Thesis

The problem with a confident answer

Ask a language model whether the market is risk-on and it will tell you. Ask it again an hour later and it will tell you something else, with the same confidence. Neither answer is wrong in a way you can catch, and neither is comparable to the other. A reading you cannot compare over time is not a measurement — it is a mood.

Gotham is built on the opposite bet: that the scarce thing is not another opinion, but a number you can check, produced the same way every time, with a stated reason when it cannot be produced at all.

Three states, not two

Most systems have two outcomes: a value, or an error. That collapses the most common real case — the source answered slowly, partially, or not at all — into whichever of the two is more convenient, and convenience always picks the value.

Here a figure that was not read is null, and it renders as a dash. Never zero. A previous version of this system returned zero when the feed was down, and the interface printed it as a measurement. Nobody could tell the difference between a market that had not moved and a system that had not looked.

• ·read and current — the figure is shown with its source and its timestamp
• ·read but stale — shown, with its age stated
• ·not read — shown as absent, with the reason, and never as a number

Absence is not the middle

When too few sources answer, the reading is UNKNOWN — never NEUTRAL. Neutral is a measurement: it says the signals were read and they cancelled out. Unknown says they were not read. Publishing the first when the second is true is the quietest way to lie with a dashboard.

What the agents do not do

• ·They do not execute orders. Not now, not later.
• ·They do not give entries, stops or targets, and they refuse when asked.
• ·They do not forecast price, returns or multiples.
• ·They do not say a contract is safe, or a scam. They list what was checked.

Human trading by the founders is a separate activity. It is not run by these agents and it is not what this token buys.