Skip to content

feat(serve): a ticker loads the published holon before EDGAR - #57

Merged
jfrench9 merged 2 commits into
mainfrom
feature/cdn-first-loading
Sep 13, 2026
Merged

feat(serve): a ticker loads the published holon before EDGAR#57
jfrench9 merged 2 commits into
mainfrom
feature/cdn-first-loading

Conversation

@jfrench9

Copy link
Copy Markdown
Member

What

Published first. The RoboSystems public data CDN writes every processed SEC filing as a holon beside the document as filed, under {year}/{cik}/{accession}/, with a per-filer catalog under companies/{ticker}.json. load_filing with a ticker or cik:accession now looks there before EDGAR:

  • the newest filing of the form, when it has a holon, loads from it — under a second for the model, the document as filed beside it so the text tools read the whole filing, a large text block's fragment fetched on load (in parallel; one that cannot be fetched stays a URL);
  • a filing that predates the artifacts falls back to EDGAR and Arelle, a minute cold and ten seconds warm;
  • the newest filing decides — an older one that happens to be published is never substituted.

XBRLKIT_ARTIFACTS_URL points elsewhere or, empty, turns the lookup off; XBRLKIT_FETCH_TEXT=false leaves fragments as URLs. Live: Intuit's FY2026 10-K (filed 2026-09-09) loads in 3.9 s with its document and all 90 text blocks inlined; NVIDIA's, filed before the artifacts, reports no holon and takes the EDGAR path.

Three fixes to what the holon carries, ahead of the corpus reprocess that rewrites every published one:

  • Calculations 1.1. The 2023 summation-item arcrole is the calculation linkbase as much as the 2003 one. Read as definition arcs, it left a filer that adopted it (Intuit, for one) with no calculation network and no footing anywhere. The parse, the property-graph projection and the holon reader all take both.
  • xbrldt:targetRole rides on the association, written and read back, so a hypercube declared across roles can be rebuilt with every axis and member.
  • The property-graph projection writes an element's type from item_type when there is no QName for it, so a type's namespace survives a second projection (salvaged from feat(deserialize): read the property graph back into the model #56).

Tests

tests/test_published.py stands a local server in for the CDN and a sentinel in for EDGAR: the catalog path, the manifest path, the document, fragment inlining and its off switch, a missing fragment, the newest-filing rule, and every fallback. Round-trip and projection tests extended for the three holon fixes.

🤖 Generated with Claude Code

https://claude.ai/code/session_019GMCxs82XWZHN8wZBC25RJ

The RoboSystems public data CDN writes every processed SEC filing as a
holon beside the document as filed, under {year}/{cik}/{accession}/,
with a per-filer catalog under companies/. A ticker or cik:accession now
looks there first: the newest filing of the form, when it has a holon,
loads from it — under a second for the model, the document beside it for
the text tools, a large text block's fragment fetched on load — and a
filing that predates the artifacts falls back to EDGAR and Arelle, a
minute cold and ten seconds warm. The newest filing decides; an older
one that happens to be published is never substituted.
XBRLKIT_ARTIFACTS_URL points elsewhere or, empty, turns the lookup off;
XBRLKIT_FETCH_TEXT=false leaves fragments as URLs.

Three things the holon carries better, ahead of the corpus reprocess
that rewrites every published one:

- Calculations 1.1. The 2023 summation-item arcrole is the calculation
  linkbase as much as the 2003 one; read as definition arcs it left a
  filer that adopted it with no calculation network and no footing.
- xbrldt:targetRole rides on the association, so a hypercube declared
  across roles can be rebuilt with every axis and member.
- The property-graph projection writes an element's type from item_type
  when there is no QName for it, so a type's namespace survives a second
  projection.

Claude-Session: https://claude.ai/code/session_019GMCxs82XWZHN8wZBC25RJ
The ClawDog round trip (#55, released as 0.14.0) and the published-first
loader both touched _load_json: main adds the clawdog kind to the served
set, this branch inlines text-block fragments on load. Both kept.

Claude-Session: https://claude.ai/code/session_01GD7d1Rjg5PPV8swJUDFdyr
@jfrench9
jfrench9 merged commit 1a73980 into main Sep 13, 2026
4 checks passed
@jfrench9
jfrench9 deleted the feature/cdn-first-loading branch September 13, 2026 18:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant