CodeSOTA · Experiment · E-0006Portable protocol · public evidence graph
Why this run exists

MMLU protocol provenance audit

Partition registry results into comparable protocol families before making cross-model claims.

Experiment
E-0006
Status
running
Outcome
pending
Forks
1 public or local starts recorded
Owner
Kacper Wikiel

Method before metrics.

Read source notes and primary references, then group rows by benchmark variant, shot count, reasoning mode and model version.

Hypothesis
H-0008 · Shot count, reasoning mode and benchmark variant explain a material part of differences published as MMLU accuracy.

The comparison contract.

A reproduction matches this protocol. A fork changes it and declares the deviation.

dataset
mmlu
controlled Variables
benchmark variant · shot count · reasoning mode · model version
independent Variable
model
benchmarks
MMLU accuracy

What this experiment produced or evaluated.

This planned experiment has not produced public evidence yet.

Change the evidence, not just the discussion.

Start locally, inspect the portable manifest, and sign in only if you choose to publish.

Attach a run to this protocol.

Record the measurement and its provenance. New evidence is unverified and pending review until CodeSOTA checks it.

Stable, portable, attributable.

Stable ID
E-0006
Visibility
public
Created
2026-08-20
Updated
2026-08-20