does not claim that the full system has been implemented, benchmarked, deployed, or validated.

## 3. SenTient in the Kristal v5 Pipeline

SenTient is designed to sit before or beside Kristal compilation.

A typical intended v5 flow is:

Raw text / document / dataset / submission
-> SenTient extraction and resolution
-> candidate assertions
-> ambiguity-aware resolved state
-> Structured Epistemic State
-> Kristal compilation
-> Working Artifact
-> Validation Reports
-> Authority Recognition
-> Reference Artifact or Runtime Pack
-> Reader Policy