Follow-up to #280, deliberately left out of it so that PR could ship without paying another review round for a section that is already true but incomplete.
The finding
runs/cloud-20260916/markdown/fema-proof-of-loss.md — 26 of the form's 124 non-blank lines are a bare $ and nothing else. Every amount field on the FEMA proof-of-loss form comes back as a naked dollar sign with its label stripped. No line in the file contains a currency amount attached to a label.
Why it matters more than the artefacts already documented
The README's "What the conversion gets wrong" section currently covers the table behaviour — spanning headers expanded across every cell, stacked tables welded into one, cells collapsed to space-joined values — plus code-block flattening, hyphen closing, and the FEMA heading-to-field separation. All true, and all about tables.
This one is a better illustration for two reasons. A table with a repeated header still reads as a table, so a reader can talk themselves out of caring. A column of 26 bare $ symbols cannot be read as anything but broken. And it is a form, the category that is hardest for any converter, because the relationship between a label and its field is purely spatial.
It is also the cleanest available proof that the checks test presence rather than structure:
contains:AMOUNTS CLAIMED passes, because the heading survives at line 9.
- Every field beneath that heading is unlabelled.
Both statements are true of the same recorded run, which scored 25 of 25.
Proposed change
Add it to the "What the conversion gets wrong" section in examples/document-to-markdown/README.md, in the paragraph that currently covers the FEMA heading-to-field separation, and extend the closing paragraph about _table_count to note that contains: checks confirm a heading is present and say nothing about the fields under it.
No artifact changes: the recorded run already contains this, and nothing needs re-recording.
Already covered elsewhere
The equivalent note is going onto the FEMA card on the sie-web /doc-to-markdown page in the same batch as the verbatim-excerpt fix, so a reader meets it at the point they actually look at the form. This issue is only about the example's own README.
Follow-up to #280, deliberately left out of it so that PR could ship without paying another review round for a section that is already true but incomplete.
The finding
runs/cloud-20260916/markdown/fema-proof-of-loss.md— 26 of the form's 124 non-blank lines are a bare$and nothing else. Every amount field on the FEMA proof-of-loss form comes back as a naked dollar sign with its label stripped. No line in the file contains a currency amount attached to a label.Why it matters more than the artefacts already documented
The README's "What the conversion gets wrong" section currently covers the table behaviour — spanning headers expanded across every cell, stacked tables welded into one, cells collapsed to space-joined values — plus code-block flattening, hyphen closing, and the FEMA heading-to-field separation. All true, and all about tables.
This one is a better illustration for two reasons. A table with a repeated header still reads as a table, so a reader can talk themselves out of caring. A column of 26 bare
$symbols cannot be read as anything but broken. And it is a form, the category that is hardest for any converter, because the relationship between a label and its field is purely spatial.It is also the cleanest available proof that the checks test presence rather than structure:
contains:AMOUNTS CLAIMEDpasses, because the heading survives at line 9.Both statements are true of the same recorded run, which scored 25 of 25.
Proposed change
Add it to the "What the conversion gets wrong" section in
examples/document-to-markdown/README.md, in the paragraph that currently covers the FEMA heading-to-field separation, and extend the closing paragraph about_table_countto note thatcontains:checks confirm a heading is present and say nothing about the fields under it.No artifact changes: the recorded run already contains this, and nothing needs re-recording.
Already covered elsewhere
The equivalent note is going onto the FEMA card on the sie-web
/doc-to-markdownpage in the same batch as the verbatim-excerpt fix, so a reader meets it at the point they actually look at the form. This issue is only about the example's own README.