Skip to content

Reconstruct captioned tables from segmented header rules - #197

Merged
silverstein merged 2 commits into
masterfrom
feat/continued-table-evidence-20260918
Sep 18, 2026
Merged

silverstein merged 2 commits into
masterfrom
feat/continued-table-evidence-20260918

Conversation

@silverstein

Copy link
Copy Markdown
Member

Outcome

Recover bounded multi-column Markdown tables when PDF ruling geometry explicitly defines the header and column intervals. Continuation pages must independently provide their own header; no cross-page stitching or inferred group values.

Safety and scope

  • Require a TABLE caption, aligned outer rules, contiguous header segments, unique source assignment and complete numeric rows.
  • Reject spanning or missing cells, competing rules, overprint, boundary crossings, missing headers and incomplete source evidence.
  • Preserve source/share parity, existing fallback warnings and exact source bindings; renderer contract advances to 1.21.0.

Verification

  • Stonebook macOS arm64, Node 26.5.0, one worker: 220 distinct targeted tests passed, 6 artifact-backed skips, zero failures.
  • Independent review passed exact head a6798fa.
  • Real MCP replay of the original public CDC PDF reconstructs two six-column tables with 30 and 25 numeric data rows. All four previously frozen sampled rows and visible footnote markers are retained.
  • Original source and first-run failures remain preserved. The PDF is now a development regression, not an untouched holdout. Overall conversion remains partial for unrelated vector, link and footnote topology limitations.

This is source integration, not a release or a comparative benchmark claim. Full CI remains the integration gate.

@silverstein
silverstein merged commit e24567c into master Sep 18, 2026
2 checks passed
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