7 7 {"run":16,"commit":"bf8645a","metric":35,"metrics":{},"status":"keep","description":"Iteration 16: Fixed extract-headings.lua to emit a real nested BulletList AST instead of escaped literal Markdown paragraphs. The filter now preserves heading hierarchy, creates actual internal links, handles skipped heading levels, and returns empty output for documents without headings. Verified Markdown list output, HTML nested ul/li output, link hrefs, and 9/9 filters pass.","timestamp":1780158253729,"segment":0,"confidence":null,"asi":{"hypothesis":"Fix existing filter correctness: escaped Markdown output is misleading; agents need real outline links/lists","bug_fixed":"extract-headings produced literal escaped '- [Heading](#id)' paragraphs instead of list/link AST","validation":"nested markdown list, HTML ul/li, internal hrefs, skipped levels, no-heading input, 9/9 filters","changed_file":"references/filters/extract-headings.lua"}}
8 8 {"run":17,"commit":"1d7d3c1","metric":36,"metrics":{},"status":"keep","description":"Iteration 17: Fixed center-images.lua LaTeX corruption bug. The prior Image/Plain approach double-wrapped images inside Figure, producing malformed minipage/center LaTeX. Rewrote filter to return explicit filter tables, use topdown traversal, skip Figure children for LaTeX/Beamer, center only image-only paragraphs, and leave inline images unchanged. Verified clean LaTeX figures, centered no-alt image paragraphs, unchanged inline images, HTML styling, and 9/9 filters pass.","timestamp":1780158416994,"segment":0,"confidence":null,"asi":{"hypothesis":"Continue fixing existing filters for real correctness; center-images had subtle LaTeX corruption for captioned figures","bug_fixed":"LaTeX Figure children were being centered, corrupting figure output","validation":"captioned figure clean, image-only paragraph centered, inline image unchanged, HTML style present, 9/9 filters","changed_file":"references/filters/center-images.lua"}}
9 9 {"run":18,"commit":"5694044","metric":37,"metrics":{},"status":"keep","description":"Iteration 18: Improved pandoc-doctor.lua accuracy for PDF diagnostics. Doctor no longer misleadingly reports pdf as normal reader/writer; it now labels PDF as writer via engine and runs real smoke tests for available PDF engines. This catches cases where engine binaries exist but PDF generation fails (verified: typst/wkhtmltopdf version OK but smoke FAIL in current env). Updated SKILL.md helper script note. 9/9 filters pass.","timestamp":1780158545607,"segment":0,"confidence":null,"asi":{"hypothesis":"Doctor version checks alone were misleading; real PDF smoke tests better diagnose agent PDF failures","bug_fixed":"pdf shown as -- format despite extension output support; engine binary OK did not mean pandoc PDF works","validation":"doctor prints PDF smoke tests, catches real failures, markdown HTML smoke OK, 9/9 filters","changed_file":"scripts/pandoc-doctor.lua"}}
10 +{"run":19,"commit":"14eda31","metric":38,"metrics":{},"status":"keep","description":"Iteration 19: Improved pandoc-md-to-pdf.lua with real PDF engine auto-detection. If PANDOC_PDF_ENGINE is unset, it smoke-tests xelatex/lualatex/pdflatex/typst/wkhtmltopdf/weasyprint and selects the first working engine; if none work, prints actionable diagnostics instead of blindly failing on xelatex. Preserves env override behavior. Updated SKILL wording. Verified no-engine diagnostic, forced-engine path, and 9/9 filters pass.","timestamp":1780158690706,"segment":0,"confidence":null,"asi":{"hypothesis":"The PDF helper should use doctor-style smoke testing instead of hardcoded xelatex; agents need actionable PDF engine diagnostics","changed_file":"scripts/pandoc-md-to-pdf.lua","validation":"auto-detect no-engine diagnostic, env override, SKILL refs, 9/9 filters","note":"Current env has no working PDF engine; script now reports all attempts clearly"}}