Preserve authored blockquote typography when lowering rich-text article wrappers

[chubes4]

Reproduction

On https://www.rugliocooks.com/post/welcome-to-my-kitchen-hamburger-with-sweet-potato-fries-an-in-depth-conversation/ the portable DLA 0.7.2 capture includes an authored <blockquote class="G16EV"> wrapping a Seneca quotation. Playwright at 1440px computes its text as italic 24px / 33.6px in an approximately 722px wide quote, rendered on two lines (top y=5929, h=67). The imported WordPress 7.1.2 page with SSI main 73a01a310a88 paired to BE header candidate 1dbda3192 retains the words and attribution but lowers the <blockquote> into nested RichText spans/divs; text becomes italic 16px / 24px, one line (top y=5908, h=24), with no blockquote element. The article actions footer begins at y=6091 in the capture versus y=6006 in WordPress, an 85px upstream shift. The comments section is independently 37px shorter, producing a total 122px document-height deficit (6704 vs 6582). Three article images before the quote match captured y and dimensions exactly, localizing this loss to the quote and following chrome rather than capture or hero layout.

Acceptance

A neutral article fixture with a styled blockquote, italic quotation and bold attribution retains authored quote typography, wrapping and block-axis size through editable Gutenberg (prefer a native Quote block when the source structure is compatible). Verify before/after rendered geometry at 390/768/1440, keep the quote text and attribution editable, and ensure the footer following the quote starts at the authored Y. Treat the separate comments height drift as a follow-up if it has a different owner.

AI disclosure: OpenAI GPT-6 Sol via OpenCode compared source/capture/WordPress DOM and computed font/geometry and filed this finding under Chris Huber direction.

[chubes4]

Regression-shape correction after inspecting the first candidate: A bare <article><blockquote><p>…</p><cite>…</cite></blockquote></article> already emits core/quote on current trunk; a test of that case passes without changing runtime code and does not exercise Ruglio. The actual retained source is a rich-text wrapper <div class="WkCxM"><blockquote class="G16EV"><span><div class="fUDW- lqKvG" data-ricos-paragraph="…"><span class="KmEWy"><em><span>“It is not …”</span></em><span>— </span><strong><span>Seneca</span></strong></span></div></span></blockquote></div>; Gutenberg output replaces the <blockquote> with a span and retains the text in a div, losing the quote’s 24px/33.6px class-owned font (imported 16px/24px). Capture quote y=5929 h=67, import y=5908 h=24; article actions footer begins y=6091 vs 6006. Fix must handle this nested rich-text shape through artifact/page lowering and CSS projection, retain authored attribution, and fail before the fix. A fixture exercising only a standalone clean blockquote must remain evidence, not the proposed resolution. The separate comments surface is another 37px shorter.

AI disclosure: OpenAI GPT-6 Sol via OpenCode reviewed the provider patch and compared captured/imported DOM and computed font geometry; Chris Huber directs parity acceptance.

[chubes4]

Second candidate rejected at integration ownership boundary. Its patch changes only packages/blocks-engine/src/theme/preserve-dom/strategy.ts and its Vitest; Studio/SSI studio create --from paired package compiles through php-transformer (the release ZIP carries PHP transformer), not the TS preserveDomStrategy. A TS quote/group assertion therefore cannot affect the real Ruglio WordPress page; source quote still loses the <blockquote> and 24px font in the generated PHP-produced blocks. No PHP implementation change or paired import evidence was included. Fix must run through php-transformer artifact/site-plan path, with an exact nested RichText regression that fails before runtime change and a fresh Studio import verifying quote 24px/33.6px and the following footer y=6091. The previous bare-PHP-quote fixture and this TS preserve-dom fixture both test paths already able to retain quotes; neither is a resolution.

AI disclosure: OpenAI GPT-6 Sol via OpenCode reviewed the candidate diff against the proven Studio import pipeline and rejected it pending PHP-path proof; Chris Huber directs the loop.

[chubes4]

Fresh paired-import rejection of PHP QuotePattern candidate (BE snapshot d95b8b0e0b35 paired with SSI zero-fallback form candidate; all eight pages imported, quality_pass=true). At 1440px Playwright still measures captured quote blockquote.G16EV, italic 24px / 33.6px, y5929/h67 and actions footer y6091. WordPress still emits spans/divs with NO <blockquote>, italic 16px / 24px, y5908/h24 and footer y6006; document height remains 6582 vs captured 6704. All three preceding article images have identical Y/dimensions. The candidate changed php-transformer/src/HtmlToBlocks/Patterns/QuotePattern.php to regex-match <span><div data-ricos-paragraph=…> and added an ArtifactCompiler fixture, but the real full-site route never reaches that conversion shape. The regex also keys on vendor-specific data-ricos-paragraph inside generic BE, contrary to provider-neutral ownership. Next diagnosis must inspect the exact post/page reconstruction path and pre-recognizer node classification that lowers the containing <div class=WkCxM> and its quote to RichText spans; add a failing full-route artifact fixture and a generic fix. No PR for this candidate.

AI disclosure: OpenAI GPT-6 Sol via OpenCode built the paired ZIP, ran Studio and Playwright against the unchanged capture, reviewed the PHP candidate and drafted this evidence under Chris Huber direction.

[chubes4]

Owning boundary found after the failed QuotePattern candidate: WP-CLI parse_blocks() on the materialized blog post (page/post ID 12) shows the entire post body containing “Seneca” lives in two generated ssi-*/gallery-* companion blocks, each with one giant attrs.content raw HTML payload. The relevant desktop block is gallery-db325f946271fcab5891ad85e1d4bd4fc70ac0eb978761f2a756e04b034b16cb. It is NOT a tree of native article/quote blocks. Browser source ancestry: blockquote.G16EV -> div.WkCxM -> section[data-hook=post-description] -> article[data-hook=post] -> div[data-hook=post-page] -> post-page-root/TPA multi-section. Materialized browser still renders the article text/images because the gallery companion outputs raw content, but the blockquote is invalidly nested/reparsed and its class-owned 24px typography is lost. The PHP QuotePattern candidate ran on a standalone fixture, whereas the real site plan has already selected a gallery companion for the entire app article; changing QuotePattern cannot affect it. This is a producer classification/editability boundary: a media-rich blog/article widget must not be swallowed wholesale by gallery fallback when its headings, prose, quote and post chrome are semantically extractable. Add a neutral full-route fixture with article + multiple images + nested quote inside an app-shell-like custom component; verify block inventory contains editable article text/quote and bounded companion islands only for irreducible interactions. Keep before/after 24px/33.6px quote and actions footer y6091 browser proof. No PR for off-path candidate.

AI disclosure: OpenAI GPT-6 Sol via OpenCode parsed materialized Gutenberg blocks with WP-CLI, walked source/candidate DOM ancestry in Playwright, reproduced the paired import, and identified this upstream ownership boundary under Chris Huber direction.

[chubes4]

Fresh paired eight-page import of semantic-article candidate 4863654546b0 (with SSI zero-fallback form candidate) proves editability gain but fails visual parity. WP-CLI now counts native core/quote (2 responsive variants), 93 core/paragraph, 20 core/heading, 6 core/image on the post, and ZERO whole-article gallery-* companions; the quote is in a native Quote subtree and its text computes the correct italic 24px / 33.6px. However the source/capture article image wells had images x430 width740 and heights 655, 610, 555 at y515,2466,4503. Imported core images expand to article width 938 with heights 830, 773, 704 at y501,2196,3938. The article-actions footer moves from captured y6091 to y5317 and page height from 6704 to 5930 (-774px). The quote appears y5236 instead of captured y5923 because preceding flow diverged. The new containsEditorialFlowContent() guard prevents the opaque gallery capture but does not carry the authored nested image container widths/positioning (or all flow geometry) into editable core blocks. A regression must assert source-owned image widths and section Y at 390/768/1440 and avoid a global veto that trades visual parity for editability. Keep the generic semantic-extraction direction; no PR until both editability and 1:1 geometry pass.

AI disclosure: OpenAI GPT-6 Sol via OpenCode built a paired dev package, ran Studio, counted post blocks with WP-CLI and measured image/quote/footer boxes in Playwright against the retained source; Chris Huber directs parity acceptance.

[chubes4]

More precise image-width localization for the article candidate: the first source/captured image is inside #viewer-oya7l1 > .HUNH- > figure.vscz7 > .KDEWi, with every one of those wrappers x=350, width=740px inside a 938px-wide article. The source image is absolutely positioned and width:100% relative to this bounded well, h=655. After native-block lowering, the same wrappers are editable core/group but each computes x=251,width=938px; image stays width:100% and grows to h=830, taking the whole article column. Source widths appear runtime-resolved rather than a direct CSS width rule on the examined selectors; the source --dim-width/image aspect-ratio evidence and capture geometry sidecars may be the proper generic signal. Restore the finite, centered image-well geometry onto the owning editable Groups at desktop/tablet while retaining fluid phone behavior; do not set 740px by site-specific constant. The full article candidate therefore fixes block inventory and quote font but FAILS the required rendered page parity as measured above.

AI disclosure: OpenAI GPT-6 Sol via OpenCode walked the exact source/WordPress image ancestor boxes and computed styles in Playwright and documented the generic width-model gap under Chris Huber direction.