Audited 37 PDF-related tools and corrected misleading capability claims, unsafe HTML previews, template-polluted localization, missing How steps, and stale keyword decisions.

Date: 2026-09-05 21:07 Summary: Audited 37 PDF-related tools and corrected misleading capability claims, unsafe HTML previews, template-polluted localization, missing How steps, and stale keyword decisions. Visibility: people

[question] 对 pdf工具进行 review和修正

[try to solve]

Context

The site had 37 PDF-related tools: 36 under the pdf-docs primary topic plus images-to-pdf. Existing mechanical SEO checks were green, but that did not establish that visible claims matched implementation, examples matched loadSample(), localized copy was natural, or similar URLs had distinct jobs.

The review therefore compared catalog metadata, page implementations, English and sampled localized shards, How/Rules/Example/FAQ sections, keyword records, and generated output. It also checked high-trust terms such as PDF/A compliance, digital signatures, and DWG conversion against what the browser code actually produced.

Process

Inventoried all 37 PDF-related catalog entries, page modules, i18n shards, generated files, briefs, and keyword records.
Reviewed the implementations for local-processing claims, sample behavior, output types, error paths, unsafe DOM writes, and feature overlap.
Audited the ten-tool PDF scene batch for repeated How steps, OCR FAQs copied onto unrelated tools, mixed-language description fragments, and generic examples.
Compared high-risk titles with actual outputs:
- The PDF/A page performed raw-byte heuristics, not ISO 19005 validation.
- The “digital signature” page drew a visible SHA-256 mark, not a PKCS#7/PAdES signature.
- The “DWG” page exported ASCII DXF text and page-border lines, not proprietary DWG.
- The flipbook page rendered one canvas page at a time and downloaded the original PDF.
Ran page wiring, coverage, SEO, full-site generation, and single-tool verification gates. When removal of template padding exposed description-length and example-keyword failures, the descriptions were expanded with real inputs, outputs, steps, and limitations instead of restoring SEO filler.

Root cause / analysis

The main content defect came from the one-shot PDF scene generator. It emitted one generic structure across unrelated jobs:

“Click the main action button” instead of the actual CTA.
An OCR/scanned-PDF failure FAQ even for tools that never extract text.
Mixed Spanish fragments such as El ejemplo al entrar and un resultado real inside other locales.
English descriptions that mentioned libraries and announced that the page contained a process or example instead of explaining the user outcome.

The SEO validator checked description length and selected words, but did not validate how_item_*, Example-to-sample consistency, localization quality, or whether a compliance-oriented title was supported by the implementation. Consequently, keyword padding could make the mechanical gate green while substantive information gain remained weak.

Three keyword intents also required a strict capability distinction. A visible integrity mark is not certificate signing; five raw-byte indicators are not PDF/A validation; and an ASCII DXF text export is not a DWG converter. These jobs remain deferred in the keyword funnel even though adjacent helper pages exist.

Solution

Capability and trust corrections

Renamed the PDF/A experience to “Inspect PDF/A indicators” across ten locales.
Replaced Pass/Fail with Found/Not found and added a prominent non-certification warning.
Replaced the counter-intuitive linearization check with five explicit indicators: PDF header, uncompressed pdfaid XMP marker, GTS_PDFA1 output intent, embedded-font stream marker, and absence of /Encrypt.
Replaced the OCR FAQ because the indicator screen does not require a text layer.
Linked to veraPDF validation and the PDF Association instead of presenting a PDF library as authority.
Fixed the PDF/A sample and clear buttons, which had passed the wrong arguments to the HTML escaping helper and could render language codes instead of labels.
Changed all locale H1s for the digital-signature page to describe a visible integrity mark while retaining the PKI/PAdES disclaimer.
Changed all locale H1s for the DWG page to describe PDF text export to AutoCAD-readable DXF.

Flipbook product correction

Added a short CSS page-turn animation to the live canvas preview.
Changed Download from the original PDF to a standalone pdf-flipbook.html.
Embedded the source PDF bytes in the HTML export and included Previous/Next rendering controls.
Documented that the exported HTML needs network access to load PDF.js, while the PDF content itself is not uploaded.
Rewrote the English How, Rules, Example, and FAQ around the actual output and removed the false scanned-PDF failure claim in all locales.

Security and implementation corrections

Sanitized pasted HTML with DOMPurify before it enters the HTML-to-PDF preview and renderer.
Sanitized Mammoth-generated DOCX HTML before preview and PDF conversion.
Escaped ampersands and angle brackets in extracted PDF text before building the comparison diff HTML.
Routed add-text initialization, sample, and clear behavior through the literal loadSample() entry point used by page wiring checks.
Updated stale Protect PDF copy that said Unlock PDF had not shipped.

Duplicate-intent correction

turn-pdf-into-editable-document and edit-pdf-text-online previously used nearly the same extraction-and-rebuild flow, and one FAQ admitted that the main difference was labels.
The editable-document tool now produces a separate editable UTF-8 .txt file and keeps rebuilt PDF as an optional secondary output.
Its H1 and FAQ were updated across all ten locales to emphasize an editable text document. The edit-text page remains focused on changing wording and returning a rebuilt PDF.

Information gain and localization

Removed mixed-language description pollution from 81 PDF scene locale shards.
Replaced false OCR FAQ pairs in 70 locale/tool combinations that do not depend on text extraction.
Replaced generic “main action button” steps with each page's actual localized CTA.
Added four localized ordered How steps to 11 older PDF tools that previously rendered only a paragraph.
Rewrote eight English scene-tool metadata sets around actual input, output, example, and boundary.
Expanded Chinese and Japanese descriptions with concrete workflow, sample, and limitation details so they pass the same content gate without keyword padding.

Planning and documentation alignment

Corrected the four high-risk tool briefs and locale review records.
Reconciled PDF keyword-pool N1–N22 entries:
- N1–N11, N13–N16, and N18–N21 now absorb into implemented tools.
- N12 certificate signing, N17 standards-based PDF/A validation, and N22 proprietary DWG/vector conversion remain deferred.
Updated the keyword tracker and bilingual README labels to match the corrected capabilities.

Verification

All 37 PDF-related pages passed lint:tool-page --require-html.
Global lint:seo passed after merging the final i18n shards.
Coverage gates passed for the four high-risk tools.
npm run build:site completed, rebuilding all 205 tools across ten languages and the sitemap.
verify:tool passed for make-pdf-flipbook, including coverage, page wiring, full build, HTML smoke, SEO, vendor, and isolation checks.
IDE lint reported no errors in the modified page and i18n sources.

Notes / boundaries

check-pdf-a-compliance keeps its existing URL for continuity, but the visible product is intentionally an indicator screen. It must not be marked as covering standards validation.
add-digital-signature-to-pdf keeps its URL but does not cover certificate signing, legal e-signatures, CMS, PKCS#7, or PAdES.
convert-pdf-to-dwg keeps its URL but produces .dxf containing text entities and page-border lines only.
The HTML flipbook is a portable client-side viewer, not a hosted publishing or analytics platform.
The full-site build also refreshes shared generated outputs. Unrelated existing working-tree changes were not reverted.

[actions]

Corrected PDF page implementations, catalog timestamps, and ten-locale i18n shards.
Updated four high-risk briefs and locale review histories.
Reconciled PDF keyword-pool and tracker decisions.
Updated bilingual README tool labels.
Rebuilt merged tool data, localized pages, home pages, and sitemap.

Audited 37 PDF-related tools and corrected misleading capability claims, unsafe HTML previews, template-polluted localization, missing How steps, and stale keyword decisions. | Dev Logs