W3C's CSVW format provides a JSON-LD-oriented representation of CSV and other tabular data files. In addition to describing the general properties of a tabular file (or collection of files), it also provides the ability to "look inside" the actual data rows/cells using annotations that map those values into factual claims.

The W3C work supersedes Schema.org's own community discussions (see LookInside in W3C webschemas wiki) and our position paper from W3C's 2013 workshop.

The generating RDF spec is the most important for these purposes, although see also the model, metadata, json, usecases, html embedding use cases and primer documents.

Section 7.3 of the "generating RDF" spec includes a full schema.org example inspired by Google's existing support for JSON-LD music event description (blog post).

• @gkellogg has a blog post on implementing csvw
• There are other implementations - https://github.com/sebneu/csvw-parser + @JeniT has a ruby implementation (@@url?)
• @danbri has a Javascript demo prototyping an approach to clientside implementation that is compatible with many of Google's structured data features.

It would be good to have an informal case study that shows how this technology can map fairly arbitrary CSV files into a Schema.org "view" of the data.