Crypto & AI
Enter the starting and ending values from a claim. The calculation shows the change between those values; you still need to check where they came from and whether they measure the same thing.
Read the result
The result will appear here with an explanation of what it does and does not establish.
What was the starting value? What is counted or measured? For a percentage, find the numerator and the total it is divided by.
Compare equal observation windows and the same measurement rule. For a model score, check the test set, scoring method and run conditions. If these changed, arithmetic alone cannot explain the difference.
Find the original table, report or experiment. Record its date and method. A screenshot or a repeated claim may omit the inputs needed to reproduce the number.
Invented example
A fictional table counts 100 transaction-sending addresses in one seven-day window and 150 in the next, on the same network using the same rule.
(150 − 100) ÷ 100 × 100 = +50%
The count rose by 50 addresses, or 50% relative to the first window's 100 addresses.
An address count is not a count of people. The increase alone does not establish adoption or explain what caused the change.
Invented example
Two fictional model runs answer 40 and 60 of the same 100 test questions correctly under the same scoring rule.
60% − 40% = +20 percentage points; (60 − 40) ÷ 40 × 100 = +50% relative change
Accuracy increased from 40% to 60%: 20 percentage points, which is a 50% increase relative to 40%.
This describes those test runs. It does not establish performance on other tasks, a typical improvement across runs or a statistically reliable difference.
Follow a fictional 50% claim from its wording to its calculation and remaining questions.
A guide to reading the original argument and keeping different voices separate.