Re: Notebook + Charter + Whitepaper Updates — offering a bind-at-source arm for the tax benchmark

Hello all,

I joined the CG in August and have been reading rather than posting. Ron's
note invited use cases and participation in the benchmarking white paper,
so here is a concrete offer and one question.

*Where I am coming from.* I work on the Semantic Data Charter, an open
specification for data that carries its meaning and its constraints bound
into the record at authoring time. Reading charter v2, the four-facet
decomposition is close to how we already split things, in different
vocabulary. What we bind maps onto Meaning, Structure and World fairly
directly, and DARK versus LIT names the failure I spend most of my time on:
an ontology link lights Meaning while Structure and World stay dark, so the
reference range, the rounding rule and the governing authority never cross
the boundary with the value.

To be clear about the relationship, because I think it matters: CGP is a
gauge and SDC is one bound source. SDC is not a context graph, and nothing
here proposes changing CGP or introducing a required data model.

*The offer: a third arm for the tax-preparation benchmark.*

   - Arm A, raw values with no bound context, all facets DARK.
   - Arm B, ontology-linked values, Meaning LIT, Structure and World DARK.
   - Arm C, SDC-bound components, Meaning, Structure, and World LIT at the
   source.

Measure Dark Fraction alongside the accuracy, cost, quality and latency
measures already planned, then stress every arm the same way by removing or
corrupting context: a wrong tax year, a changed rounding rule, a swapped
filing status. The hypothesis worth trying to falsify is that Arm C starts
at the lowest Dark Fraction and degrades least under that stress, because
the interpretive prerequisites were bound at the source rather than
reconstructed at the boundary. If it does not, that is a more interesting
result than if it does.

Concretely I would contribute the Form 1040 components for the benchmark's
task, openly licensed, and the adapter that turns a bound instance into the
canonical facet table. If time-to-first-usable-model is worth measuring
alongside accuracy and cost, I would like to see that included too, since
the authoring-time cost is the honest objection to any bind-at-source
approach and the benchmark is the right place to settle it rather than
argue about it.

*The question.* SDC emits SHACL alongside XSD 1.1, RDF, JSON-LD and W3C
PROV from a single component, so a bound instance should be straightforward
to decompose into a facet table without translation loss. Under charter v2,
which committee owns that adapter work? Format-level concerns look like
Serialization and Specification from the outside, but I would rather ask
than assume.

A one-page mapping of SDC onto the four facets, with a Form 1040
standard-deduction example and the three-arm proposal above, is here:
https://semanticdatacharter.com/W3C/context-graphs-mapping.html

Happy to bring this to whichever subcommittee is the right home, or to
leave it as a written input if that is more useful.

Tim Cook
Semantic Data Charter

Timothy W. Cook, CEO, Founder

-- 
Sent from Axius SDC <https://axius-sdc.com/>, Inc.