Why VCC cancels out

Two fractions of the same thing

The divider gives SIGNAL as a fraction of VCC. At 25 °C that fraction is one half, whatever VCC is.

An Uno's ADC measures the same way. Its reference is AVCC, which on an Uno is the same 5 V as the 5V pin, and analogRead returns SIGNAL as a fraction of it, out of 1023. So the count is the divider's fraction, and VCC appears nowhere:

count / 1023 = R_ntc / (R_ntc + 10 kΩ)

A measurement where the supply cancels out like this is called ratiometric. It is the reason the Uno is the easy case for this block.

Pick VCC (Uno) and move the slider. A USB port is often a little under 5 V. SIGNAL moves with it, but so does the top of the ADC's scale, and the count stays at 512. The temperature does not move.

The Pico works the same way, close enough: its ADC measures against the Pico's own 3.3 V, filtered on the board, and the block runs from the same 3.3 V.

Where it breaks

An ESP32 or ESP32-S3 measures against a reference inside the chip, not against VCC. Its readings are in volts, not in fractions of anything. To get back to the fraction, the sketch has to divide by what it believes VCC to be, and that is 3.3 V unless you tell it otherwise.

Pick its own reference (ESP32) and move the slider. With the 3V3 rail 1 % high, the sketch reports about half a degree low; 3 % low, about 1.4 °C high. The ADC is reading correctly. The sketch's assumption is what is wrong.

The fix is one number. Measure the 3V3 pin with a multimeter while the board is running and put it into the sketch as VCC_MV. Reading it on an ESP32 covers the rest of what is different there.