Using it
Reading an Answer
Every answer comes with what you need to check it yourself.
The status
| Card says | Meaning |
|---|---|
| ANSWERED | Every stage worked. The number is shown with its unit. |
| PARTIAL | A stage couldn't finish, so no number is shown. The card says which stage stopped and why, instead of guessing. |
| REFUSED | Not a physics question ("how much does the color blue weigh?"). You get the reason and, when there is one, a physics question you could ask instead. |
The parts of an answer
- The value and unit. Computed by Noether, never by a language model.
- Equations used. Each one by name and id, with a page in Equations giving its source and license.
- Inputs. Every value plugged in, and where it came from: given (you wrote it), constant (from Constants), or assumption (for example, "released from rest" means an initial speed of 0).
- Assumptions. Everything the answer depends on that you didn't say: no air resistance, constant acceleration, and so on.
- Caveats. Anything a check didn't like.
- Confidence. A label and a score from 0 to 1.
- Models. Which Fermi model handled each stage, and how many plans were tried.
Confidence
The score comes from a fixed, published formula, never from a model grading itself:
a = 1 / (1 + 0.25 * number_of_assumptions)
score = 0.35 * retrieval + 0.30 * dimensions + 0.20 * magnitude + 0.15 * aHigh is 0.75 or more, medium 0.45 or more, low below that. Some results are capped:
| Cap | When |
|---|---|
| 0.2 | The units don't work out, the result is negative where it can't be (a negative mass or resistance), or it is 0 only because the plan assumed a 0. |
| 0.4 | The question states a value the plan didn't use, or the question fits two look-alike equations and doesn't say which (series or parallel resistors). |
| 0.6 | A Fermi estimate: the inputs are educated guesses. |
A low-confidence answer comes with caveats saying why. Take it as "probably, but check this part". An impossible result (a negative resistance, say) is never shown at all: the answer becomes PARTIAL and says what went wrong.