docs: list .pptx/.xlsx as extracted formats; retire "no XLSX/PPTX extractor"

[thomas-villani]

What this changes, and why

#1117 (PPTX) and #1124 (XLSX) shipped in v1.0.0-rc.81, but the site and the fact-check gate still describe both formats as missing. Two published pages now make false claims:

• answers/search-word-documents-in-a-folder: "XERJ has no extractor for .xlsx, .pptx … and refuses a file in any of those formats".
• compare/xerj-vs-localsynapse: "there is no XLSX or PPTX extractor at all". The TL;DR, description, FAQ and "choose LocalSynapse" section all concede Excel workbooks on that basis.

Changes:

• scripts/seo/claims_rules.py: the RED "Excel / PowerPoint" THING row is split, the same way the mail row was for feat(autoindex): mbox and Google Takeout ingest #949:
  • "Excel workbooks (.xlsx)" is GREEN and cites extract/xlsx.rs. Its gate lists the known limits: one table per sheet, merged cells not expanded, and no value for a formula without a cached result.
  • "PowerPoint decks (.pptx)" is GREEN and cites extract/pptx.rs.
  • A new RED row covers what is still refused: legacy .xls/.ppt, .xlsb and OpenDocument.
  • The FC-THING-RED rewrite text no longer lists XLSX/PPTX as missing.
• Fixtures: bad_thing_red.md now targets legacy .xls. It still expects FC-THING-RED, which now comes from the new RED row. A new good_xlsx_answer.md covers the GREEN row.
• LocalSynapse comparison: rewritten, not just flipped. XERJ now reads workbooks as typed tables. LocalSynapse still keeps cell coordinates and merged ranges, so a workbook that depends on its layout still fits LocalSynapse better. The mail and desktop concessions are unchanged.
• README, llms.txt, llms-full.txt: the extractor lists now include PPTX and XLSX. In llms.txt only the "Then, from an unknown folder…" list changed. Line 3 is left alone on purpose: it already goes past the ~300-character first-screen limit (rule 7 in docs/research/llms-txt-2026-09). No other open PR edits llms.txt (checked before branching).
• landing/answers and landing/compare were regenerated with build_articles.py, and the sitemap was regenerated in a separate commit after the content commit (own-commit rule).

Evidence

factcheck.py --self-test       self-test OK: 56 rules, 55/37/32 tiered numbers, 34 THING rows
factcheck.py --fixture-check   58 TP, 0 FN, 0 FP; true negatives 12 (of 12); rules with no fixture: 0
factcheck.py --fail-on error   92 file(s), 0 ERROR, 93 WARN   (same 0 ERROR / 93 WARN as main)
build_articles.py --check      ok (189 files; 92 selected article(s))
gen_sitemap.py --check         ok (183 URLs)
fix_heads.py / fix_links.py / mk_og_card.py --check, test_article_schema.py: ok
landing-constants-guard.sh     all checks passed

Checks

• cargo fmt --all: N/A, no Rust changed
• Scoped release build: N/A
• cargo test -p <crate>: N/A
• ES-YAML conformance suite: not applicable, this change is docs/landing/fact-check-only
• New ES-compatible behavior YAML case: N/A
• Docs updated
• CONTRIBUTION_REVIEW audit: N/A, no operational document or runnable recipe changed

Not run:

• seo_lint.py fails locally with sitemap.stale, and it fails the same way on a clean main checkout. This Windows clone has core.autocrlf=true, so sitemap.xml is checked out with CRLF line endings. gen_sitemap.py --check passes.
• The first-time-agent harness (llms-txt rule 11) was not re-run, because this change only adds two names to an existing list.
• ste_check.py is advisory. On the LocalSynapse page it reports one fewer ERROR than on main. The remaining ERROR is the existing mail sentence.

Provenance

• Written by: AI agent (Claude Code, Claude Opus 5.5), run by @thomas-villani, who reviewed it. I also wrote feat(autoindex): extract PowerPoint decks (.pptx), one record per slide #1117 and feat(autoindex): extract Excel workbooks (.xlsx), one dataset per sheet (rebase of #1120) #1124.
• Verified: every command in Evidence, run on this branch. The format limits named in the gate and on the pages come from the doc comments in extract/xlsx.rs and extract/pptx.rs, which were tested in feat(autoindex): extract PowerPoint decks (.pptx), one record per slide #1117/feat(autoindex): extract Excel workbooks (.xlsx), one dataset per sheet (rebase of #1120) #1124.
• Assumed: the LocalSynapse claims (cell coordinates, merged ranges) are carried over unchanged from the page's existing evidence: sources and were not re-checked. Legacy .xls/.ppt (OLE compound files) have no sniff rule and were not run through autoindex here. The RED row says only "no extractor", which sniff.rs confirms: it has no CFB branch.

🤖 Generated with Claude Code

[thomas-villani]

Merge order: I've added commit docs: list man pages; describe XLSX merged-cell handling to this PR. It describes the code in #1132 (vertical merges filled down), #1133 (two-row grouped headers → Q1_Jan) and #1134 (man pages) as shipped. Please merge this PR after those three. I put it here rather than in a new PR because only one open PR should edit landing/llms.txt at a time.

Until #1134 lands, factcheck.py --fixture-check reports one FC-EV-DANGLING on the new good_man_answer.md, because it cites extract/man.rs. That error also flags a merge in the wrong order. With man.rs from #1134 present it reports 0 false positives and 13/13 good fixtures clean. The other gates (--self-test, --fail-on error, build_articles --check, gen_sitemap --check, landing-constants-guard) pass on the branch as it is.

This comment and commit were written by an AI agent (Claude Code) on behalf of @thomas-villani.