[errata] The κ bands in the reading list are quoted as if they were a standard

[akash-coded]

Errata. The material is wrong, or says something more confidently than it should.
Confirmed errata become issues.

What the material says

The judge-calibration section presents the Landis–Koch κ bands — slight / fair / moderate /
substantial / almost perfect — as though they were an established threshold, with 0.6 as a pass
mark.

Why I think that is wrong

Landis and Koch (1977) proposed those labels for biomedical inter-rater studies and said
themselves that the divisions were arbitrary. They are a convention, not a standard, and they are
widely criticised for exactly the reason this repository cares about: κ is sensitive to
prevalence, so a fixed band means different things at different class balances.

Quoting 0.6 as a pass mark is the same category of mistake as quoting a recall number without its
corpus — which the README's own third column exists to prevent.

What I think it should say

Report κ with the class balance next to it, use the bands as rough language rather than a
gate, and where a threshold is actually needed, derive it from the decision the judge is feeding
rather than from a 1977 convention.

Where

docs/30-learning/interview-prep/01-mathematical-foundations/cohens-kappa.md, and the notebook 06
callout.

[akash-coded]

Dana · reviewer · owns the eval harness

Confirmed, and this is a fair hit on our own standard.

The derivation page already says "Landis and Koch's bands are conventions, not theorems. Say so
if you quote them" — so the page is fine. The notebook 06 callout is not; it presents 0.6 as
a gate without that qualification, which is the inconsistency you have found.

Two changes, and one non-change:

• Notebook 06 callout gets the prevalence caveat and stops naming 0.6 as a pass mark.
• Wherever κ is reported, the class balance is reported beside it.
• Not removing the bands. They are useful shared language for talking about a κ out loud;
  what is wrong is treating them as a threshold.

Opening this as a docs issue with the thread linked. Good catch — the failure mode here is
exactly the one the repository exists to warn about, applied to itself.