TILOS · The Integrity Layer Operating System

What matters should shape what happens.

TILOS: a values optimization engine for governance

An open-source, domain-agnostic engine being built to make values-directed governance computationally actionable. For individuals. For organizations. For communities. For AI systems.

We choose what we optimize for. We set values, make commitments, and decide what we want to stand for. But what we say matters and what we actually do can quietly drift apart. TILOS is being built to compute that distance, evaluate how it changes, and inform the responses that bring action back toward what we chose. Its longer-term research frontier: modelling what different value configurations could produce.

Why TILOS is open source Explore the vision Reference architecture →

Why TILOS exists

The distance
between what we value
and what we doSometimes that distance is small. Sometimes it grows gradually, almost imperceptibly. And sometimes, by the time we recognize it, the consequences are already significant. Until now, it has been difficult to compute, evaluate, and act on systematically.

We choose

Values, commitments, and boundaries are declared.

We act

Decisions, behaviours, and outcomes unfold in the real world.

We observe

Evidence of what actually happened is gathered, with its provenance and limits made explicit.

We recognize

We can see how closely what we do reflects what we declared, and how that relationship changes over time.

We respond

Reflection, oversight, and correction happen earlier, while change is still easy.

TILOS does not decide what anyone should value. Values are configured by the people and governance authorities they belong to. The engine evaluates behaviour against those choices, subject to evidence, context, and human oversight. That distinction protects human agency, and it is why one engine can serve every domain.

Four worlds

One engine. Many expressions

TILOS is domain-agnostic. The relationship between what we value, what we do, and what happens as a result exists wherever decisions are made. These are not four products. They are four places the same engine can be applied.

01 / The individual

What if we could see ourselves more clearly?

Recognizing the distance between what you value and how you are actually living, without judgment.

Imagine

Being able to recognize patterns in your choices that no longer reflect the life you want to create. Not to judge yourself. Not to pursue perfection. To understand where intention and behaviour align, where they don't yet, and what might bring them closer together.

The possibilityGreater self-awareness, agency, and intentional change.

02 / The organization

What if integrity could be observed, not just declared?

Seeing the gap between commitments and operational reality before it becomes a crisis.

Imagine

An organization that does not have to wait for an audit, a complaint, or a headline to discover its operations have drifted from its commitments. Where governance becomes more than policies and periodic reporting, and accountability rests on evidence rather than assurances alone.

The possibilityMore accountable decisions, earlier intervention, and demonstrable governance.

03 / The community

What if public trust could rest on observable accountability?

Examining whether institutions follow through on the promises they make.

Imagine

Communities with better ways to see whether the institutions serving them honour their commitments. Public promises examined alongside observable decisions and outcomes, with people participating in defining shared priorities and evaluating progress, under appropriate transparency and safeguards.

The possibilityStronger civic accountability, meaningful participation, and evidence-informed trust.

04 / The AI system

What if AI could be governed against the values it is meant to serve?

Governance that does not end when a model is deployed.

Imagine

AI systems whose observable behaviour is continuously evaluated against the values, objectives, and boundaries set for them. Emerging inconsistencies detected, examined, and escalated to human oversight. Not by assuming AI will always behave as intended, but by building the capacity to recognize when it does not.

The possibilityContinuous, evidence-grounded evaluation of AI behaviour and more responsive human oversight.

Different values. A shared need

Individual choices shape organizations. Organizations shape communities. AI systems increasingly participate in all three. The same engine can follow the thread through each.

Declared↔Observed

How it works

From commitment to consequence, and back

At its heart, TILOS is a closed computational loop. Declared values and observable behaviour become evaluable governance states. The engine computes the Integrity Delta between them, evaluates how it changes over time, and informs a response whose effects are then evaluated in turn.

01 · Define what matters

Establish the values, commitments, and boundaries against which behaviour will be evaluated. Values are set through the appropriate governance authority; the engine never supplies them itself.

02 · Observe what happens

Gather evidence of decisions, actions, interactions, and outcomes, with explicit attention to provenance, uncertainty, and evidentiary limits.

03 · Evaluate the difference

Represent declared commitments and observable behaviour as evaluable states, then compute the distance between them, accounting for context. That measurement is the Integrity Delta.

04 · Recognize emerging patterns

Evaluate how the Integrity Delta changes over time, weighing evidence, context, provenance, and uncertainty, so meaningful drift is distinguished from isolated noise.

05 · Support a response

Inform governance responses designed to improve alignment with declared values: reflection, oversight, intervention, or correction, with humans in the loop for decisions that matter.

06 · Learn and reassess

Preserve relevant history and evaluate whether subsequent behaviour reflects the intended change. Assessments never become their own evidence.

A communication model of the loop. The formal specification, including all modules, is in the TILOS Reference Architecture.

Values optimization

A values optimization engine for governance

First, align behaviour with what we value. Then, explore what our values make possible. TILOS is being developed around two complementary horizons, in that order.

Primary engineering objective Foundation

Optimize toward declared values

How can our decisions and actions better reflect what we have chosen to value?

Explore the foundation

TILOS evaluates observable behaviour against a configured values baseline, shows where actions have moved away from commitments, and supports governance responses designed to bring actions into closer alignment with commitments. The values are the reference point; the engine helps improve alignment with them.

This is the foundation. The working computational core demonstrates parts of this evaluation today; the full optimization and feedback loop is under active development and has not yet been validated.

SafeguardOptimization stays bounded by governance constraints. The goal is never to maximize a score while producing harm elsewhere. That would be optimizing the metric, not the values.

Future capability Subject to validation

Model alternative value configurations

What becomes possible when we change what we optimize for?

Explore the possibility

With sufficient evidence, rigorous testing, and independent validation, TILOS could develop the capability to model how different values, priorities, and weightings influence observable behaviour.

Imagine an organization that holds both innovation and safety as core values. What patterns might emerge if innovation is weighted more heavily? What trade-offs become visible if safety takes precedence? Exploring alternative configurations before consequential decisions, rather than discovering the outcomes after the fact.

Honest statusThis is a research objective, not a demonstrated capability. It depends on reliable models, predictions tested against observed outcomes, and evidence that they generalize. Validation can establish whether it works within defined conditions; it cannot promise it will work in every domain.

Horizon 1 builds the evidence Horizon 2 needs

First, we learn to better honour the values we have chosen. Then, with enough evidence and validation, we may understand and anticipate more of what choosing those values produces.

Measure→Optimize→Explore

From measuring alignment to modelling possibility.

Proof of concept

From possibility to working computation

TILOS is more than a proposal. Its working computational core already evaluates what real companies declared against what the public record shows they did, across a decade of history.

Demonstration

TILOS live compute Live

Longitudinal cases evaluating declared commitments against documented events: Wells Fargo, Nucor, and Air Canada, with more cases in development.

→

Specification

TILOS Reference Architecture Zenodo

The full modular architecture: evidence, values configuration, relational evaluation, drift detection, feedback, and governance continuity.

→

Open source

The open build Coming soon

The engine's open-source implementation, under active development, to be released on GitHub so anyone can run, examine, and extend it.

→

What the demonstration does, and does not, establish

It shows that parts of the engine's evaluation run on real evidence: declared commitments and documented events can be represented as evaluable states, compared, and evaluated over time using public records.

It does not yet show the full optimization and feedback loop in operation, deployment readiness, or effectiveness across all four domains. Parameters in the computable core carry provenance tags so placeholders are never presented as findings, and further validation, implementation, and domain-specific testing are still ahead. We say this plainly because a governance engine should hold itself to the standard it asks of others.

For organizations

Responsible AI needs more than responsible intentions

Organizations have invested in AI principles, policies, risk assessments, and governance frameworks. TILOS adds what connects them to reality: a reusable computational engine that evaluates those commitments against evidence of actual behaviour and informs the governance response, for your organization and for the AI systems acting on its behalf. Openly specified. Designed for integration. Developed for independent examination.

The goal is not another governance platform. It is to make values-directed governance computationally actionable for every platform and organization that needs it.

Earlier

See misalignment sooner

Notice when everyday decisions begin moving away from declared commitments, while it is still easy to respond.

Evidence

Connect fragmented signals

Bring scattered evidence together, with provenance, uncertainty, and evidentiary limits kept explicit.

Responsive

Make governance act

Support decisions and corrective action, rather than reports that sit on shelves.

Continuity

Preserve institutional learning

Keep memory across assessments, decisions, and changing conditions, so lessons persist past staff turnover.

Integrate, don't replace

TILOS is being built to work as a computational layer inside existing GRC, audit, AI governance, and operational platforms, not as one more suite to adopt.

Your platform+TILOS layer

Open architecture

Built to be examined, challenged, and extended

Accountability infrastructure that cannot itself be inspected asks for trust it has not earned. That is why the TILOS reference architecture is published openly, and why the engine's implementation will follow as open source.

Reference architecture · open now Computable core · CC0 on release Public domain · CC0 Domain-agnostic

Why TILOS is open source Read the reference architecture →

The invitation

What could TILOS make possible in your world?

Not because alignment is a permanent state, but because maintaining it is an ongoing practice.TILOS is being built to make that practice computable, so it can be evaluated and improved. Whether you are a researcher, developer, institution, or governance platform, there is a place for your work here.

Discuss a pilot Experience the demonstration → Read the reference architecture →