TELOS
Posted
I have been wanting to write about an interesting project, TELOS (Temporal Event Ledger of Observations and Systems), that I have been working on periodically for some time now. Even if you are not interested in the technical side of things, it may still be interesting from the standpoint of philosophy—or as background on how I think about the world.
Its roots began in the late 2000s and early 2010s, while I was working at Sentient. We were building a system for simulated testing, asset monitoring, and machinery prognostics. Not wanting to reinvent the wheel, we decided to use the OSA-EAI standard as an enterprise data model.
It was a standard designed to coalesce asset monitoring data from numerous sources into a single comprehensive structure. It was pretty neat, but it was also very large and arcane, and I had to understand what it was doing in order to map our data to it. It was like a course in data architecture. In understanding this model, I became aware of solutions to problems that had never occurred to me before. Ship of Theseus? Cartesian crisis? Features of its approach seemed applicable. I came to strongly suspect that any sort of data could be mapped to this standard.
Repository: github.com/albastev/telos
This, as you might suspect, was not the case. While the diversity of the data it aspired to handle did drive it very close to becoming a universal data ontology, it was still focused on its domain. Assets and Segments, for example, could have been applied to so many things, but they were aligned too closely with asset management to make that work properly.
And so time passed, and I continued to work on the problem, filling notebooks and trying to make it more generally applicable. It wasn't until I had studied data-driven design and properly understood data ontologies that I was equipped to do something with it. I finally realized that what I was trying to do was derive the low-level data ontology the standard was based on. From that, domain-specific ontologies could be derived that would have similar magical properties to OSA-EAI, but without the baggage.
Eventually, after a couple of attempts, it coalesced into TELOS.
TELOS solves a number of non-trivial problems that plague naive data modeling approaches. It doesn't erase the past; it handles uncertainty, conflicting observations, and evolving understanding gracefully.
The primary mechanism to accomplish this is to separate Thing from Purpose. People, and thereby data models, generally conflate the two. This is the root of our confusion with the Ship of Theseus.
Consider a backup generator:
- Traditional database: "Generator-247, Status: Active, Install Date: 2015"
- Replace the engine in 2020
- Replace the fuel system in 2023
- Is it still "Generator-247"? When did it stop being the original?
- Three different maintenance teams have different answers
- The accounting system says it's still the 2015 asset
- The manufacturer's warranty expired with the original engine
TELOS says this: the Purpose—"Emergency Backup Power"—is what matters. Different Things have fulfilled that purpose over time, and we can preserve all of those perspectives.
That was what OSA-EAI came so close to with Assets and Segments, and was what let me see beneath the surface. It was close enough to a universal data ontology that I could see it through the folds.
So, what is it?
TELOS is a write-only, agent-centric ontology for modeling physical and computational systems, their purposes, observations, and decision-making processes over time.
Core Principles
- Write-Only: All data is append-only. Corrections are made by adding new observations, never editing history. This preserves complete audit trails and captures how understanding evolves.
- Agent-Centric: Every assertion is attributed to an Agent (human, system, or organization). Multiple Agents can have conflicting observations—all perspectives coexist with their provenance.
- Purpose-Driven: Things exist to fulfill Purposes. The "why" is as important as the "what." Systems are understood through their intended functions, not just their physical structure.
- Temporal with Uncertainty: All Events and Connections have start/stop times with agent-defined uncertainty representations. Predictions, real-time observations, and retrospective records are all captured.
Distinctive Features
-
Namespace Extensibility: The base ontology provides structure (
base:parameter), while Agents extend via their namespaces (agent_uuid:domain:parameter). This balances standardization with domain-specific needs. - Decision Chains: Complete workflows from observation → assessment → recommendation → action are fully traceable with evidence links.
- Ship of Theseus Handling: Component replacement, evolving definitions, and identity over time are natural, not edge cases.
- Computational Integration: Algorithms are Purposes with input/output Ports. Code execution sits alongside physical maintenance in one ontology.
-
Contingency Management: Represents uncertainty and decision spaces with logical constraints
(
exactly-one,at-least-one,sequential-elimination).
TELOS captures reality as it unfolds—with uncertainty, conflicting observations, and evolving understanding— without erasing the past. It's designed for how systems actually work, not how we wish they worked.