fix(pdf): keep positioned math out of tables

[asionesjia]

Confirmed #77: glyph runs that pdftotext -layout already has were turned into ragged GFM tables, and equation numbers / math lead-ins were promoted to headings. Formulas that are images or paths (empty in pdftotext) stay out of scope.

Changes

• Reject borderless grids that look like stacked math (operators, ∂/∑, equation numbers)
• Merge smaller, vertically offset script runs (hᵖ⁺¹, H⁺) without treating the next paragraph as subscripts
• Join d/dt and ∂L/∂q stacks into d/dt style fractions
• Do not heading-ize equation numbers, math lines, lowercase fragments, or colon lead-ins

Checked against the issue examples: Lagrange / Richardson formulas stay as text, (3.18) is not a heading, or inversely is not a heading, and real tables (year headers, P-value grids) still convert.

fixes #77

---

Note

Medium Risk
Heuristic changes affect table detection and heading/script merging across all PDFs; misclassification could hide real tables or alter typography, though tests target the reported regressions.

Overview
Improves PDF→markdown for scientific text so aligned math glyphs stay as prose instead of GFM tables, and display lines are less often turned into headings.

Text merging: Adds mergeStackedMath in the finish pipeline (after script merge) to join vertically stacked pairs like d/dt into d/dt. Superscript/subscript attachment now covers letters and operators (hᵖ⁺¹, H₂), with tighter font-size and vertical-offset rules so baseline hyphens and unrelated lines are not scripted.

Table detection: Borderless table acceptance rejects grids that looksLikeMathLayout (equation numbers, math operators/glyphs, short operator-heavy cells) while still allowing compact numeric tables with ±/−.

Headings: Skips isolated-line and header promotion for equation numbers, math-heavy lines, lowercase lead-ins, colon endings, and similar false positives.

Adds math.test.ts, table/math regression tests, and a small snapshot tweak for endnote superscript attachment.

^{Reviewed by Cursor Bugbot for commit d468c44. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
✅ Deployment successful! View logs	mdgate-demo	d468c44	Aug 27 2026, 04:57 AM

[cursor]

Stale comment

Left a non-blocking comment. Cursor Bugbot finished as skipped and reported 2 unresolved issues, so this was not approved. Human review is needed; no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: Minus-only equations become tables
  • The eqNums path now still treats − and ± as operators, so numbered minus-only formulas are rejected as tables while signed numeric grids stay tables.

_{You can send follow-ups to the cloud agent here.}

^{Reviewed by Cursor Bugbot for commit 91d20f3. Configure here.}

[cursor]

Stale comment

Left a non-blocking comment. Cursor Bugbot finished as skipped and left 1 unresolved issue on the current head, so this was not approved. Human review is needed; no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[cursor]

Approved. Cursor Bugbot passed on the current head and all prior findings are resolved, so no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}