Check implementation backlog

The implementation now includes explicit source and bibliography transformations, independent document packages, review/reporting controls, and additional bounded PDF, metadata and structure checks. These additions close specific absence claims; they do not complete the full requirements. Broader syntax/geometry support, image preparation, richer PDF review output and platform/tool validation remain engineering work. Interpretation of scientific meaning, accessibility adequacy and anonymity still requires human judgment.

This inventory reconciles the original requirements (a historical design record) and the original check ideas with the checker implementations. The catalogue lists every registered diagnostic code, including subchecks and preparation guards, and is generated from the registered rules. The number of codes is not a denominator for completed requirements; the grouped rows below also have different sizes and must not be used to compute a completion percentage. No code is reserved here for an unimplemented check.

How to read this backlog

• Implemented means the narrowly stated behavior exists, within the stated boundary. It does not mean every related requirement or platform is covered.

• Partial means related behavior exists, but the row names the missing part.

• Not implemented means no corresponding checker or operation currently exists.

• Unverified in a row means implementation exists but the named evidence or live-platform validation is missing; it is not an implementation claim.

• Intentionally excluded means it must not become an implicit product feature.

• Next prioritizes remaining first-release requirements; Extended covers deeper analysis; Optional requires explicit selection and, for remote services, explicit data-sharing authorization. None means no pending work for the narrow row or an intentional exclusion. These are planning priorities, not new release commitments.

All numerical limits, required fields, package policies and identity terms must come from explicit user settings. An inventory is not a policy pass. Deterministic observations and derived measurements must remain separate from heuristic suspicions and semantic advice. Missing tools or uncertain interpretation must produce an incomplete result, not a pass. Current settings are defined in Settings, ManuscriptOptions, PdfInspectionOptions, and the typed options in the additional *_checks.py modules. The additional-check example shows the nested tables. No template registry, package policy or publisher limit is maintained by the application. Online checks require both selection and network permission; unavailable providers are not passes.

The preparation options and example describe the new opt-in operations. rendered_pdf_artwork applies to the final document; included_figure_artwork applies separately to input PDFs. Regions and sample patches belong to each inspected PDF’s own coordinate system. Final-document coordinates are not reused for figures.

Concrete remaining engineering includes image optimization/externalized-figure adapters, annotated PDF output, broader literal source forms and drawing/clip/mask coverage, OCR/automatic figure-table segmentation, Windows isolation and a web interface. Linux Bubblewrap, qpdf and MuPDF live validation is incomplete on the macOS development host. These gaps are separate from semantic questions that no current check answers, such as reference identity, declaration adequacy and natural reading order.

Compilation, templates and source structure

Status	Priority	Requirement and remaining behavior	Current related behavior
Implemented	None	Clean build, final-log errors/rerun requests, missing glyph and font warnings; reject stale or partial PDFs.	BLD001–BLD009; fresh build directories, bounded latexmk, complete log/recorder requirements. Missing-character evidence comes from logs, not visual glyph recognition. See build evidence.
Partial	Next	Dependency completeness for every explicitly selected document; extend literal adapters without guessing execution context.	TEX005–TEX007 and recorder checks cover each selected package; TEX008 is an advisory naming the file TeX’s lookup order selects. With a complete build trace, remaining lexical uncertainty is a package.dependencies warning, while flattening and source transformations still need a complete literal graph. Literal import/subimport aliases, constrained subfiles, selected data readers and explicit local font files now resolve and rewrite paths. Dynamic/custom readers, inferred font faces, legacy subfiles preambles and content-dependent subfile checks remain unsupported or inconclusive.
Implemented	None	Apply a user-supplied overfull-box tolerance and report the threshold comparison.	BLD101 compares supported numerical final-log overflow with an explicit tolerance in TeX points. Incomplete logs/builds cannot pass; BLD006/BLD007 retain the underlying advisories.
Implemented	None	Check expected literal document class and required/forbidden class options.	TEX101, TEX102; indirect, conditional or macro-generated declarations are inconclusive. See manuscript evidence.
Partial	Extended	Inventory actual loaded classes/packages/options and identify existing local shadow copies.	BLD301 records actual normalized/raw kernel option registers after execution, including forwarded options, and matches origins to the recorder. Defaults, options consumed without recorded registers, and later package setup semantics remain outside that inventory. BLD104 checks local-shadow candidates through sandboxed system lookup; PKG003 protects flattening from creating a new shadow.
Partial	Extended	Compare supplied class/template files for integrity and version compatibility against a user-provided reference.	PKG107 compares explicit local project/reference pairs byte-for-byte and reports supported declared-version agreement. BLD103 checks explicit minimum header dates for loaded packages. Semantic compatibility, authenticity and official/outdated status are not inferred.
Implemented	None	Enforce package deny and allow policies over a clearly stated loaded-package scope.	TEX103 and MAN001 cover direct literal declarations; BLD102 applies explicit allow/deny/required sets to recorder-observed transitive .sty/.cls inputs. Their scopes remain separate, and incomplete recorder evidence is inconclusive. No policy lists are maintained.
Partial	Extended	Diagnose draft leftovers and suspicious spacing/layout manipulation: negative spacing, manual page breaks, shrinking, forced floats, margin/text-width/line-spacing overrides.	TEX001/TEX002 detect markers/editing commands (\color and \textcolor are not editing commands; local .sty/.cls files are not scanned); MAN002 adds literal spacing, page-break, shrinking, forced-float and layout-override candidates. Supported literal examples/comments are excluded. These advisories do not infer intent or discover every template-specific dummy field.
Implemented	None	Require a nonempty supported abstract; apply explicit abstract-word and keyword limits.	TEX104–TEX106 use documented lexical counting, not rendered word counts. Unknown macros and ambiguous declarations are inconclusive.
Partial	Extended	Detect abstract citations and apply an explicitly requested abbreviation policy.	MAN003 enforces an explicit no-citation policy for supported literal abstracts; MAN004 applies selected forbid/define heuristics to uppercase abbreviation candidates. Contextual abbreviation meaning and arbitrary declaration forms remain unverified.
Partial	Extended	Heading depth, configured numbering, apparently empty sections, and hard-coded reference numbers.	TEX107 and MAN005–MAN007 check literal source headings and candidates. PDF401 matches supplied title/number text on explicit PDF pages. Automatic rendered-heading detection, appendix/counter semantics and generated headings remain outside this interpretation.
Partial	Extended	Unique labels and resolved references with generated/custom syntax support where feasible.	TEX003/TEX004 scan literal labels/references as heuristics; BLD002/BLD003 consume TeX diagnostics. No general macro expansion or proof of reference semantics.

Pages, geometry and fonts

Status	Priority	Requirement and remaining behavior	Current related behavior
Implemented	None	Configured total page budget, expected page dimensions and consistent dimensions.	PDF001, PDF101, PDF102; total pages include references and appendices. Page boxes are inventoried. See PDF evidence.
Implemented	None	Section-specific page budgets, with explicit treatment of references, appendices and supplements.	PDF201 checks author-supplied inclusive page ranges against explicit section budgets. Out-of-document ranges are inconclusive. Section boundaries, reference sections and supplements are not inferred.
Partial	Extended	Detect unintended spillover/leaking pages, near-empty pages and a blank final page.	PDF205 measures bounded grayscale ink coverage against a supplied threshold/background and explicit exemptions. CMP001 compares page counts. Blank/spillover candidates remain advisory; intent and image-only/scan classification are not established.
Partial	Extended	Printable-region and column bounds, including rotated/cropped pages and legitimate full-width material.	PDF203 checks extractable text against explicit allowed regions; PDF304 adds affine image and straight filled-path bounds after supported rectangular clipping. Arbitrary rotations/crops, curves, strokes, masks and complete content bounds remain unmeasured or inconclusive.
Partial	Extended	Overflowing equations, figures, tables and captions; overlap and clipping candidates.	PDF203/PDF204 report text bounds and overlap candidates. PDF303/PDF305 add boundary-ink and supported image/polygon clip crossings. Semantic object attribution, complex clips/masks and intentional-cropping decisions remain unresolved.
Implemented	None	Configured embedded-font and Type 3 policies for font resources reported in the final PDF.	PDF104, PDF105. Zero reported fonts cannot establish embedding coverage of outlined/image-only text. Type 3 is a font type, not proof of bitmap glyphs.
Partial	Extended	Distinguish bitmap painting from vector painting in Type 3 glyph programs.	PDF308 inspects supported reachable declared Type 3 programs and nested Form XObjects, with an explicit bitmap policy. Opaque or unsupported programs are inconclusive; actual glyph-use coverage and every PDF operator/compositing form are not established. The adapter needs qpdf; live validation remains outstanding.
Partial	Extended	Minimum rendered manuscript, equation, caption, table and figure-label text sizes.	PDF202 checks reported extractable-text sizes against global or explicitly located regional minima. Integer rounding, partial region overlap, absent text and unsupported geometry are inconclusive. Semantic text categories, outlined/raster lettering and OCR remain unimplemented.

Figures and tables

Status	Priority	Requirement and remaining behavior	Current related behavior
Implemented	None	Check both effective raster-resolution axes at each placement reported by Poppler against an explicit minimum.	PDF103 retains repeated placements and reports unmeasured DPI. It consumes Poppler’s placement-scale measurements, including masks; it does not map them to source figure labels. See PDF evidence.
Partial	Extended	Classify observed artwork composition and apply supplied category thresholds.	PDF306 inventories raster/vector/mixed drawing operations for a selected PDF; text and backgrounds count. PDF307 applies explicit category DPI thresholds to repeated placements, including shear/rotation sampling. Automatic figure segmentation, category inference and final-page-to-source attribution remain unimplemented; empty/unsupported evidence is inconclusive.
Partial	Extended	Inspect included PDF figure fonts and attribute final-PDF font problems to an input figure when supported.	PDF212 separately inspects supplied input PDF figures for explicit embedding/Type 3 policies and retains source paths. Zero reported fonts is inconclusive. Reliable mapping of final-PDF font failures back to a particular included figure is not established.
Partial	Extended	Measure excessive figure whitespace/borders and possible cropping at supplied boundaries.	PDF302 measures rendered outer ink-box whitespace under an explicit background/threshold; PDF303 reports boundary-band ink. Regions are supplied for final pages or use input-PDF whole pages. Automatic figure boundaries, arbitrary page frames and intent are not inferred; edge ink does not prove clipping.
Partial	Extended	Check rendered line weight, including scaled figure strokes.	PDF213 covers supported uniform transforms; PDF309 measures straight-segment normal thickness after nonuniform affine transforms. Curves/text need uniform transforms. Hairlines, caps, joins, filled outlines and raster lines remain outside measured coverage.
Partial	Extended	Inspect configured color encoding and measurable foreground/background contrast.	PDF214 applies a rendering color-space allowlist; PDF215 uses an explicit background assumption. PDF312 compares actual composited RGB pixels in supplied foreground/background patches with explicit homogeneity tolerance. Automatic patch selection, general text/graphic attribution, source encoding and perceptual adequacy remain unresolved.
Partial	Extended	Figure caption/label presence, label placement, unreferenced figures and first-reference order.	MAN011–MAN015 check supported source float captions, labels, label placement, external references and first-reference order. Custom counters/subfloats may be inconclusive; rendered float order and arbitrary template-generated floats remain unverified.
Partial	Extended	Table caption/label presence, label placement, references and first-reference order.	MAN011–MAN015 also check supported table environments, separately ordering table references. Complex/custom float structures and rendered order remain outside the source model.
Partial	Extended	Table overflow, minimum rendered font size and rasterized-table candidates.	PDF202/PDF203 check explicit table text regions. PDF310 measures union coverage of axis-aligned raster placements in selected regions and reports candidates with text evidence. Automatic table identification, OCR and raster/outlined lettering measurements remain unimplemented; raster coverage does not diagnose a table or scan.

Bibliography and citation correctness

Status	Priority	Requirement and remaining behavior	Current related behavior
Implemented	None	Local syntax, duplicate keys, repeated fields, undefined strings and supported relationship targets.	BIB001–BIB005; bounded scanner preserves unknown fields and leaves damaged files unchanged. Incomplete parse coverage is explicit. See bibliography evidence.
Implemented	None	Resolve used citation keys before compilation as well as after it.	BIB101 resolves supported literal citation keys against bibliography resources declared by the selected graph, or explicitly selected all-root graphs. Dynamic/incomplete evidence is inconclusive; BLD001 remains the independent compilation check.
Partial	Next	Detect uncited entries and offer a conservative cited-only bibliography.	BIB102 reports usage candidates in one finding that lists every entry; opt-in BIB203 proposes cited-only resources while retaining supported nocite/alias/relationship/set closure and declarations. Incomplete/dynamic scope, subfile content, filters, missing keys and ambiguous relationships block pruning. Arbitrary-TeX usage proof remains outside scope.
Partial	Extended	Distinguish exact duplicate identifiers from likely duplicate works.	BIB007/BIB002 report exact identifiers; BIB108 adds bounded title-similarity candidates. Explicit BIB204 merge mappings retain chosen target metadata and redirect supported uses atomically. Work identity, conflict resolution and deciding which record to retain are not inferred.
Implemented	None	Validate supported literal DOI syntax and offer visible DOI-prefix normalization.	BIB006 plus normalize_dois; neither performs DOI resolution nor verifies that the DOI denotes the cited work. Macros and ambiguous markup remain unchanged.
Implemented	None	Entry-type/user-configured required fields; missing title/authors/year/venue; page-range and URL syntax; capitalization protection.	BIB103 applies explicit per-type requirements and field alternatives; BIB105/BIB106 check supported page-range/URL syntax. Optional BIB107 flags capitalization-protection candidates without editing braces or values. Macros, inheritance and unfamiliar syntax remain inconclusive.
Implemented	None	Report missing DOI fields when applicable.	BIB104 reports missing/empty DOI fields for explicitly selected entry types as advisories. It does not assert that a DOI exists or that references without DOIs are invalid.
Partial	Optional	Resolve DOIs and compare title, authors, year, venue and version with authoritative metadata.	NET001 resolves selected literal DOIs; NET002 compares supported fields with Crossref metadata. Resolution, provider agreement and cited-version identity remain separate. Requests are bounded and explicitly authorized; provider coverage and unsupported metadata remain visible.
Partial	Optional	Suggest missing DOI candidates and published versions of preprints; merge only after explicit review.	NET003/NET004 surface Crossref candidates with ambiguity. BIB204 can apply a user-supplied donor-to-target mapping; it does not turn a remote candidate into an authorized match or merge. Version identity and replacement decisions remain unverified.
Implemented	Optional	Check reference URLs and replication/artifact links.	NET006/NET007 make explicitly selected bounded public-network requests for literal reference/source URLs and supplied artifact links. Redirects, authentication, throttling and temporary failures remain distinct. Availability does not establish artifact completeness or reproducibility.
Partial	Optional	Surface authoritative correction/retraction notices.	NET005 surfaces attributed Crossref update metadata and related notices, distinguishing correction/retraction types. Absence or agreement from one provider is not exhaustive evidence; broader provider coverage and human interpretation remain necessary.
Partial	Next	Bibliography formatting, explicit field removal, cited-only output, reviewed key/merge mappings, guarded metadata edits and optional ordering.	BIB201–BIB206 implement copy-only proposals for the documented literal subset, preserving unaffected values/comments/macros and rejecting partial/ambiguous plans. Citation/key/relationship edits are atomic. Automatic metadata correction, work-identity decisions, dynamic citation expansion and arbitrary backend ordering semantics remain unsupported; preparation rebuilds and compares rendered output.

Author metadata, declarations and anonymity

Status	Priority	Requirement and remaining behavior	Current related behavior
Partial	Next	Required author, affiliation, email and corresponding-author fields.	MAN008 checks supported literal metadata presence. MAN201 checks each sequential preamble author record against explicitly selected associated commands. Shared-field association, arbitrary template grammars, generated metadata and identity/ownership remain unverified.
Implemented	None	ORCID syntax and checksum.	MAN009 validates supported literal ORCID commands with the ISO 7064 MOD 11-2 checksum. Syntax/checksum success does not establish identity or ownership.
Partial	Next	Required title/author/publication/rights metadata, source-to-PDF agreement and explicit sanitization.	PDF206/PDF207 and source matching check supported properties. PRV101 proposes exact expected-before literal source metadata replacements; PRV102 rechecks selected rebuilt-PDF properties and removed values. Generated/ambiguous declarations, binary-object sanitization and publication/rights semantics remain outside coverage.
Partial	Extended	Presence of Data Availability, competing interests, funding, CRediT/author contributions, AI-use and ethics declarations.	MAN010 requires literal content under user-named headings or supported declaration commands; TEX107 also checks heading presence. Template-specific/generated declaration forms, applicability and statement adequacy are not inferred.
Not implemented	Optional	Semantic adequacy of declarations or consistency with the study.	Human review is required; optional advice needs explicit selection. Do not infer that ethics/AI-use statements are applicable or generate their contents.
Partial	Extended	Anonymous-review scans of names/affiliations, acknowledgements, comments, filenames and metadata.	PRV001–PRV003 scan configured terms and identifying-language candidates. PRV001/PRV002 match whole words unless identity_term_matching = "substring". PRV103 adds bounded PNG text/EXIF and pre-scan JPEG EXIF/XMP/extended XMP/IPTC/comment inspection; MakerNote and other unsupported blocks are recorded as limitations. Unsupported image formats/metadata, raster text, indirect clues and semantic anonymity remain unverified; image metadata is not rewritten.
Partial	Extended	Identifying artifact URLs, institution/domain hints, self-identifying language and cross-document leaks.	Configured PRV001 terms match literal/URL-decoded identities across supplied text files and document roots. PRV003 separates heuristic self-identifying phrases from direct matches. Arbitrary institutional inference and semantic cross-document anonymity are not established.
Partial	Next	Required separate deliverables: title page, highlights, graphical abstract, cover/response letter, supplements and replication link.	PKG104 checks explicit paths and file/PDF/ZIP/text evidence. workflow.documents and PKG301 assign independent source packages and verify each archive rebuild. Content adequacy, actual separate delivery and unselected roots remain unverified; noncompiled deliverables are not forced through TeX.

Accessibility

Status	Priority	Requirement and remaining behavior	Current related behavior
Implemented	None	Required nonempty figure descriptions, including supported Description markup, and obvious placeholders.	MAN016 requires nonempty supported literal Description commands per figure and flags obvious placeholders. Unknown/generated content stays inconclusive. This is a source-presence check, not description-quality or PDF-accessibility validation.
Not implemented	Optional	Description quality and adequacy.	Requires human judgment or explicitly enabled, labeled advice with data-sharing consent; source presence cannot prove quality.
Partial	Extended	Document language, PDF tagging, reading order, accessible links and table structure.	PDF208/PDF209 inspect tagging/language. PDF402–PDF405 check supported declared table/link/Figure structure, Alt presence, parent links and MCID/backlink consistency. Complete stream coverage, natural reading order, meaningful headers/names/descriptions and PDF/UA conformance remain unverified. Object adapters require qpdf and are not live-validated on this host.
Partial	Next	Searchable/readable text and possible scan-only or rasterized regions.	pdf.text measures page extraction; PDF310 adds explicit-region raster-coverage candidates with text evidence. No OCR or reliable semantic scan classification exists. Image-only, outlined and blank pages are not treated as equivalent.
Partial	Extended	Color-only encoding checks, measurable contrast, and grayscale previews.	PDF215/PDF216 retain nominal-color checks; PDF312 measures explicit rendered foreground/background samples. PDF311 exports selected bounded grayscale previews outside the source archive. Color-only meaning, automatic sample/region choice, raster palette semantics and perceptual separability still require further analysis or review.

Submission bundle and PDF object checks

Status	Priority	Requirement and remaining behavior	Current related behavior
Partial	Next	Remove known debris and select needed project inputs without guessing uncertain dependencies.	Preliminary plan_cleanup preserves observed dependencies, unknown files and .bbl files, and removes OS/editor debris and rebuildable auxiliary files. Preparation separately selects inputs from complete build traces and the selected literal dependency graph, retaining files listed in [package] include (and, for compatibility, required deliverables and template files) and omitting unrelated files only in staging. A missing or incomplete build trace blocks packaging; an incomplete literal graph alone is advisory. Optional PKG106 remains an advisory inventory. No global minimality or all-root unused-file proof is claimed.
Partial	Next	Comment/private-note cleanup and unfinished-source checks.	TEX001/TEX002 and optional PRV005 inspect draft material. Opt-in TEX201 proposes removal of selected comment bodies in .tex/.ltx/.latex files only, retaining percent signs, line endings, literals, directives and license blocks. Dynamic lexical behavior and unclosed regions block proposals. Active editing-command removal and arbitrary TeX semantics remain unsupported.
Implemented	None	Reject unsafe archive paths/links and supported literal external source paths.	Import rejects links, special files, absolute paths, traversal and normalized collisions; names that are only nonportable (`<>:”
Implemented	None	Explain source constructs that require shell escape before building.	Optional advisory PRV006 flags supported literal write18/shell-execution constructs before building. Runtime still prohibits shell escape and confines processes; the lexical preflight does not replace enforcement.
Implemented	None	Scan likely secrets/private tokens with redacted findings and false-positive handling.	Optional heuristic PRV004 scans supported local text, comments, bibliography notes, filenames and supplied PDF evidence for bounded credential patterns. Candidate values are redacted, false positives require review, and unscanned binary content remains outside scope.
Implemented	None	Configurable allowed filename characters/lengths/types and final ZIP size.	PKG101–PKG103 apply explicit basename character-count, literal character-set and final-extension policies. PKG105 applies separate max_archive_bytes to actual compressed output. Existing import limits and PKG001–PKG004 flattening guards remain in force; extensions do not prove file formats.
Implemented	None	Deterministic ZIP and fresh exact-extraction rebuild for each explicitly selected source package.	create_archive, run_job and PKG301 verify each package independently, including flat-layout checks. Multi-document output is released only when every selected package passes required gates. Unselected roots and external delivery are outside that result.
Implemented	None	Configured absence of PDF encryption/password protection, forms, JavaScript and embedded attachments as reported by current tools.	PDF106–PDF109; inaccessible/unrecognized evidence is inconclusive. These checks are limited to pdfinfo/pdfdetach reporting, not exhaustive object inspection.
Partial	Extended	Inspect annotations and arbitrary active content/media.	PDF210 inspects resolved page-annotation subtypes against an explicit forbidden list; PDF211 scans reachable dictionaries for forbidden action-name candidates. Opaque streams, orphan objects, arbitrary media and exhaustive active-content absence remain unverified.
Partial	Next	Detect located crop/registration-mark candidates against explicit trim boundaries.	PDF301 identifies supported paired corner strokes and outside-trim registration crosses, with explicit thresholds and exemptions. Raster/outlined marks, unsupported transforms and intent remain unmeasured; candidates stay advisory and do not establish universal crop-mark detection.

Preparation and workflow backlog, outside the check-code count

Status	Priority	Requirement and remaining behavior	Current related behavior
Partial	Next	Automatic flat hierarchy and command-aware rewriting for a documented literal subset.	plan_flatten adds literal import/subimport aliases, constrained subfiles, selected data/font paths and PKG201 explicit filename overrides to existing inputs/graphics/bibliography/class/style handling. Protected identities and ambiguous/dynamic contexts block unsafe maps. Broader readers and source execution semantics remain engineering work.
Implemented	None	Explicit tex-fmt check/diff/application, project exclusions, protected regions and blank-line policies for supported source.	FMT001/FMT002 retain isolated proposals/idempotence; FMT101–FMT103 verify protected bytes, report exclusion scope and enforce formatter/preserve/collapse blank-line policies. The installed formatter exposes no blank-line flag, so this policy is adapter-side. Custom literal semantics remain outside the subset; final PDF comparison is still required.
Implemented	None	Assign explicit document roots and independent supplementary source packages; rebuild each solely from its archive.	workflow.documents materializes each selected directory/file set with its own root, optional engine and reference PDF. PKG301 requires all selected packages to verify before output release. Selection is explicit; no automatic assignment or content-adequacy claim is made.
Partial	Next	Validate engines and explicit bibliography backends on restricted platform runtimes.	pdfLaTeX/BibTeX workflows and XeLaTeX/LuaLaTeX smoke builds pass live on macOS. BLD201/BLD202 validate fresh bibliography controls and required tool evidence. On macOS, Biber runs from a runtime-owned, sealed single-architecture copy (see Biber on macOS) and passes live, including a realistic biblatex project; no unconfined fallback exists. Linux Bubblewrap has controlled-command tests but no live Linux run. Windows isolation is unimplemented.
Implemented	None	Compare baseline, prepared, archive-rebuilt and explicitly supplied PDFs separately.	CMP001–CMP003 compare page count, extracted text and RGB renders at 144 DPI. Supplied-PDF, repeated-baseline and archive comparisons remain exact; explicit tolerances/reviewed changes apply only to baseline versus prepared. Unavailable required evidence blocks preparation.
Partial	Next	Baseline stability, explicit comparison tolerances and reviewed expected-output changes.	CMP101 checks 2–5 configured independent baselines when repeats are requested. Explicit channel/pixel tolerances and eligible scoped exceptions apply only to baseline versus prepared; archive/reference comparisons cannot be relaxed. Inspection accepts up to 5000 pages, but renders (comparisons and rendered measurements) remain bounded to 300 pages/16 million pixels per page and do not prove semantic equivalence, PDF internals or subpixel fidelity.
Implemented	None	Scoped review controls, standalone PDF inspection and explicitly unverified diagnostic exports.	RPT101–RPT103 cover unmatched/ineligible decisions and diagnostic export; CLI pdf check, exact scoped suppressions/accepted exceptions with reasons and original evidence retention exist. Missing evidence, build/isolation failures and archive mismatches cannot be waived. Accepted exceptions remain a distinct outcome.
Partial	Extended	Offline HTML reports, located-region diagrams, source before/after views and CI annotations; richer PDF review remains.	Escaped HTML, source diffs, SVG coordinate diagrams, local PDF page links, CI annotations and bounded embedded before/after PDF pages with difference overlays exist. Annotated PDF output remains unimplemented. Preview omissions and capture failures do not change comparison findings. Reports and diagnostics remain outside source archives.
Not implemented	Optional	Web interface using the same core, with service-wide budgets and isolated cancellation.	CLI only. This is an interface/workflow item, not an additional check. See the separate parallel execution plan for scheduling work.
Partial	Optional	Reviewed source merging, bibliography inlining, image optimization and externalized-figure preparation.	TEX202 merges bounded standalone literal input lines; TEX203 inlines an explicitly selected standalone BibTeX thebibliography file. Both propose copies/diffs under rebuild comparison; include/import merging and biblatex inlining remain unsupported. Image optimization and externalized-figure preparation still have no adapters and must remain opt-in.
Intentionally excluded	None	Maintained publisher/venue/year profiles, package catalogs, template registries and scraped submission policies.	Users provide current constraints/reference files; generic checks must not claim publisher certification.
Intentionally excluded	None	Implicit online checks/LLM advice, manuscript rewriting, scientific fact checking, venue upload and guaranteed acceptance/anonymity/accessibility.	Optional remote/advice modules require explicit opt-in; human decisions and unsupported semantic claims cannot become deterministic passes.

Evidence map and verification boundary

The entries below identify inspected implementations and behavioral tests supporting the implemented portions. They do not claim a fresh full-suite or live-tool pass from this documentation audit. Controlled-output unit tests establish adapter behavior; live tests are a separate gate and require an operational sandbox and tools. Absence statements above were checked against the implementation modules and configuration schema, not inferred from an unassigned code.

Additional-check evidence

The catalogue links each new code to intended-behavior tests, including valid, violating and unavailable/ambiguous evidence. The implementation/test pairs are:

Scope	Implementation	Behavioral tests
Bibliography use and fields	bibliography_checks.py	test_bibliography_checks.py
Build tolerance and package policy	build_checks.py	test_build_checks.py
Actual loaded option registers	loaded_options.py	test_loaded_options.py
Manuscript, metadata and floats	manuscript_checks.py	test_manuscript_checks.py
Privacy and submission policies	submission_checks.py	test_submission_checks.py
PDF measurements and object policies	pdf_checks.py	test_pdf_checks.py
Optional online references	online_checks.py	test_online_checks.py
Source transformations	source_transform.py	test_source_transform.py
Bibliography transformations	bibliography_transform.py	test_bibliography_transform.py
Artwork geometry, pixels and glyph programs	pdf_artwork.py, pdf_artwork_options.py, pdf_artwork_trace.py, pdf_artwork_drawings.py, pdf_printer_marks.py, pdf_type3_glyphs.py, pdf_artwork_rendered.py	test_pdf_artwork.py
Metadata edits and image metadata	metadata_privacy.py, metadata_source_edits.py, image_metadata.py	test_metadata_privacy.py
Author records	author_records.py	test_structure_checks.py
PDF structure	pdf_structure_checks.py	test_structure_checks.py
Review decisions and offline outputs	review_policy.py, html_report.py, ci_annotations.py, diagnostic_bundle.py	test_review_policy.py, test_reporting.py, test_diagnostic_bundle.py
Independent packages and repeated baselines	documents.py, document_run.py, workflow_options.py	test_multi_document.py, test_workflow_review.py
Linux command construction and bibliography backend	sandbox.py, latexmk_build.py	test_runtime_backends.py

Online tests use fake transports, not live provider requests. Tool-output fixtures exercise Poppler/qpdf/MuPDF interpretation without claiming that each optional tool is installed or works on every PDF. Missing qpdf/mutool, unsupported object encoding, geometry uncertainty and unmeasured text cannot become successful checks. These component evidence boundaries also apply to the rows above.

Current macOS live smoke evidence covers two independent flat source packages with selected bibliography formatting/pruning, private-comment removal, input merging, repeated baselines and exact archive rebuilds. Separate literal-import, subfile and input-merge fixtures passed page-count, extracted-text and exact 144-DPI render comparisons with unchanged originals. Actual kernel-option instrumentation passed against an uninstrumented pdfLaTeX build; XeLaTeX and LuaLaTeX option/build smoke tests also passed. This does not validate all options or combinations. qpdf and MuPDF are not installed, and Linux isolation has not run on Linux. The live Biber test and a realistic biblatex/Biber preparation test pass on macOS with the sealed Biber copy; they run only with FLEDGE_RUN_INTEGRATION=1.

Build evidence

build_project and log_findings are exercised by BuildTests: test_failed_tool_never_accepts_partial_pdf, test_final_rerun_request_blocks_success, test_build_copy_has_fresh_output_and_safe_main, test_external_recorder_dependency_blocks_success, and test_build_log_check_codes_match_diagnostics. RuntimeTests.test_missing_backend_never_runs_command covers unavailable isolation. LiveIntegrationTests has separate test_confined_process_cannot_read_or_write_outside_its_workspace and test_confined_process_cannot_connect_to_network tests.

Source evidence

SourceGraph and plan_flatten are exercised by the source tests (test_source_analysis.py, test_source_search_paths.py, test_flat_layout.py and test_source_commands.py): test_markers_labels_and_refs_are_heuristics_not_blockers, test_nested_paths_and_collisions_are_rewritten_not_globally_replaced, test_dynamic_and_unsupported_commands_are_flat_blockers, test_relocated_style_cannot_shadow_system_style, test_sidecar_relationship_breakage_is_blocked, and test_flattening_is_idempotent_and_keeps_utf8_bom_crlf. New context/adapter cases include test_import_contexts_resolve_nested_inputs_and_restore_outer_paths, test_literal_subfile_body_and_parent_reference_are_flattened, test_literal_reader_adapters_preserve_nonfile_arguments, test_subfiles_bibliography_path_and_legacy_mode, and test_filename_overrides. The separate source-transformation tests cover original-byte retention, significant-percent/newline preservation, comment-wrapper retention, atomic blocked plans, recursive input merging and supported/unsupported bibliography inlining.

Manuscript evidence

check_manuscript is exercised by ManuscriptTests: test_expected_document_class_matches_and_mismatch_blocks, test_class_options_report_missing_and_forbidden_values, test_forbidden_packages_check_direct_local_graph_declarations, test_required_abstract_rejects_missing_and_empty_content, test_abstract_word_limits_use_documented_math_and_markup_counting, test_keyword_limits_count_phrases_and_common_environments, test_required_sections_match_literal_headings_not_body_claims, and test_conditional_and_macro_generated_declarations_are_inconclusive.

PDF evidence

inspect_pdf is exercised by PdfConstraintTests: test_page_limit_boundary_and_overflow, test_expected_dimensions_measure_every_page_with_pdf_point_tolerance, test_page_consistency_uses_full_range_not_just_first_page, test_image_resolution_checks_both_axes_and_repeated_placements, test_font_embedding_violation_changes_status_and_zero_fonts_are_inconclusive, test_forbidden_type3_font_policy_detects_only_type3, test_encryption_policy_uses_explicit_reported_feature, test_forms_policy_detects_acroform_and_xfa, test_javascript_policy_requires_explicit_reported_value, and test_attachments_policy_counts_embedded_files. test_unconfigured_constraints_do_not_report_compliance and test_partial_text_extraction_is_not_a_readability_pass guard evidence boundaries.

Bibliography evidence

check_bibliography_integrity and normalize_dois are exercised by BibliographyTests: test_duplicate_keys_and_dois_include_conflicts_without_merging, test_literal_relationships_and_builtin_months_are_checked, test_invalid_plain_dois_and_ambiguous_markup_are_distinct, test_doi_case_is_not_collapsed_for_exact_duplicate_detection, test_malformed_file_is_never_partially_normalized, and test_prefix_edits_preserve_braces_comments_case_and_unknown_fields.

Packaging evidence

import_project, plan_cleanup and create_archive are exercised by ProjectTests: test_unsafe_zip_names_are_rejected, test_normalized_case_and_unicode_collisions_are_rejected, test_zip_symlinks_and_special_files_are_rejected, test_zip_actual_stream_bytes_are_bounded_independently_of_metadata, test_cleanup_retains_dependencies_bibliography_styles_and_unknown_files, and test_archive_bytes_ignore_input_modes_timestamps_and_creation_order.

Those cleanup tests cover the preliminary debris pass. Preparation’s mandatory dependency selection is a separate stage, described under package contents; --no-cleanup disables only the preliminary pass.

Formatting evidence

format_project is exercised by FormattingTests: test_only_idempotent_output_is_proposed, test_formatting_differences_preserve_original, and test_resource_and_output_limits_block_even_with_zero_return_code. test_project_exclusions, test_protected_regions and test_blank_line_policy exercise matching, rejected modifications and incomplete protected contexts. These are adapter-level checks, not proof that arbitrary TeX can be formatted.

Workflow evidence

run_job is exercised by WorkflowTests: test_flat_bundle_from_exact_extraction_and_serial_equivalence and test_failed_final_build_never_releases_output. compare_pdfs is exercised by PdfTests: test_rendered_change_is_detected_when_text_matches, test_extracted_text_change_is_detected_when_raster_matches, test_page_count_difference_identifies_missing_pages, and test_missing_renderer_is_inconclusive_and_blocking. The separate live acceptance test is LiveIntegrationTests.test_nested_sources_bibliography_images_formatting_and_zip_rebuild.

Before closing a backlog row, implement the stated generic behavior, expose any required explicit settings, and add meaningful valid/violating/uncertain fixtures. Measure false positives for visual heuristics before promoting them to blockers. Update this inventory and the code catalogue together; passing a nearby test or adding a registry entry alone does not close the requirement.