Skip to content

Explainers

Guides to the methods

15 essays on how the possession simulator thinks, what calibration actually means, why the graveyard of null results is the point, and what the on-page AI can and cannot tell you. Every number inside traces back to a committed artifact — see each essay’s sources at the bottom of the page.

01
How a Possession Simulator Thinks

The correlation between teammates is never programmed in. It falls out of five players fighting over one finite pile of possessions.

2 sources cited

02
What Calibration Means, and Why We Grade Ourselves Against the Market

Being accurate is the easy bar. Matching a sharp closing line within noise is the real one -- and here is exactly where we fall short.

3 sources cited

03
The Graveyard: Why We Publish What Failed

A search that only ever confirms is not a search. Nulls outnumber confirms here by about two to one, and every one is kept.

3 sources cited

04
How AI Agents Built This, and What a Human Decided

About 96 percent of the commits are an agent's keystrokes. The gate strict enough to reject the agent's best ideas is the human's.

2 sources cited

05
Reading an Entity Card Without Fooling Yourself

Every card names the floor that let it in and the thing it is not. Here is how to read one and not overreach.

2 sources cited

06
What the Ask AI Can and Cannot Tell You

It can only relay what the engine returns or admit it does not know. Three things it does well, three ways it refuses.

4 sources cited

07
How to Read a Reliability Diagram

A reliability diagram compares what a probability said, on average, with what happened in the published group.

3 sources cited

08
What a Brier Score Can and Cannot Tell You

The Brier score is a mean squared probability error. Its components help separate calibration from information concentration.

4 sources cited

09
Why Repeated Ticks Are Not Repeated Evidence

Within a game, a smooth probability path can produce many ticks that carry closely related residual information.

4 sources cited

10
Reading an Atlas Analysis

An entity atlas is a committed collection of descriptive entity cards, not a single score or a current feed.

3 sources cited

11
Calculation Inputs: How a Derived Row Is Reproduced

A derived row exposes its source paths and substituted operands so its displayed calculation can be checked against the committed artifact.

5 sources cited

12
Reading a Score Decomposition Receipt

A worked score row keeps the published Brier value, its binned components, and the reconstruction remainder separate.

1 source cited

13
Reading Residual Anatomy by State

A state row reports recorded absolute error with its volume; it does not identify why that error occurred.

1 source cited

14
Reading Blowout Timing

A permanent-margin table describes when a published game margin remained in place through the final clock.

1 source cited

15
Reading Adjacent State Contrasts

A contrast compares neighboring published state buckets while preserving the labels and support of both rows.

1 source cited