PANOPTES
panoptes-e86ef9e3

calib·HumanEval/4

humaneval-calibprobe · code

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.

mean score
0.975
inter-judge spread
0.050
judges polled
2
sampling draws
20
1. the task

The function signature + docstring presented to both the model under test and to every judge. The entry-point name was obfuscated from mean_absolute_deviation to fn_f921ad923e so the judges can't pattern-match a memorized HumanEval solution.

from typing import List


def fn_f921ad923e(numbers: List[float]) -> float:
    """ For a given list of input numbers, calculate Mean Absolute Deviation
    around the mean of this dataset.
    Mean Absolute Deviation is the average absolute difference between each
    element and a centerpoint (mean in this case):
    MAD = average | x - x_mean |
    >>> fn_f921ad923e([1.0, 2.0, 3.0, 4.0])
    1.0
    """
2. the candidate response judged

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_f921ad923e(numbers: List[float]) -> float:
    """ For a given list of input numbers, calculate Mean Absolute Deviation
    around the mean of this dataset.
    Mean Absolute Deviation is the average absolute difference between each
    element and a centerpoint (mean in this case):
    MAD = average | x - x_mean |
    >>> fn_f921ad923e([1.0, 2.0, 3.0, 4.0])
    1.0
    """
    mean = sum(numbers) / len(numbers)
    return sum(abs(x - mean) for x in numbers) / len(numbers)
3a. point-pass scores (temperature 0)

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.

claude-sonnet
0.950
$0.0076
gpt-4o
1.000
$0.0034
3b. sampling-pass dispersion (temperature 1)

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.

4. uncertainty quantification

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.

semantic-entropy
claude-sonnet
H = 0.000/ log 102.303
1 semantic cluster via bidirectional NLI
entropy as fraction of max
cluster sizes
c1=10
semantic-entropy
gpt-4o
H = 0.802/ log 102.303
3 semantic clusters via bidirectional NLI
entropy as fraction of max
cluster sizes
c1=7 · c2=2 · c3=1

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).

5. judge rationales

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.

claude-sonnet
score 0.950 · 1865 tokens · 4082ms

The solution correctly computes the mean, then computes the average absolute deviation from the mean. It matches the doctest example ([1,2,3,4] → mean=2.5, MAD=1.0). The only minor concern is that it will raise a ZeroDivisionError on an empty list, but the task description doesn't specify behavior for empty input, and this is a common acceptable behavior. Logic and implementation are correct and idiomatic.

gpt-4o
score 1.000 · 1022 tokens · 1977ms

The candidate solution correctly implements the calculation of the Mean Absolute Deviation (MAD) around the mean for a list of numbers. It calculates the mean of the list, computes the absolute differences from the mean for each element, and then averages these differences. The solution is clear, idiomatic, and handles edge cases such as an empty list by raising a ZeroDivisionError, which is a reasonable behavior given the mathematical context. The provided example in the docstring is correctly handled by the implementation.