Shipped batch-convert-jpg-to-text-with-ocr with queue, skip, ZIP, combined TXT, ten-locale copy, and a back-link from the single-image OCR page.

Date: 2026-09-08 19:45 Summary: Shipped batch-convert-jpg-to-text-with-ocr with queue, skip, ZIP, combined TXT, ten-locale copy, and a back-link from the single-image OCR page. Visibility: people

[question] 继续

Tool links (English)

batch-convert-jpg-to-text-with-ocr convert-a-jpg-to-text-with-ocr

[try to solve]

Context

The brief in work-tasks/batch-convert-jpg-to-text-with-ocr/ was already ready. The job is not a second “JPG to text” URL: N1 (convert-a-jpg-to-text-with-ocr) keeps the single-photo query. This page must carry extra Information Gain as controls — a multi-file queue, Convert all, skip a failed row, Download ZIP of {stem}.txt plus default combined.txt — or it would be a doorway.

Prior session wrote catalog, icon, batchConvertJpgToTextWithOcrPage.ts, and five locales (en, zh, es, ja, ar). This session finished the remaining five locales, linked N1 back, ran ship gates, and recorded what still could not be clicked in a live browser.

Process

Confirmed the page already serializes one Tesseract worker (MAX_FILES = 10), does not auto-loadSample() on landing, and exposes Convert all / Download ZIP / Copy combined text.
Wrote independent pt / id / fr / ru / de shards from 03-locale-briefs.md (local H1: Converter vários JPG… / Ubah banyak JPG… / Convertir plusieurs JPG… / Пакетно преобразовать JPG… / Mehrere JPG mit OCR…). How steps use the same button labels as the UI in that language.
Re-scanned sample locales en, zh, es, ja for coverage pass 2b: master H1 stays “Batch convert JPG to text with OCR”; secondary “batch OCR / bulk OCR” (and local equivalents) sit after Steps so the snippet keeps Convert all / ZIP. Lengthened the Chinese _description after lint:seo warned it was 118 characters (need ≥120).
Cross-tool (CROSS_TOOL_UPDATE=1): N1 catalog related now leads with this slug; all ten N1 locales gained FAQ q8/a8 naming the batch H1 (not the slug). README English and Chinese tool lists each got one line next to N1.
Gates: coverage:gate --phase=4 and --phase=all green; merge:tools; build:site --full; lint:tool-page --require-html; lint:seo; lint:vendor; isolation with CROSS_TOOL_UPDATE=1. First verify:tool failed only on the short zh description; after the two-character fix the SEO step was re-run green and the two OCR slugs were re-prerendered.
Static HTML check (not a click-through): English prerender contains Convert all, Download ZIP, Copy combined text, function loadSample, multiple, combined.txt. Chinese contains 全部转换 / 下载 ZIP / 复制合并文本. N1 English FAQ asks “Can I convert several JPGs at once?” and names the batch H1. Home / and /zh/ already list the new card. Sitemap includes the English URL.

Root cause / analysis

A batch OCR URL only earns a separate indexable page if the extra work is on the first screen. Copy that says “also works for several files” while the form stays one <input> would fail both the brief and Google doorway / scaled-content rules. Serial WASM is a product constraint, not a copy claim: parallel workers would compete for memory on phones.

N1 used to say “one image per run — not batch ocr,” which became false once this page existed. FAQ q8 points people to the queue/ZIP page by tool name, so crawlers and humans see a pair instead of two titles fighting jpg to text.

Solution

Catalog: page.style: "opts", localProcessing: true, related N1 + extract-text-from-pdf.
Status: 03 → i18n-done, 02 → implemented.
Engine reuse: same /vendor/tesseract/ as N1; ZIP via /vendor/fflate/index.js.
Browser MCP was not available. Local wrangler on port 8788 still served pre-upload R2 HTML (404 for the new slug). Verification used prerendered public/_pages/{lang}/tools/*.html and home HTML, not an interactive Convert all / ZIP download.

Notes / boundaries

Still not a scanned-PDF, Word, CSV, handwriting, or shop-scan product.
jpg to text stays on N1; this H1 is the batch task sentence, not “Batch OCR” alone.
Clicking Load sample → two real OCR lines → ZIP enablement was not exercised in a browser this session.
Not deployed; npm run deploy / production upload:r2 were not run.

[actions]

Added pt/id/fr/ru/de i18n shards; N1 FAQ q8 + related; README tool-list lines
03 i18n-done / 02 implemented; coverage 2b named en,zh,es,ja
npm run coverage:gate --phase=4; lint:seo; lint:tool-isolation with CROSS_TOOL_UPDATE=1; prerender of both OCR slugs