What we are exploring
Digital content and records are easy to copy, edit and reattribute, which makes reasoning about their origin and integrity genuinely difficult. This lab studies how software can capture and present provenance - where something came from, how it has changed and whether anything looks inconsistent - in a way a human reviewer can actually understand. We prototype experimental approaches to lineage tracking, tamper-evidence and verification workflows, and we test how clearly they communicate uncertainty.
Throughout, our aim is to make trust signals more legible rather than to pronounce verdicts. A tool that flags a document as suspicious is only helpful if it can also explain why, in terms a person can check. We are equally interested in the failure cases: false alarms, missed tampering and signals that look more authoritative than the evidence warrants. The work is exploratory and is not a substitute for the human judgement and formal processes that ultimately decide what to trust.
What is research-stage today
It is important to be plain about maturity. The provenance, tamper-evidence and verification techniques in this lab are research-stage prototypes, not accredited products or assurance services. They can help surface that content has a recorded history, or that something has changed unexpectedly, but they do not - and cannot - certify that a document is genuine or legally valid. Those are determinations for accountable people and recognised authorities, not for our software.
We avoid any framing that implies a guarantee. A clean signal means our prototype found no anomalies under the checks it ran, which is different from proof of authenticity. We document the assumptions behind each technique, the conditions under which it is meaningful and the ways it can be fooled, so that anyone building on the research understands its limits rather than overrelying on a reassuring indicator.
- Provenance capture shows recorded history, not proof of authenticity.
- Anomaly flags prompt human investigation rather than deliver a verdict.
- Every check logs its assumptions and the conditions it relies on.
- No trust marks, accreditations or legal attestations are issued.
- Known evasion cases are documented alongside each technique.
How verification and human sign-off fit together
In our prototypes, automated checks and human approvals are designed to work as one transparent workflow rather than as a black box that emits a score. Each step is logged, each signal carries an explanation, and the responsible person is given the context to accept, reject or escalate. The software organises and surfaces evidence; the decision - and the accountability for it - stays with people and the processes they answer to.
This separation is deliberate and would carry into any client work. We are not a certification authority, so verification features we research are framed as decision-support that strengthens an existing approval process, never as a replacement for it. The honest position is that good tooling can make trust easier to reason about and harder to fake unnoticed, while final responsibility remains human.
