The Glass

Last Prompt is complicated to explain. The architecture is precise, the evaluation is rigorous, and the design decisions are each the result of something specific I observed or experienced. But none of that is why it exists.

It exists because I have been in rooms where the ground moved. And I noticed something in those rooms that most training environments never replicate: the thing that ended the situation was never the marble. It was always the glass.

The marble is the event. The crisis in front of you. The refugee column at the gate, the server breach, the regulatory deadline, the position that needs to close. It is vivid, immediate, and demands everything in the room. Everyone watches the marble.

The glass is the stats. The structural conditions that were quietly shifting while the marble absorbed all the attention. Cohesion declining across four cycles while the security crisis was handled brilliantly. Cash flow eroding while the compliance team produced excellent work. The dimension nobody was measuring because the event didn't mention it.

Last Prompt ends when a stat hits zero. Not when a decision scores poorly. Not when the immediate crisis is mishandled. When the glass breaks — even if every individual score was high.

That is how real systems fail. Not through a single catastrophic wrong decision, but through the accumulated neglect of what nobody was watching while everyone was focused on what was visible.

The purpose of reasoning is not to control the future. It is to remain viable when the future refuses to cooperate.

I have watched this happen at scale three times before I built anything.

*Brandeis, 1996.* I was twenty-three, less than two years into financial markets, when the Herbie Black/Sumitomo crisis hit. Tens of millions of dollars moved in ways that nobody fully controlled. The marble — the positions, the decisions, the daily management — had looked fine. The glass had been cracking for months in places nobody thought to look.

*Y2K, 1999.* I was chosen by the head of operations to liaise between the IT consultants and the traders — to translate between what the system needed and what the traders wanted, regardless of which side was comfortable with the answer. What I discovered was that systems have dependencies nobody notices until they are forced to notice them. The marble was the technology transition. The glass was every assumption baked into infrastructure that had never been stress-tested because it had never needed to be.

*Fuerst Day Lawson, early 2000s.* I arrived not knowing what a bill of lading was. I was brought in to look at how the business worked and find ways to make it work better. What I learned was that ignorance of a domain is not the same thing as inability to reason within it. The marble was the unfamiliar industry. The glass was the structural inefficiency that people inside the domain had stopped seeing because they were too close to it. Someone who didn't know the marble could still read and stablise the glass.

*London, 2008.* A Monday lunchtime. I watched people return from their offices carrying boxes, often already drunk, to tell their families they had lost their jobs, could not pay the mortgage, and that nobody was hiring. The system had appeared stable right up until the moment it was not. Every marble decision had looked reasonable. The glass had been breaking for years in ways that the marble obscured.

After 2008 we made a decision that most people around us would not have made.

Sell. Move to France. Avoid debt or dependency.

Not to optimise inside the system. To change the system.

This is the glass/marble distinction made personal. Staying in London and trying to continue would have been balancing the marble while the glass quietly cracked. The decision to leave was not about the immediate event — it was about the structural condition the event had revealed. We were not solving the crisis. We were protecting a stat in our lives.

The objective was not: make the crisis disappear. The objective was: survive the reality that now exists.

This is what Last Prompt is built around.

Most strategy games teach quietly: the right move fixes the problem. Last Prompt teaches something different: sometimes the best move merely limits the damage. The orange note on a Black Swan event says it plainly — some events carry consequences that no response can fully prevent. An adequate outcome means you mitigated the impact as effectively as possible. Residual damage is the nature of the crisis, not a reflection of your decision.

The refugee column still reaches the gate. The hurricane still hits. The recession still arrives. The market still crashes. The decision is not whether the event happens. The decision is how the system absorbs it.

Most people unconsciously assume: good decisions produce good outcomes. Last Prompt is built around something more precise: good decisions increase the probability of acceptable outcomes. It is not the same thing.

A lot of real decisions look like:

- How do we lose less? - How do we preserve options? - How do we avoid collapse? - How do we absorb the shock?

Rather than: how do we win?

The three skins reflect three different relationships to knowledge — and three different versions of the same underlying truth.

*Colony:* I know this world. Food, shelter, safety, community. Everyone knows this world. But knowledge of the domain does not guarantee good reasoning within it. The marble is familiar. The glass still breaks when nobody watches it.

*Corporate Reckoning:* I don't know this world. The AI middleware, the ERP cascade, the regulatory stress test. Domain ignorance is real and the clarifying question exists for exactly this reason. But lack of knowledge does not prevent good reasoning. The Fuerst Day Lawson lesson. You can read the glass even when the marble is unfamiliar.

*Lockwood:* Nobody knows this world. The Traveller arrives at a moment before the consequences of a discovery are understood by anyone. There is no expertise to defer to. Good reasoning still matters — perhaps matters most — when certainty is impossible and the only available tool is the quality of the thinking itself.

I did not build Last Prompt because I read about decision theory. I built it because I was in the rooms where the systems moved, watched the marble absorb every available unit of attention, and understood — only afterwards, and then more clearly with each passing year — that the glass had been the thing that mattered all along.

The evaluation is rigorous. The architecture is precise. But underneath both is something simpler: the conviction that the conditions under which real decisions are made can be replicated honestly, and that practising inside those conditions — with a world that remembers every choice and ends when the glass breaks — develops something that no credential, no course, and no simulation of a specific scenario can develop.

Not what you choose. How you reason. And whether you were watching the glass.