Everything here answers one question: how you prove what a decision actually was, months after it was made. Start with the white paper for the argument, the experiments for the evidence, or the short reads if you have ten minutes and a regulator’s letter on your desk.
18 pieces
Featured · Article
Article
What a day at Mastercard's RiskX event taught me about payment fraud, agentic payments and the work around a transaction.
Every figure on these pages is re-derivable from the published corpus.
Featured · Article
Article
We connected MeshQu to selected decisions across our coding task lifecycle. The records helped us identify bottlenecks, question our measurements and begin improving both the work and the rules around it.
Every figure on these pages is re-derivable from the published corpus.
Featured · White paper
A white paper on the structural gap between AI governance frameworks and the moment a consequential decision is made, and what it would take to close it.
01 · adoption vs supervision
81% vs 24%Firms running AI in production, against regulators collecting structured AI data.
02 · liability disagreement
38% vs 18%Regulators who say the firm bears AI harm, against industry that agrees.
Every figure on these pages is re-derivable from the published corpus.
Swipe to change slide.
Category
Short reads, newest first — each card says what it is and how long it takes.
A general statement about your AI governance process cannot answer why one payment was made, and neither can a test showing the agent…
An EU policy process, UK public opinion and a Singapore legal working group have reached for the same requirement in three different…
We broke the previous experiment apart to find which piece was doing the work, then ran the same records through a second AI from a…
We ran the same AI over the same 283 UK procurement records five times, adding a layer of governance context each round. The decisive…
An AI and a written rulebook reviewed the same UK government purchase records at the same moment. The AI named the same problems the…
AI explainability shows how a model behaved. It does not show which policy, threshold and model version were in force when the decision was…
A route through the research: the argument first, then the three experiments that test it, in the order they were run.
Each experiment answers the question the last one left open. The white paper is the argument they were built to test.
AI governance frameworks describe how decisions should be made — but the moment a decision is actually executed, the evidence to defend it almost never exists in a form anyone can verify.
They almost never agreed — but not because the agent was reckless. The policy engine gave a straight allow or deny; the agent kept reaching for 'needs review' when the evidence was incomplete.
More context didn't make it steadily more decisive. It stayed on the fence until we showed it past decisions, then committed in a single jump — and eased back off once it saw the full policy.
On the same records, GPT-5.4 and Claude Opus 4.7 reasoned the same way, yet committed to verdicts on very different fractions of decisions — 23% vs 80%.
Category
Experiments on 283 real UK procurement decisions, every verdict bound to a signed receipt.
Why two AI models can reason the same way about a decision and still commit to it on very different fractions of records — and what that…
A 283-decision corpus measuring how an LLM agent’s verdicts shift as governance context is added one rung at a time, from no policy to full…
A 283-decision corpus comparing an LLM agent’s verdicts against MeshQu’s policy evaluator on UK Contracts Finder filings — every record…
Category
Updates from MeshQu.
News · 1 min read
MeshQu has been selected for the FCA Digital Sandbox. We will test a single claim against synthetic authorised push payment fraud data…
News · 6 min read
The EU AI Act has been delayed again and its timelines are still being argued over. The decisions already made under it cannot move with…