Repository navigation
v0.21.0
Spreadsheets are published
(ADR-046, proposed): a
Google Sheet, an Excel workbook or a CSV file in a published folder gets a
page with each visible sheet as tables, and a sheet with formulas says how it
calculates.
Added
- Spreadsheet pages.
sourceType: drive-sheet, markedSheetin the
sidebar,sheetindata/docs-index.json,llms.txtand the MCP server's
format. A Google Sheet is exported as.xlsx; an uploaded.xlsx,
.xlsm,.csvor.tsvfile is downloaded, and not again while Drive's
checksum is unchanged. Hidden sheets, rows and columns are left out, and
the page never names a hidden sheet. A CSV file may be UTF-8, UTF-16 or
Windows-1251. - "How it is calculated" under a sheet with formulas: each distinct
formula once, the cells it uses named by their row and column labels or the
workbook's defined names, then its inputs and results. - A spreadsheet's tables hold their header, filter and sort on its page.
- Content health lists a spreadsheet cut short (
sheet-truncated), one
with charts or images the page does not show (sheet-not-shown), and a long
one (sheet-long). The sync report'ssummary.publishedgainssheets. - The manifest records
exportMode: sheetandsheetVersion. numfmt3.2.6 is a dependency of the sync package, for Excel's number
formats.
Changed
- The Markdown version writes tables without padding, and splits a table
longer than about 3,000 characters into parts that each repeat its header,
so a passage an assistant reads still says what each column is. Every
page's Markdown, and so each document's stored copy and its measured length,
changes once. - A link to a Google Sheet in the corpus leads to its page, and one
outside it is noted like a link to a Google Doc outside it. .xlsand.odsfiles are listed as "Spreadsheets in an older format",
with the advice to save them as Google Sheets.
Upgrade note
Bump the packages and the workflow pins, then run a sync. Every spreadsheet
in the published folders gets a page on that sync, and every page's Markdown
version is rewritten once for its tables, so the MCP server stores every
document again on its next schedule.
Spreadsheets are published to everyone their folder's rule names. Before the
first sync, tell editors that a spreadsheet in a published folder is now on
the site, and that hiding a sheet, a row or a column, or moving the file out
of the published folders, keeps it off.
A platform older than this cannot read a manifest that records a spreadsheet.