TILOS · Open architecture

Built to be inspected.

Why the TILOS reference architecture is published openly, why the computable core is following it to GitHub, and how you can examine, integrate, and extend both.

Accountability infrastructure that cannot itself be inspected asks for trust it has not earned. A system that evaluates whether others honour their values has to meet that standard first. So every assumption, formula, and placeholder in TILOS is visible.

Read the reference architecture What's in the build →

Why open

The measure of
values cannot belong
to one companyIf a single vendor controlled how alignment between values and behaviour is computed, that vendor would quietly shape what counts as integrity. TILOS takes the opposite position. The method should be a shared public resource, like an accounting standard or an open protocol, so that anyone using it can see exactly how a result was reached.

Inspect

Every formula, threshold, and assumption can be read, so findings can be traced back to how they were produced.

Challenge

Researchers, auditors, and critics can test the method, find its flaws, and publish what they find.

Adapt

Organizations configure their own values and evidence sources. TILOS never decides what anyone should value.

Integrate

Governance, audit, and AI platforms can build TILOS in as a computational layer without negotiating access to a black box.

Endure

Open infrastructure outlives any single steward. The method stays available whatever happens to the people who first built it.

We chose stewardship over ownership. TILOS is not patented and will not be. Publishing the reference architecture openly also places its methods on the public record, which makes them far harder for anyone else to enclose.

What's open

Available now, and what comes next

The reference architecture is published. The demonstration is live. The engine's open-source implementation is under active development, starting with the computable core.

Specification

TILOS Public Reference Architecture v1.1 Published

The complete modular specification: values configuration, evidence, relational evaluation, drift detection, adaptive control, and governance continuity. Archived on Zenodo with a permanent DOI.

→

Demonstration

TILOS live compute Live

Longitudinal cases computed in the browser from public records: Wells Fargo, Nucor, and Air Canada.

→

Source code

TILOS computable core Coming to GitHub

Standalone Python modules implementing the mathematically specified parts of the reference architecture, with a full test suite. Release follows integration with the live demonstration.

The build

What the computable core contains

Each module implements a numbered section of the reference architecture, so code and specification can be read side by side. Dependencies are kept minimal: the core runs on Python and numpy alone.

8specified modules implemented

153automated tests passing

5provenance tags on every output

1call runs a full governance cycle

ModuleWhat it doesSpec

VCLValues Configuration Layer. Standard, custom, or integrated values sets, with a review gate before use.7A.1

M-DSDecision Classification. Scores decisions into three tiers and activates the modules each tier needs.7A.4

M2Stakeholder Dynamics. Influence weights and attribution of who is shaping behaviour.2A

M3Context and Constraint. Adjusts the coherence window for real-world constraints, so pressure is not mistaken for drift.2B

M5Relational Calculus. Computes alignment and the Integrity Delta, with a stability check.4

M6Drift Detection. Separates adaptive change from genuine drift, scores persistence and severity, and flags possible gaming.5

M9Adaptive Controller. The optimization core: a guarded feedback controller that proposes corrective responses, with authorization checks.7.0 to 7.2

M6bResidue Repository. Institutional memory of unresolved issues across cycles. It advises, and never overrides.7A.3

Provenance: how the build stays honest

Every parameter and every output carries one of five tags: REAL ELICITED SPEC_DEFAULT PROXY PLACEHOLDER. An output inherits the weakest tag of anything that fed it.

Many calibration values are still placeholders, and the code says so on every result. A run demonstrates the mechanics; it does not become a finding until its inputs are elicited from the people they describe or measured from real evidence.

Not yet built, and why

Some modules depend on things code cannot supply. Pattern recognition needs a library of validated pattern cards. Scenario modelling needs scenario libraries and human respondents. Resolved signatures need real deployment history. These will be built as those foundations exist, not before.

Openness at work: what building it found in our own spec

Implementing the reference architecture line by line surfaced nine issues in the published specification: arithmetic slips in a worked example, a tier label that contradicted its own table, an ambiguity in the order of a calculation, and several quantities the spec names but never defines.

These will be published as errata alongside the code. This is exactly why the method is open: it got better the moment it was tested, and it will keep getting better as more people test it.

Licensing

In the public domain

TILOS is dedicated to the public domain under CC0. Any person, organization, or platform can adopt, integrate, and build on it, commercially or otherwise, without asking.

Documentation

Reference architecture

Dedicated to the public domain. Use, adapt, and redistribute it for any purpose, with no permission needed.

CC0 1.0 Universal

Source code

Computable core

Will be released to the public domain on completion, free for any use including commercial integration.

CC0 1.0 Universal · on release

Values

Bring your own values

The configuration layer accepts any values set. Organizations supply their own, or use a published registry under that registry's terms.

Patents

None, by design

TILOS is not patented. The patent track was ended deliberately so the method could remain a commons.

Participate

Ways to take part

Open infrastructure is only as strong as the people who examine and use it.

Researchers

Test the method. Replicate the demonstration cases, challenge the mathematics, propose validation studies, and publish what you find, including where TILOS falls short. Critical findings are as valuable as confirming ones.

Developers

When the core reaches GitHub: run the test suite, port modules to other languages, build connectors to evidence sources, and help close the open questions documented in the repository.

Governance, audit, and AI platforms

Integrate TILOS as a computational layer inside the platform your customers already use. No licence negotiation is required. We welcome early conversations about data contracts and integration patterns.

Organizations

Pilot TILOS on your own declared commitments and AI systems. Pilots generate the real evidence that turns placeholders into calibrated parameters, and they move the whole commons forward.

Stewardship

Shared infrastructure for responsible governance

Not another product competing for market share.TILOS is offered as part of the common ground every organization using AI can build on, so that integrity can be examined by anyone, not just asserted by those being evaluated.

Talk to us about a pilot Read the reference architecture → Back to TILOS →