sciwrite-lint 0.5.0__tar.gz → 0.6.0__tar.gz

{sciwrite_lint-0.5.0/sciwrite_lint.egg-info → sciwrite_lint-0.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sciwrite-lint
-Version: 0.5.0
+Version: 0.6.0
 Summary: A linter for scientific manuscripts — reference verification, consistency checking, and structural validation.
 Author: Sergey Samsonau
 Maintainer-email: Authentic Research Partners LLC <info@arpconnect.com>
@@ -158,11 +158,11 @@ Three models — a vision model (Qwen3-VL-2B default, or 8B FP8 via `--vision-ba
 - **14-source full-text cascade** — early exit on first successful download across arXiv, Semantic Scholar, OpenAlex, PubMed Central, Europe PMC, Unpaywall, bioRxiv, NBER, RePEc/IDEAS, HAL, ERIC, NASA ADS, OSF Preprints, CORE
 - **Live monitoring** *(advanced)* — `sciwrite-lint containers monitor` shows service health, VRAM usage, and KV cache utilization in a terminal dashboard
 
-Full pipeline on a 50-reference paper: ~30 minutes initial (dominated by downloads and claim verification), minutes on cached reruns. The pipeline automatically swaps vLLM containers to free GPU for embedding and vision stages (~50× faster than CPU) — same code path on WSL2 and native Linux.
+The pipeline automatically swaps vLLM containers to free GPU for embedding and vision stages (~50× faster than CPU).
 
 ## Install
 
-**Assumed setup:** A workstation with an NVIDIA GPU (16+ GB VRAM). Tested on WSL2 with NVIDIA driver 546.01+. Native Linux is likely to work with GPU memory allocation tuning (see [docs/services.md](https://github.com/authentic-research-partners/sciwrite-lint/blob/main/docs/services.md#embedding-device)). Not tested on macOS.
+**Assumed setup:** A workstation with an NVIDIA GPU (16+ GB VRAM). Tested on WSL2 with NVIDIA driver 546.01+. Native Linux uses the same GPU-sequencing code path with no extra configuration (see [docs/services.md](https://github.com/authentic-research-partners/sciwrite-lint/blob/main/docs/services.md#embedding-device)), though it isn't actively tested. Not tested on macOS.
 
 Requires [uv](https://docs.astral.sh/uv/getting-started/installation/), a container runtime (podman or docker), CUDA drivers, and [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html).
 
@@ -207,6 +207,8 @@ sciwrite-lint containers start         # start GROBID + vLLM (needs GPU for vLLM
 sciwrite-lint containers monitor       # live dashboard: service health, VRAM, KV cache
 ```
 
+`init` generates a `.sciwrite-lint.toml` config. With no argument it auto-detects the manuscript — a `.tex` file (bibliography read from `\bibliography{}`) or a markdown file with a sibling `{name}.bib` (so ordinary `.md` files aren't mistaken for papers); if it finds **several**, it lists them and asks you to pick one. Name a manuscript explicitly — including a **PDF** — with `sciwrite-lint init <file>`. Review to confirm the right file was registered.
+
 ### WSL2: disable CUDA Sysmem Fallback (recommended)
 
 By default Windows silently spills GPU overflow into system RAM via PCIe (30–100× slower than VRAM); you'll see this as a sudden throughput collapse during the heaviest pipeline stages with no error in the logs. Tell the NVIDIA driver to fail loudly instead:
@@ -217,8 +219,6 @@ Driver 546.01+. After this, WSL2 behaves like native Linux on overflow (CUDA OOM
 
 ![Monitor dashboard](https://github.com/authentic-research-partners/sciwrite-lint/raw/main/docs/monitor.png)
 
-`init` detects `.tex` files and their `.bib` references and generates a `.sciwrite-lint.toml` config. Review to confirm the right files were detected.
-
 **Providing references manually — two drop folders.** sciwrite-lint reads two local directories before going to the open-access waterfall, kept separate because the folder itself signals how much to trust the source:
 
 - `local_pdfs/` — **academic sources**. Accepts `.pdf` (primary) and `.md` summaries you've written yourself. Use this for paywalled papers, for OA papers whose publisher requires a browser (Cloudflare/JS walls, Cell, Springer, some PMC pages), and for anything you'd cite as peer-reviewed evidence.
@@ -275,11 +275,52 @@ sciwrite-lint check                             # all papers (batch-staged when
 sciwrite-lint check --concurrency 4             # batch parallelism (default 2, validated up to 4)
 sciwrite-lint check paper.tex                   # text + LLM rules on a .tex file
 sciwrite-lint check paper.pdf                   # check a PDF (GROBID required)
+sciwrite-lint check paper.md                    # markdown manuscript (prose, structure, [@key] citations, @fig:/@sec: cross-refs)
 sciwrite-lint check --checks prose-quality paper.tex   # run only the listed checks (comma-separated)
 sciwrite-lint contributions --paper my_paper    # add contribution axes to SciLint Score
 sciwrite-lint contributions paper.pdf           # standalone file scoring
 ```
 
+**Supported inputs.**
+
+*The manuscript you check:*
+
+| Format | Notes |
+|---|---|
+| `.tex` | LaTeX — hand-written, or pandoc-generated (e.g. from Markdown); pandoc's heading anchors are unwrapped automatically. Full pipeline. |
+| `.pdf` | Parsed via GROBID (container required). Full pipeline. |
+| `.md` | Markdown (pandoc). Prose and structure checks run; pandoc `[@key]` citations are checked against a sibling `{name}.bib` and verified against external databases in the full pipeline; pandoc-crossref `@fig:`/`@sec:`/`@tbl:` references and `{#fig:…}` figures are checked for broken/unreferenced targets. Sections split by heading. |
+
+*What runs on each format:*
+
+| Capability | `.tex` | `.pdf` | `.md` |
+|---|:---:|:---:|:---:|
+| Prose & structure checks | ✓ | ✓ | ✓ |
+| Citation ↔ bibliography (`dangling-cite`) | ✓ | ✓ | ✓ |
+| Cross-references (`dangling-ref`) | ✓ | ✓ <sup>1</sup> | ✓ |
+| Unreferenced figures | ✓ | ✓ <sup>2</sup> | ✓ |
+| Reference verification (external APIs) | ✓ | ✓ | ✓ |
+| Claim support / citation purpose | ✓ | ✓ | ✓ |
+| Footnote-URL web sources | ✓ | — <sup>3</sup> | ✓ |
+| `ref-health` (no external APIs) | ✓ | ✓ <sup>4</sup> | ✓ |
+| `init` auto-detection | ✓ | — <sup>5</sup> | ✓ <sup>6</sup> |
+
+<sup>1</sup> PDF detects broken references rendered as `??`; `.tex`/`.md` match each reference to its label.
+<sup>2</sup> A PDF has no `\label`s, so figures are matched by the number GROBID reads from each figure's caption against the "Figure N" mentions in the prose (needs GROBID). Only cleanly-labelled figures are checked, so it stays conservative.
+<sup>3</sup> Footnote-URL web sources are matched for LaTeX and markdown; not yet for PDF.
+<sup>4</sup> A PDF's references come from GROBID (a local parser — still no external/network APIs), so PDF `ref-health` needs the GROBID container running.
+<sup>5</sup> *Auto-scan* skips PDFs — no signal distinguishes a manuscript PDF from a project's reference/figure PDFs, the way a `.tex` or a `.md` with a sibling `.bib` is recognizable. Register one explicitly with `sciwrite-lint init paper.pdf` (the explicit form takes any type). When auto-scan finds several `.tex`/`.md`, it asks you to name one rather than registering all.
+<sup>6</sup> Auto-detected when a sibling `{name}.bib` is present.
+
+Markdown citation support targets pandoc `[@key]` citations with a bibliography. Numeric `[1]` citations are not supported by design — they carry no stable key (renumbering on every edit) and their reference list is free text rather than structured entries; a document using them is reported, not partially parsed.
+
+*Cited sources you optionally supply (drop folders — see [Setup](#setup)):*
+
+| Folder | Accepts | Trust level |
+|---|---|---|
+| `local_pdfs/` | `.pdf`, `.md` (your own summaries) | academic |
+| `local_web/` | `.mhtml` / `.mht`, `.md` | web capture |
+
 `check` runs the full pipeline in one command: text checks → figure analysis → LLM consistency → reference verification via APIs → download and parse cited papers → claim verification → consistency checks on cited papers → bibliography verification → aggregate reliability scores → SciLint Score. An initial run on a 50-reference paper takes up to 30 minutes (dominated by downloads and claim verification); subsequent cached runs complete in minutes.
 
 ### Iterative editing — no `--fresh` needed

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/README.md

@@ -92,11 +92,11 @@ Three models — a vision model (Qwen3-VL-2B default, or 8B FP8 via `--vision-ba
 - **14-source full-text cascade** — early exit on first successful download across arXiv, Semantic Scholar, OpenAlex, PubMed Central, Europe PMC, Unpaywall, bioRxiv, NBER, RePEc/IDEAS, HAL, ERIC, NASA ADS, OSF Preprints, CORE
 - **Live monitoring** *(advanced)* — `sciwrite-lint containers monitor` shows service health, VRAM usage, and KV cache utilization in a terminal dashboard
 
-Full pipeline on a 50-reference paper: ~30 minutes initial (dominated by downloads and claim verification), minutes on cached reruns. The pipeline automatically swaps vLLM containers to free GPU for embedding and vision stages (~50× faster than CPU) — same code path on WSL2 and native Linux.
+The pipeline automatically swaps vLLM containers to free GPU for embedding and vision stages (~50× faster than CPU).
 
 ## Install
 
-**Assumed setup:** A workstation with an NVIDIA GPU (16+ GB VRAM). Tested on WSL2 with NVIDIA driver 546.01+. Native Linux is likely to work with GPU memory allocation tuning (see [docs/services.md](https://github.com/authentic-research-partners/sciwrite-lint/blob/main/docs/services.md#embedding-device)). Not tested on macOS.
+**Assumed setup:** A workstation with an NVIDIA GPU (16+ GB VRAM). Tested on WSL2 with NVIDIA driver 546.01+. Native Linux uses the same GPU-sequencing code path with no extra configuration (see [docs/services.md](https://github.com/authentic-research-partners/sciwrite-lint/blob/main/docs/services.md#embedding-device)), though it isn't actively tested. Not tested on macOS.
 
 Requires [uv](https://docs.astral.sh/uv/getting-started/installation/), a container runtime (podman or docker), CUDA drivers, and [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html).
 
@@ -141,6 +141,8 @@ sciwrite-lint containers start         # start GROBID + vLLM (needs GPU for vLLM
 sciwrite-lint containers monitor       # live dashboard: service health, VRAM, KV cache
 ```
 
+`init` generates a `.sciwrite-lint.toml` config. With no argument it auto-detects the manuscript — a `.tex` file (bibliography read from `\bibliography{}`) or a markdown file with a sibling `{name}.bib` (so ordinary `.md` files aren't mistaken for papers); if it finds **several**, it lists them and asks you to pick one. Name a manuscript explicitly — including a **PDF** — with `sciwrite-lint init <file>`. Review to confirm the right file was registered.
+
 ### WSL2: disable CUDA Sysmem Fallback (recommended)
 
 By default Windows silently spills GPU overflow into system RAM via PCIe (30–100× slower than VRAM); you'll see this as a sudden throughput collapse during the heaviest pipeline stages with no error in the logs. Tell the NVIDIA driver to fail loudly instead:
@@ -151,8 +153,6 @@ Driver 546.01+. After this, WSL2 behaves like native Linux on overflow (CUDA OOM
 
 ![Monitor dashboard](https://github.com/authentic-research-partners/sciwrite-lint/raw/main/docs/monitor.png)
 
-`init` detects `.tex` files and their `.bib` references and generates a `.sciwrite-lint.toml` config. Review to confirm the right files were detected.
-
 **Providing references manually — two drop folders.** sciwrite-lint reads two local directories before going to the open-access waterfall, kept separate because the folder itself signals how much to trust the source:
 
 - `local_pdfs/` — **academic sources**. Accepts `.pdf` (primary) and `.md` summaries you've written yourself. Use this for paywalled papers, for OA papers whose publisher requires a browser (Cloudflare/JS walls, Cell, Springer, some PMC pages), and for anything you'd cite as peer-reviewed evidence.
@@ -209,11 +209,52 @@ sciwrite-lint check                             # all papers (batch-staged when
 sciwrite-lint check --concurrency 4             # batch parallelism (default 2, validated up to 4)
 sciwrite-lint check paper.tex                   # text + LLM rules on a .tex file
 sciwrite-lint check paper.pdf                   # check a PDF (GROBID required)
+sciwrite-lint check paper.md                    # markdown manuscript (prose, structure, [@key] citations, @fig:/@sec: cross-refs)
 sciwrite-lint check --checks prose-quality paper.tex   # run only the listed checks (comma-separated)
 sciwrite-lint contributions --paper my_paper    # add contribution axes to SciLint Score
 sciwrite-lint contributions paper.pdf           # standalone file scoring
 ```
 
+**Supported inputs.**
+
+*The manuscript you check:*
+
+| Format | Notes |
+|---|---|
+| `.tex` | LaTeX — hand-written, or pandoc-generated (e.g. from Markdown); pandoc's heading anchors are unwrapped automatically. Full pipeline. |
+| `.pdf` | Parsed via GROBID (container required). Full pipeline. |
+| `.md` | Markdown (pandoc). Prose and structure checks run; pandoc `[@key]` citations are checked against a sibling `{name}.bib` and verified against external databases in the full pipeline; pandoc-crossref `@fig:`/`@sec:`/`@tbl:` references and `{#fig:…}` figures are checked for broken/unreferenced targets. Sections split by heading. |
+
+*What runs on each format:*
+
+| Capability | `.tex` | `.pdf` | `.md` |
+|---|:---:|:---:|:---:|
+| Prose & structure checks | ✓ | ✓ | ✓ |
+| Citation ↔ bibliography (`dangling-cite`) | ✓ | ✓ | ✓ |
+| Cross-references (`dangling-ref`) | ✓ | ✓ <sup>1</sup> | ✓ |
+| Unreferenced figures | ✓ | ✓ <sup>2</sup> | ✓ |
+| Reference verification (external APIs) | ✓ | ✓ | ✓ |
+| Claim support / citation purpose | ✓ | ✓ | ✓ |
+| Footnote-URL web sources | ✓ | — <sup>3</sup> | ✓ |
+| `ref-health` (no external APIs) | ✓ | ✓ <sup>4</sup> | ✓ |
+| `init` auto-detection | ✓ | — <sup>5</sup> | ✓ <sup>6</sup> |
+
+<sup>1</sup> PDF detects broken references rendered as `??`; `.tex`/`.md` match each reference to its label.
+<sup>2</sup> A PDF has no `\label`s, so figures are matched by the number GROBID reads from each figure's caption against the "Figure N" mentions in the prose (needs GROBID). Only cleanly-labelled figures are checked, so it stays conservative.
+<sup>3</sup> Footnote-URL web sources are matched for LaTeX and markdown; not yet for PDF.
+<sup>4</sup> A PDF's references come from GROBID (a local parser — still no external/network APIs), so PDF `ref-health` needs the GROBID container running.
+<sup>5</sup> *Auto-scan* skips PDFs — no signal distinguishes a manuscript PDF from a project's reference/figure PDFs, the way a `.tex` or a `.md` with a sibling `.bib` is recognizable. Register one explicitly with `sciwrite-lint init paper.pdf` (the explicit form takes any type). When auto-scan finds several `.tex`/`.md`, it asks you to name one rather than registering all.
+<sup>6</sup> Auto-detected when a sibling `{name}.bib` is present.
+
+Markdown citation support targets pandoc `[@key]` citations with a bibliography. Numeric `[1]` citations are not supported by design — they carry no stable key (renumbering on every edit) and their reference list is free text rather than structured entries; a document using them is reported, not partially parsed.
+
+*Cited sources you optionally supply (drop folders — see [Setup](#setup)):*
+
+| Folder | Accepts | Trust level |
+|---|---|---|
+| `local_pdfs/` | `.pdf`, `.md` (your own summaries) | academic |
+| `local_web/` | `.mhtml` / `.mht`, `.md` | web capture |
+
 `check` runs the full pipeline in one command: text checks → figure analysis → LLM consistency → reference verification via APIs → download and parse cited papers → claim verification → consistency checks on cited papers → bibliography verification → aggregate reliability scores → SciLint Score. An initial run on a 50-reference paper takes up to 30 minutes (dominated by downloads and claim verification); subsequent cached runs complete in minutes.
 
 ### Iterative editing — no `--fresh` needed

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "sciwrite-lint"
-version = "0.5.0"
+version = "0.6.0"
 description = "A linter for scientific manuscripts — reference verification, consistency checking, and structural validation."
 requires-python = ">=3.13,<3.14"
 license = "MIT"

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/sciwrite_lint/__init__.py

@@ -2,7 +2,7 @@
 
 from sciwrite_lint.exceptions import LLMConnectionError, SciWriteLintError
 
-__version__ = "0.5.0"
+__version__ = "0.6.0"
 
 from sciwrite_lint.oa import (
     DownloadResult,

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/sciwrite_lint/__main__.py

@@ -155,6 +155,13 @@ def main(argv: list[str] | None = None) -> int:
     p_init = sub.add_parser(
         "init", help="Initialize project: config + references directory"
     )
+    p_init.add_argument(
+        "file",
+        nargs="?",
+        help="Manuscript to register (.tex/.pdf/.md). Omit to auto-detect "
+        "(a single .tex / .md+.bib); name one explicitly when there are "
+        "several or for a PDF.",
+    )
     p_init.add_argument(
         "--force", action="store_true", help="Overwrite existing config"
     )
@@ -175,6 +182,13 @@ def main(argv: list[str] | None = None) -> int:
     )
     p_config_email.add_argument("email", help="Your email address")
     p_config_email.add_argument("--config", help="Path to .sciwrite-lint.toml")
+    p_config_email.add_argument(
+        "--global",
+        dest="set_global",
+        action="store_true",
+        help="Store as the machine-wide default (~/.sciwrite-lint/) instead of "
+        "this project's .sciwrite-lint.toml",
+    )
     p_config_email.set_defaults(func=run_config)
 
     p_config_set_key = config_sub.add_parser(

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/sciwrite_lint/api.py

@@ -41,6 +41,20 @@ _openlibrary_limiter = MonotonicRateLimiter(5, 1.0)  # ~5 req/s (be polite)
 _loc_limiter = MonotonicRateLimiter(5, 1.0)  # ~5 req/s (be polite)
 
 
+def _openalex_params(polite_email: str, **params: Any) -> dict[str, Any]:
+    """OpenAlex query params with the polite-pool ``mailto`` merged in.
+
+    Single source of truth for every OpenAlex call site: requests carrying
+    a ``mailto`` are routed into OpenAlex's polite pool, which has a higher
+    and more stable shared rate limit than the anonymous common pool. This
+    keeps reference verification off the common pool, where sustained use
+    draws HTTP 429s. No-op when no contact email is configured.
+    """
+    if polite_email:
+        params["mailto"] = polite_email
+    return params
+
+
 class CitationAPI:
     """Verify citations against CrossRef, OpenAlex, Semantic Scholar, Open Library, and Library of Congress."""
 
@@ -129,6 +143,15 @@ class CitationAPI:
     # OpenAlex
     # ------------------------------------------------------------------
 
+    def _oa_params(self, **params: Any) -> dict[str, Any]:
+        """OpenAlex query params for this client's configured contact email.
+
+        Thin instance-scoped wrapper over :func:`_openalex_params` so the
+        four ``_openalex_lookup`` call sites don't each repeat
+        ``self._config.polite_email``.
+        """
+        return _openalex_params(self._config.polite_email, **params)
+
     async def _openalex_lookup(self, citation: Citation) -> dict[str, Any] | None:
         try:
             # Try DOI lookup
@@ -137,7 +160,7 @@ class CitationAPI:
                 resp = await rate_limited_get(
                     _openalex_limiter,
                     f"https://api.openalex.org/works/doi:{clean_doi}",
-                    params={"select": _OA_FIELDS},
+                    params=self._oa_params(select=_OA_FIELDS),
                     label="OpenAlex DOI",
                     client=self._client,
                     service="openalex",
@@ -155,7 +178,7 @@ class CitationAPI:
                 resp = await rate_limited_get(
                     _openalex_limiter,
                     f"https://api.openalex.org/works/doi:{arxiv_doi}",
-                    params={"select": _OA_FIELDS},
+                    params=self._oa_params(select=_OA_FIELDS),
                     label="OpenAlex arXiv DOI",
                     client=self._client,
                     service="openalex",
@@ -173,10 +196,10 @@ class CitationAPI:
                 resp = await rate_limited_get(
                     _openalex_limiter,
                     "https://api.openalex.org/works",
-                    params={
-                        "filter": f"ids.pmid:https://pubmed.ncbi.nlm.nih.gov/{citation.pmid}",
-                        "select": _OA_FIELDS,
-                    },
+                    params=self._oa_params(
+                        filter=f"ids.pmid:https://pubmed.ncbi.nlm.nih.gov/{citation.pmid}",
+                        select=_OA_FIELDS,
+                    ),
                     label="OpenAlex PMID",
                     client=self._client,
                     service="openalex",
@@ -192,11 +215,11 @@ class CitationAPI:
             resp = await rate_limited_get(
                 _openalex_limiter,
                 "https://api.openalex.org/works",
-                params={
-                    "filter": f"title.search:{query}",
-                    "per_page": 10,
-                    "select": _OA_FIELDS,
-                },
+                params=self._oa_params(
+                    filter=f"title.search:{query}",
+                    per_page=10,
+                    select=_OA_FIELDS,
+                ),
                 label="OpenAlex search",
                 client=self._client,
                 service="openalex",
@@ -481,13 +504,12 @@ async def batch_openalex(
     doi_filter = "|".join(f"https://doi.org/{doi}" for _, doi in all_dois)
     doi_by_key = {doi.lower(): key for key, doi in all_dois}
 
-    params: dict[str, str | int] = {
-        "filter": f"doi:{doi_filter}",
-        "per_page": 200,
-        "select": _OA_FIELDS,
-    }
-    if config.polite_email:
-        params["mailto"] = config.polite_email
+    params = _openalex_params(
+        config.polite_email,
+        filter=f"doi:{doi_filter}",
+        per_page=200,
+        select=_OA_FIELDS,
+    )
 
     # Secondary index by arXiv ID: OpenAlex may return a different
     # canonical DOI than the query DOI (e.g. arXiv DOI → publisher DOI).

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/sciwrite_lint/checks/_diagnostics.py

@@ -38,7 +38,7 @@ Two orthogonal axes describe a Finding:
    issue". Same level vocabulary, different semantics — the panel /
    JSON-field disambiguates.
 
-The three diagnostic rule_ids:
+The diagnostic rule_ids:
 
 - ``llm-unavailable``  — ``llm_query`` exhausted its retry ladder
   (medium → low → off) and returned ``None``. One per skipped check
@@ -59,6 +59,17 @@ The three diagnostic rule_ids:
   during ``build_queries`` or ``process_results`` / a text check's
   body raised mid-run. One per failing check. Emitted by the LLM
   batch runner and the text-check runner.
+- ``source-unsupported`` — one or more checks declare
+  ``supports_markdown=False`` (they parse LaTeX citation/cross-reference/
+  figure syntax) and the manuscript is markdown. The text-check runner
+  skips those checks and emits one summary finding listing them, so the
+  coverage gap is explicit rather than a silent zero-result.
+- ``no-references`` — citation extraction found zero references in the
+  manuscript, so the citation / reference / claim stages had nothing to
+  verify. Emitted once per run by the pipeline setup. A genuinely
+  citation-free note is harmless, but a zero usually means the
+  bibliography was not detected — surfacing it stops a clean-looking
+  report from hiding that the reference checks never ran.
 
 Adding a new system rule_id:
 
@@ -92,6 +103,8 @@ SYSTEM_RULE_IDS: frozenset[str] = frozenset(
         "vision-incomplete",
         "internal-error",
         "parse-failed",
+        "source-unsupported",
+        "no-references",
     }
 )
 
@@ -235,3 +248,56 @@ def internal_error_finding(check_id: str, error: BaseException) -> Finding:
         message=f"{check_id}: check raised an unexpected exception; check did not run",
         context=f"{type(error).__name__}: {error!s}"[:200],
     )
+
+
+def source_unsupported_finding(
+    source_type: str, skipped_check_ids: list[str]
+) -> Finding:
+    """Build a finding for checks skipped because the source type lacks the structure they need.
+
+    Emitted once per run by the text-check runner when the manuscript is
+    markdown and one or more registered checks declare
+    ``supports_markdown=False`` (they parse LaTeX citation/cross-reference/
+    figure syntax or need a populated bibliography). Listing the skipped
+    checks makes the coverage gap explicit — a markdown manuscript still
+    gets prose and structure checks, but these did not run.
+
+    Parameters
+    ----------
+    source_type:
+        The manuscript source type (e.g. ``"markdown"``).
+    skipped_check_ids:
+        Registered check IDs that were skipped for this source.
+    """
+    checks = ", ".join(sorted(skipped_check_ids))
+    return Finding(
+        level="warning",
+        rule_id="source-unsupported",
+        message=(
+            f"{len(skipped_check_ids)} check(s) need citation/reference structure "
+            f"not extracted from {source_type} manuscripts and did not run: {checks}. "
+            "Prose and structure checks ran normally."
+        ),
+    )
+
+
+def no_references_finding() -> Finding:
+    """Build a finding for a run that extracted zero references.
+
+    Emitted once by the pipeline setup when citation extraction yields an
+    empty list. The reference, citation, and claim stages then have
+    nothing to verify — surfacing this keeps a clean-looking report from
+    hiding that those checks never ran (usually a bibliography that was
+    not detected, not a genuinely citation-free manuscript).
+    """
+    return Finding(
+        level="warning",
+        rule_id="no-references",
+        message=(
+            "No references found in the manuscript — citation, reference, and "
+            "claim checks had nothing to verify. If the manuscript cites sources, "
+            "check that its bibliography is detected (LaTeX \\bibliography{} / "
+            ".bib, or for markdown the YAML bibliography: field or a sibling "
+            "{name}.bib)."
+        ),
+    )

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/sciwrite_lint/checks/dangling_cite.py

@@ -48,12 +48,13 @@ def _check_latex(tex_path: Path, config: LintConfig) -> list[Finding]:
     return findings
 
 
-def _check_pdf(ctx: object) -> list[Finding]:
+def _check_context(ctx: object) -> list[Finding]:
     """Check that inline citations link to a reference in the bibliography.
 
-    Uses ManuscriptContext's inline_citations (from GROBID TEI <ref> tags)
-    and parsed_references. Citations with keys not in the reference list
-    are dangling.
+    Serves any ManuscriptContext that populates ``inline_citations`` and
+    ``parsed_references`` — PDF (GROBID TEI <ref> tags) and markdown
+    (pandoc ``[@key]`` + sibling ``.bib``) both do. Citations whose key is
+    not in the reference list are dangling.
     """
     from sciwrite_lint.manuscript_store import ManuscriptContext
 
@@ -86,8 +87,13 @@ def _check_pdf(ctx: object) -> list[Finding]:
     description="Citation has no matching bibliography entry.",
 )
 def check_dangling_cite(tex_path: Path, config: LintConfig) -> list[Finding]:
-    """Check dangling citations. Supports both LaTeX and PDF input."""
-    if config.is_pdf:
-        return _check_pdf(config.manuscript_context)
+    """Check dangling citations. Supports LaTeX, PDF, and markdown input.
+
+    PDF and markdown both populate ``inline_citations`` +
+    ``parsed_references`` on the context, so the same key-membership
+    check (``_check_context``) serves both; LaTeX is parsed from source.
+    """
+    if config.is_pdf or config.is_markdown:
+        return _check_context(config.manuscript_context)
 
     return _check_latex(tex_path, config)

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/sciwrite_lint/checks/dangling_ref.py

@@ -2,6 +2,8 @@ r"""Check: dangling-ref — cross-reference has no matching target.
 
 For LaTeX: \ref{X} has no matching \label{X}.
 For PDF: rendered "??" patterns in section text (broken references).
+For markdown: pandoc-crossref ``@fig:`` / ``@sec:`` / ``@tbl:`` reference
+with no matching labelled element.
 """
 
 from __future__ import annotations
@@ -102,6 +104,32 @@ def _check_pdf(ctx: object) -> list[Finding]:
     return findings
 
 
+def _check_markdown(md_path: Path) -> list[Finding]:
+    """Flag pandoc-crossref references with no matching labelled element.
+
+    ``@fig:x`` / ``@sec:x`` / ``@tbl:x`` whose target is not defined by any
+    block id in the document is dangling.
+    """
+    from sciwrite_lint.markdown_cites import analyze_markdown
+
+    analysis = analyze_markdown(md_path.read_text(encoding="utf-8"))
+    findings: list[Finding] = []
+    seen: set[str] = set()
+    for xref in analysis.crossrefs:
+        if xref.target not in analysis.labels and xref.target not in seen:
+            seen.add(xref.target)
+            findings.append(
+                Finding(
+                    level="error",
+                    rule_id="dangling-ref",
+                    message=f"Cross-reference '@{xref.target}' has no matching label",
+                    file=md_path.name,
+                    context=xref.target,
+                )
+            )
+    return findings
+
+
 @check(
     id="dangling-ref",
     category="manuscript",
@@ -109,7 +137,9 @@ def _check_pdf(ctx: object) -> list[Finding]:
     description="Cross-reference has no matching target.",
 )
 def check_dangling_ref(tex_path: Path, config: LintConfig) -> list[Finding]:
-    """Check dangling references. Supports both LaTeX and PDF input."""
+    """Check dangling references. Supports LaTeX, PDF, and markdown input."""
+    if config.is_markdown:
+        return _check_markdown(tex_path)
     if config.is_pdf:
         return _check_pdf(config.manuscript_context)
 

{sciwrite_lint-0.5.0 → sciwrite_lint-0.6.0}/sciwrite_lint/checks/full_paper_consistency.py

@@ -570,6 +570,29 @@ If FIGURE DESCRIPTIONS says "Not available.", return {{"issues": []}}.\
 ]
 
 
+def selected_check_defs(
+    config: "LintConfig", *, has_figures: bool
+) -> list[tuple[str, str, str, str, bool]]:
+    """Full-paper check defs that should actually run.
+
+    Single source of truth for two filters applied together:
+
+    1. Figure availability — figure checks are pointless without figure
+       descriptions.
+    2. The enabled-check gate (``config.is_check_enabled`` — honors both
+       ``--checks`` and ``[lint] disabled_rules``).
+
+    Query-building and result-processing both call this so they select the
+    same checks in the same order. If they diverge, batched LLM results get
+    zipped against the wrong ``check_id``.
+    """
+    return [
+        d
+        for d in _CHECK_DEFS
+        if (has_figures or not d[4]) and config.is_check_enabled(d[0])
+    ]
+
+
 # ---------------------------------------------------------------------------
 # Factory: build_queries / process_results for each check
 # ---------------------------------------------------------------------------