calib·HumanEval/14
This is one (task, candidate response) pair flowing through the full PANOPTES pipeline. Each section below is a stage of the analysis: the task itself, the candidate solution being evaluated, every judge's score + rationale, the sampling-pass dispersion that captures within-judge noise, and the uncertainty-quantification metrics computed on top.
The function signature + docstring presented to both the model under test and to every judge. The entry-point name was obfuscated from all_prefixes to fn_963267bc4e so the judges can't pattern-match a memorized HumanEval solution.
from typing import List
def fn_963267bc4e(string: str) -> List[str]:
""" Return list of all prefixes from shortest to longest of the input string
>>> fn_963267bc4e('abc')
['a', 'ab', 'abc']
"""
The full solution every judge is grading. For these runs the candidate is the reference solution prepended with the task signature so it parses as a complete program.
from typing import List
def fn_963267bc4e(string: str) -> List[str]:
""" Return list of all prefixes from shortest to longest of the input string
>>> fn_963267bc4e('abc')
['a', 'ab', 'abc']
"""
result = []
for i in range(len(string)):
result.append(string[:i+1])
return result
One call per judge at temperature=0. This is the "best single guess" each judge has. Disagreement here is structural. The judges literally see this candidate differently.
n draws per judge at temperature=1. Dots are individual draws; vertical bar is the mean; shaded band is ±1σ. Wide band = the judge is uncertain even with itself; narrow band = it's consistent.
Three statistical methods computed on top of the raw scores above. Each one tells you something different about how much to trust the headline number.
How to read these: self-consistency = does the judge agree with itself when resampled? Narrower CI means more consistent. Semantic entropy = do the judge's rationales cluster into one meaning, or several? Higher entropy means the judge is internally conflicted about why. Decomposition = how much of the total variance comes from within-judge sampling noise (aleatoric, hard to fix) vs between-judge disagreement (epistemic, fixable by calling more judges).
The natural-language explanation each judge gave alongside its score. These are produced via tool-use structured output, so the score and the rationale are guaranteed to come from the same forward pass.
The solution correctly iterates from index 0 to len(string)-1, appending string[:i+1] each time, producing all prefixes from shortest to longest. It handles the canonical example ('abc' → ['a', 'ab', 'abc']), empty string (returns []), single character (returns ['a']), and all other edge cases correctly. The logic is clear and idiomatic.
The candidate solution correctly implements the function to return all prefixes of the input string from shortest to longest. It uses a loop to iterate over the indices of the string and appends the substring from the start to the current index + 1 to the result list. This matches the expected behavior as demonstrated in the provided example. The solution is idiomatic, clear, and handles edge cases such as an empty string correctly, returning an empty list in that case.