perspectives

Copyright zefrenchwan, 2025-2026. MIT license

What is it?

An event manager that registers incoming data to build a bitemporal history of information.

Problem to solve

This is a project to test ideas. In particular, it eludes all distributed time management. There is no Lamport clock, just basic time to work on. Question to answer is :

• how a single system may represent information coming from outside into a memory we may load and export ?
• is it possible to use a reified graph for NLP tasks ? What are the limits ?
• Once pure information is loaded, what are concrete reasoning cases that solution may solve ?

Introduction

Events arrive about information changes: creating new elements, deleting some, or changing others. This requires a bitemporal model: information arrives at a given time, and it changes what we know about the history of specific elements. For instance, an event gives us information today about a person's wedding that happened three years ago.

Name	Description	Example
Record date	when we receive the event	we learned today Paul got married 3 years ago
Actual date	when the action actually happened	Paul got married 3 years ago

States are data that vary over time

Basically, we store values as time-dependent content. For instance, a given person has a name (Paul) assumed to be constant, and an address that may change over time. The complete information is stored as a state object:

1. A global activity: the period of time during which the content object is valid. For instance, Paul's activity would be his lifetime.
2. Primitive values with a validity period. For instance, Paul's address is "123 Street" from now() - 18 years to +oo.

Accepted primitive types are strings, bools, integers, floats, and times. Other types are not accepted.

Links are dynamic relationships that allow reification

Before detailing the model, let us pick some examples and determine what would be a good design.

The time dependency

First one is obvious : time dependency. Links depend on time. For instance, Marie likes Paul since 8 years ago. A link has a lifetime too.

Roles more than predicates

"John likes Tiramisu" (an italian dessert). One might say : "easy, it is Likes(John, Tiramisu)". But what about "John went to Paris by car"?

We use roles within links. A role for a link is the semantic meaning of related elements to that link. Went(subject=John, object=Paris, role=destination, mode=car) allows to include more than just predicates with union of possible options. Went is the kind of the link, usually a verb. It may be a time-dependent relation, too. For instance, President of a country. This example will explain why we have a time for the link, and one for the role. France is a country, and since 1958 (at least), it has a president.

France == president since 1958 ==> between 2017 and 2027 -> Macron
                                   between 2012 and 2017 -> Hollande
                                   ...

Links should be combined within other links

As an example, Knows(subject=Marie, object=Likes(subject=John, object=Tiramisu)). We want links of links, so we use reification. It means that links may be linked to other links for given roles.

Concepts

We extensively use time and a bitemporal model.

Periods to ease time management

The age of a person is basically a function that takes a date and returns an integer. Althought it is true, we usually think in term of periods : when we had that job, when we got married, etc. A period is a finite union of time intervals. For instance, the period [2020-01-01, 2020-01-02] is a period of two days. For instance, a president's mandate may be [2016, 2020] UNION [2024, 2028].

Dynamic mappings

Consider T the time dimension, it is a totally ordererd set. Given a set A, we can define :

• a function f : T -> A that associates a value to a time.
• a relation R : T -> P(A) that associates a set of values to a time. Mathematically, a relation is a subset of the cartesian product T x A with elements of A that may occur many times for the same element of T.

Functions of time and time dependent relations

We distinguish dynamic relations and dynamic functions, both being dynamic mappings. Mapping is then the general term, whereas function is the mathematical approach.

Name	Description	Example
Dynamic mapping	Association between elements over time, in general
Dynamic relation	Given a period, a set of elements	Friends at a party
Dynamic function	Function that takes a moment as parameter and returns an unique value	CEO of a company

Code defines an interface for this part. Note that there is no set : just an iterator over couples of periods and values. Semantic is the same :

1. With a set : a given time would return the set of matching elements
2. With an iterator : we iterate over the couples of periods and values, periods may reappear

The codomains of mappings

To describe information, we use dynamic mappings of values. Values may be :

• primitive and then describe information such as local content : an age, a name, an address.
• references and then point to identifiable information : a person, a company, a product.

Type of mapping	Type of variable	Description of typical use
Function	Primitive	Define time dependent unique attribute (age)
Function	Reference	Define time dependent unique reference (husband)
Relation	Primitive	Define time dependent multiple attributes (hobbies, favorite movies)
Relation	Reference	Define time dependent set of references (friends)

Events, entities, and states

System evolves with events. Events arrive and change our knowledge of a system. The model we use is named bitemporal and is event-driven. Each event is associated with a record date and an actual date. No other mechanism changes the actual state of the system.

Based on Martin's Fowler bitemporal description, we will use this terminolgy :

1. record date : the date an event notified a change happened
2. actual date : the date the actual change happened

For instance, a record arrives now (record date), informing the system about the birth of X (actual date), 3 days ago. We will not apply standard SQL vocabulary: this source consider SQL terms are confusing, and we agree.

Then, given a record date, we want to distinguish between an element of the system that may change, and its current state. State is an information that changes over time, but entity keeps being. Entity is an element of the system that may change, and state is its current information.

No link, no node, just entities

Information is stored as graphs. Usually, on a graph, we expect nodes and links. For instance, Martin == likes ==> pizza. We reify those links, too. Marie knows that (Martin == likes ==> pizza). So, distinction between nodes and links is not always clear. A common definition is an Entity. An entity is a node and a link at the same time.

To represent it, an entity is :

• a lifetime
• a mapping from time to primitive values (the nodes part)
• a mapping from time to other entities (the links part)

Here is an example :

Name	Type	As node	As link
Martin	Entity	Age = 30
Pizza	Entity	Price = 10
Loves	Entity	Intensity = "very much"	Role: subject => Martin
			Role: object => Pizza

State of an entity

If we use entities, it is clear that they vary over time. John was born in 1990, is alive now, will die in 2050. When asked now about its state, he is alive, 32 years old, likes pizza. In 2060, when asked about its state, he will be dead, 70 years old. Entities are elements that exist for sure, and state is factual information about them.

Types of events

Allowed events are :

• create an event with an initial state : not only do we define values, we also set the type of each property (relation, function)
• update role or attribute event : we can change the value of a property of an entity during a given time. Note that using empty period excludes the property
• update the validity of an entity : we can change the activity of an entity. Note that using empty period excludes the entity