Skip to content
DIXON.AI The AI Reliability Tracker
Menu

Posts · 14 in this category

Guardrails

The expensive mistakes came from process gaps, not the model.

The evidence so far

5 AI errors and 2 catches logged in 4 of these 14 posts, up to 21 Aug 2026.

The evidence

06

How to tell if an AI answer is true in 30 seconds

Two questions, and a prompt you can copy, that catch a confident, wrong AI answer before you act on it.

Guardrails 3 min read

07

Make it show its working

An AI hallucination is a model blending what it knows with what it invents, in one tone.

Guardrails 3 min read

08

The whole method, in four questions

The four questions I run any AI answer through before I trust it, shown end to end on one everyday example, with a prompt you can copy.

Guardrails 4 min read

10

The AI prompt I run before every sell decision

Every other sell-decision prompt asks AI whether to sell.

Guardrails 11 min read

11

What AI stock research comparisons should test

Most AI stock research comparison pieces test retrieval and issue verdicts about reasoning.

Guardrails 9 min read

14

5 questions to ask AI before buying any stock (2026)

Five questions to ask AI before buying any stock, for the checkpoint after the research and before you commit capital.

Guardrails 8 min read

About Guardrails

Posts here are the site's slower layer: the guardrails you keep. The checklists and frameworks that decide how AI fits into a decision, and where it doesn't get a vote. Less what I ran this morning, more what I run every time. Want the quick version? The seven paste-in checks live in the guardrails toolkit.

That includes the whole method in four questions, run end to end on one everyday example, and how to tell if an AI answer is true in 30 seconds. It also covers the money decisions: the five questions I put to AI before buying any stock, the thesis audit I run before selling, and why most AI tool comparisons end in verdicts you can't act on.

The common thread: the discipline matters more than the model. The expensive mistakes in my own record came from process gaps, not tool gaps: holding through an earnings release with no written sell trigger, buying without a thesis I could state in three sentences. The guardrails here exist to close those gaps. The AI checks the work; the checklist makes sure there's work to check.