Linking between data

Lexicon design and versioning

https://atproto.com/guides/lexicon#versioning says:

If a schema must change a previously-published constraint, it should be published as a new schema under a new NSID.

(annotated here: https://hyp.is/zLtkZKqoEfCN8-P0_SUCdg/atproto.com/guides/lexicon)
Paul Frazee (Bluesky CTO) has written this awesome guide:

Guidance on Authoring Lexicons
Lenses (https://github.com/lexicon-community/governance/issues/14) could help solve version changes in the future

Thoughts on endorsements / validations / reviews:

To me these feel like they're all the same thing.
I prefer "review" because "endorsement" implies a review which is positive, but we also need to support negative sentiment.
They can be applied to not just claims but also evaluations, measurements, evidence, media, ... even other reviews i.e. reviews of reviews (the "reviews / claims all the way down" idea).
In a review field lexicon, this can easily be modelled with a subject field which is a strongRef and can therefore point to any other entity. You could even do interesting things like review someone else’s profile, if you wanted to vouch for them or claim that they’re a scammer!
This could also support AI-generated reviews (e.g. from AI / bot identities) although we’d want a way to distinguish humans.
If we have a single unified approach to this, it effectively creates a web of trust which opens the door to interesting ways of doing computation on the graph to generate trust metrics, whether it’s simplistic approaches like https://maxflow.one/ or more sophisticated multi-dimensional ones.
Addition (holke), see https://www.figma.com/board/BMEkEvkszSMkPptSD8ZvMn/Hypercerts-strategy?node-id=840-1437&t=ARnfmkgLeUqHSX1p-1

Evidence: Any form of supporting material, such as texts, images, or other data sources.
Measurement: One or more quantitative observations reported with their respective units.