---
title: "Scholar Profiles And Indexing Research"
canonical_url: "https://scholarxiv.com/developers/docs/scholar-profiles-and-indexing-research"
markdown_url: "https://scholarxiv.com/developers/docs/scholar-profiles-and-indexing-research.md"
---
> For the complete documentation index, see [llms.txt](/llms.txt).
# Scholar Profiles And Indexing Research
URL: /developers/docs/scholar-profiles-and-indexing-research
LLM index: /llms.txt
# Discoverability, Citations & Researcher Profiles — Research Dossier
Companion to [journal-platform-research.md](journal-platform-research.md). Covers two questions:
1. **Should we accept PDF and DOCX alongside required LaTeX source?** (§1)
2. **How do Google Scholar indexing, citation counting, and author profiles actually work — and how
do we build a researcher profile with real citation data?** (§2–§7)
Sources are primary and fetched July 2026. API field shapes in §4 were verified with live calls, not
recalled.
---
## 1. PDF and DOCX — the elaboration you asked for
### 1.1 What the incumbents actually do
| Server | LaTeX source | PDF | DOCX |
|---|---|---|---|
| **arXiv** | **preferred** | accepted **only if not TeX-derived** | **rejected outright** |
| bioRxiv / medRxiv | rejected (must convert to PDF first) | accepted | accepted → auto-converted to PDF |
| Nature | accepted at acceptance stage only, then converted to Word | initial submission only | **preferred** |
| PLOS ONE | rejected (submit the PDF) | accepted for LaTeX users | accepted (DOC/DOCX/RTF) |
arXiv's rule is the sharp one: *"We do not accept dvi, PS, or PDF created from TeX/LaTeX source."*
Their stated reason is portability and stability over time — a PDF is a dead end, source can be
re-rendered forever.
**That rule is mechanically enforceable.** Every TeX engine stamps the PDF's `/Producer` and
`/Creator` metadata (`pdfTeX-1.40.x`, `XeTeX`, `LuaTeX`). We can detect a TeX-derived PDF in
milliseconds and reject it with "you have the source — upload it." No judgement call needed.
### 1.2 The tension in your own decisions
You chose **all disciplines** *and* **LaTeX required**. Those pull against each other, hard:
- LaTeX is the norm in physics, math, CS, astronomy, quantitative economics, and statistics.
- Word is the norm in **biology, medicine, chemistry, psychology, most social sciences, and the
humanities** — the majority of the world's researchers by headcount.
bioRxiv exists precisely because life scientists don't write TeX. If we require source-only, we
build an arXiv competitor, not an all-disciplines preprint server. That's a fine product — it's just
a different one than "all disciplines" implies, and it should be a deliberate choice rather than a
side effect.
### 1.3 What we gain by having the source (this is the real argument)
The compile pipeline you already have (`src/lib/write/` + Tectonic in E2B) makes source-in
disproportionately valuable:
| Capability | With LaTeX source | With DOCX | With PDF only |
|---|---|---|---|
| **Automated accept gate** (compiles = valid) | ✅ free, replaces the human moderator you're not hiring | ⚠️ via conversion | ❌ nothing to check |
| **HTML full text** (accessibility, mobile, SEO, Scholar likes it) | ✅ | ✅ | ❌ |
| **Structured metadata** (title/authors/abstract/refs read from source) | ✅ exact | ✅ good | ❌ guessed by a parser |
| **Reference list extraction** → citation graph | ✅ from `.bbl`/`.bib` | ✅ | ⚠️ heuristic |
| **MathML / accessible equations** | ✅ | ⚠️ | ❌ |
| **Semantic version diffs** ("what changed in v2") | ✅ | ⚠️ | ❌ |
| **JATS XML export** (indexers, PMC, preservation) | ✅ | ✅ | ❌ |
| **Format migration in 20 years** | ✅ | ✅ | ⚠️ |
| **Spam resistance** | ✅ high friction | ⚠️ medium | ❌ upload anything |
That last row matters more than usual for us. **With no human moderation, the compile gate is the
only gate we have.** A PDF-only lane has no gate at all.
### 1.4 Recommendation — a three-tier acceptance model
Don't make it a binary. Make the *tier* visible on the paper's page, and let it drive features:
**Tier 1 — `source` (LaTeX).** Required path, the default, and the only one available at launch.
Full feature set: HTML view, MathML, structured metadata, version diffs, JATS export, "Source
available" badge, higher default ranking in our own search.
**Tier 2 — `converted` (DOCX).** Ship right after launch. Pipeline: DOCX → pandoc → LaTeX → Tectonic
→ PDF + HTML. DOCX carries real structure (styles, headings, footnotes, equations as OMML), so
conversion is genuinely good — far better than anything you can do from a PDF. Keep the original
DOCX as an archived asset. This is what unlocks "all disciplines" honestly.
**Tier 3 — `unstructured` (PDF).** Accept only with these guardrails:
- **Reject if `/Producer` or `/Creator` matches a TeX engine** — arXiv's exact rule, with a message
telling them to upload the source.
- **Reject if there is no extractable text layer** (scans). Google Scholar rejects these too, so
accepting them would mean hosting papers that can never be indexed or cited.
- **Reject if >5 MB** (Scholar's hard cap — see §2.2). Or warn and offer compression.
- Metadata typed by hand by the submitter, flagged as unverified.
- No HTML view, no diffs, no JATS. Visibly a second-class object.
**My call:** launch Tier 1 only (it's what you already have and it's the strongest product), add
Tier 2 within the first few iterations because "all disciplines" demands it, and treat Tier 3 as a
grudging escape hatch that we may never need. If you want to keep it maximally simple, ship
**Tier 1 + Tier 3-with-guardrails** and add DOCX when the first biologist complains.
### 1.5 One hard constraint you should design around now
**Google Scholar will not index a PDF over 5 MB, and will not index a PDF without a text layer.**
Our compiler should surface the output PDF's size at compile time and warn above ~4 MB, with the
same advice arXiv gives: JPEG for photographs, PDF/PNG/PS for line art, and watch for images above
34 megapixels. A paper we host that Scholar can't index is a paper that will never accrue citations —
which makes the whole profile feature in §6 worthless for it.
---
## 2. Google Scholar — how content gets in
Scholar has **no public API and no submission form.** You do not submit to Scholar; Scholar crawls
you. Getting indexed is purely a matter of meeting their technical guidelines.
### 2.1 Site-level requirements
- The site must host **primarily scholarly articles** — journal papers, conference papers,
dissertations, preprints, technical reports. (Scholar generally excludes book reviews, editorials,
and **documents without authors**.)
- **Every article needs its own unique URL.** Not multiple abstracts on one page, not multiple papers
in one PDF, not one paper split across several files.
- **Complete abstracts must be visible** without signing in, installing anything, dismissing a popup,
or scrolling past other content.
- **Browse interface:** every article URL must be reachable from the homepage in **≤10 simple HTML
links**. Recommended shapes: a single publications page (small collections); browse by publication
date or entry date (thousands); plus a **"last two weeks" listing** for frequent recrawl once you
pass ~100,000 papers.
- **No Flash, JavaScript, or form-based navigation** for those links — plain HTML `GET` links.
(Relevant to us: SvelteKit must SSR the browse pages and the article pages.)
- **robots.txt** must not block article or browse URLs; *should* block search results, carts,
comment forms and other dynamic junk.
- **Uptime and correct status codes:** `5xx` for temporary errors, `4xx` for permanent ones,
**HTTP 301 when an article moves — never redirect to the homepage.**
### 2.2 File requirements
- **HTML or searchable PDF only. Maximum 5 MB.** (Over 5 MB → they tell you to use Google Books.)
- PDFs must have text extractable by Acrobat Reader. No scans.
### 2.3 Metadata — the Highwire Press tags
Scholar reads four schemes; **Highwire Press `citation_*` is preferred** (also supported: BE Press
`bepress_citation_*`, PRISM `prism.*`, Dublin Core `DC.*` — DC is explicitly the least reliable).
Minimum required: **title, first author, publication year.**
```html
```
Also available: `citation_conference_title`, `citation_volume`, `citation_issue`,
`citation_firstpage`, `citation_lastpage`, `citation_isbn`,
`citation_dissertation_institution`, `citation_technical_report_number`.
Rules: author names as `Smith, John` or `John Smith`; **omit affiliations and degrees**; present
values as they'd appear in a citation (`Trans. Mag. Real.`, not `Magic Realism, Transactions on`);
escape HTML special characters properly.
### 2.4 If you have no meta tags, the PDF's *visual layout* becomes the metadata
Worth knowing because it constrains our LaTeX templates:
- **Title = the largest text on page 1, minimum 24pt, at the top.**
- **Authors = 16–23pt**, immediately before or after the title.
- **A formal bibliographic citation line alone on its own line** in the header or footer.
- **Use Type 1 fonts, not Type 3** — i.e. Times/Helvetica/Palatino in LaTeX. (Old `latex`+`dvips`
bitmap-font setups produce Type 3 and break Scholar's text extraction. Tectonic/pdfTeX with
standard font packages is fine; this is a template-lint rule we should enforce.)
- Journal/repository name must be in **smaller** type than the title and authors.
We will emit meta tags, so this is a belt-and-braces concern — but our default templates should
comply anyway, because the PDF travels beyond our site.
### 2.5 Reference formatting — this is how citations happen
Scholar **parses reference lists out of the full text.** Their requirements:
- Mark the section with a standard heading — **"References" or "Bibliography"** — on its own line.
- Format entries as **numbered**: `1.`/`2.`/`3.` or `[1]`/`[2]`/`[3]` in PDFs; an `` in HTML.
- Each entry must be a **formal bibliographic citation with no commentary mixed in.**
- *"Incorrect formatting may cause exclusion or low ranking."*
Practical consequence for us: our LaTeX templates should default to a numeric bibliography style
(`plainnat`/`unsrt`/`ieeetr`-like), and our HTML renderer must emit the bibliography as a real
`` under an `References
`. Author-year styles are riskier for parsers.
### 2.6 Timeline, and the risk our no-moderation choice creates
- Scholar updates **several times a week**; new papers typically appear **within weeks**.
- **Corrections to already-indexed papers take 6–9 months, sometimes years.** Get the metadata right
the first time — a bad `citation_author` is effectively permanent for the better part of a year.
- The `site:` operator's result count is explicitly unreliable for measuring coverage.
**The risk, stated once:** Scholar's first content requirement is that the site host *primarily
scholarly articles*. A free, all-disciplines, zero-moderation server is a spam magnet, and the
penalty for failing that test is de-indexing — which would destroy the citation and profile features
that are the whole point of §6. I'd still build it your way, but with an **automated** gate that
costs no staff time:
1. Compile must succeed (already have it).
2. Similarity check against the corpus + web (Crossref Similarity Check or an open equivalent).
3. Account verification + per-account rate limits (arXiv caps back-catalogue posting at 3/day).
4. Structural sanity: has an abstract, has a references section, above a minimum length, is in
English or declares its language, isn't a duplicate of an existing entry.
5. LLM scope/type classifier that flags — not blocks — likely non-research uploads.
That is a machine, not a moderation team, and it keeps us inside Scholar's content requirement.
---
## 3. Google Scholar — how citations are actually counted
- Scholar crawls full text, **extracts each paper's reference list, and matches those references to
records it has already indexed.** Citation counts are the size of the resulting inbound set.
- Their own caveat: counts *"reflect the state of the web as it is currently visible to our search
robots."* They are an estimate, not a ledger.
- **Version clustering is the important mechanic.** Scholar automatically identifies different
versions of the same article — preprint, publisher version, repository copies — and groups them
into one record ("All versions"). **Citations aggregate across the cluster.** This is why an arXiv
preprint and its published journal version share a citation count.
**Why this matters enormously for us:** if a ScholarXIV preprint is correctly recognised as a version
of the eventual journal article, our copy inherits the whole citation count and appears in "All
versions" — free credibility and traffic. To make that clustering work we need: identical/near
title and author strings, a DOI, and ideally an explicit `is-preprint-of` relationship registered
with the DOI (§5).
- An **asterisk (\*)** on a profile means the count includes citations that might not match that
article — i.e. automated matching wasn't confident.
---
## 4. Citation data we can actually use in-product
**Google Scholar cannot be a data source for us.** No API, and scraping violates their terms. Every
"Scholar-like" product is built on the open stack below. Field shapes here were verified with live
API calls today.
### 4.1 OpenAlex — the best default
Free, CC0, ~2× the coverage of Scopus/WoS by their claim. A free API key now gives $1/day of usage;
heavier use needs a paid plan or the quarterly snapshot.
`GET https://api.openalex.org/authors?search=Yoshua%20Bengio` returns, per author:
```jsonc
{
"id": "https://openalex.org/A5086198262",
"display_name": "Yoshua Bengio",
"raw_author_names": ["Bengio Y.", "Bengio, Yoshua", "Y. Bengio", "YOSHUA BENGIO", …],
"works_count": 1290,
"cited_by_count": 475760,
"summary_stats": { "2yr_mean_citedness": 7.90, "h_index": 186, "i10_index": 723 },
"affiliations": [
{ "institution": { "ror": "https://ror.org/01sdtdd95", "display_name": "CIFAR",
"country_code": "CA", "type": "facility" },
"years": [2025, 2024, 2023, …] }
]
}
```
**`summary_stats` hands us h-index and i10-index precomputed** — the two headline numbers on a Google
Scholar profile — and `raw_author_names` is a ready-made name-variant list for disambiguation.
`GET https://api.openalex.org/works?filter=doi:…` returns, per work: `doi`, `title`,
`publication_date`, `type` (`"preprint"`), **`indexed_in: ["arxiv","datacite"]`**,
`primary_location { landing_page_url, pdf_url, version: "submittedVersion", source { type:
"repository", … } }`, `open_access { is_oa, oa_status: "green", oa_url }`, and `authorships[]` with
each author's OpenAlex id, institutions (with **ROR**), `is_corresponding`, and
`raw_affiliation_strings`.
Note the `primary_location.id` on that arXiv record: **`pmh:oai:arXiv.org:2205.01833`**. OpenAlex
ingested it **through OAI-PMH**. That is the doorway (§5.2).
### 4.2 Semantic Scholar Academic Graph (S2AG)
Free; API key optional but strongly recommended — unauthenticated calls get 429'd quickly (mine did).
- Paper: `citationCount`, **`influentialCitationCount`** (their distinctive signal), `referenceCount`,
`externalIds` (ArXiv, DOI, DBLP, MAG, ACL, PubMed, CorpusId), `openAccessPdf`, `tldr`,
`embedding.specter_v2`, plus `/citations` and `/references` endpoints.
- Author: `paperCount`, `citationCount`, `hIndex`. *(Verified live: Oren Etzioni → 258 / 43,671 / 88.)*
- Batch: `POST /paper/batch`, ≤500 ids, ≤10 MB, ≤9,999 citations per call. Accepts
`ARXIV:2106.15928` style ids directly — useful for us.
### 4.3 Crossref
- Any DOI's inbound count is on the work record as **`is-referenced-by-count`** (verified:
`10.1038/nature14539` → 74,490). Free, no key.
- **Cited-by** service is free but requires you to **deposit your reference lists**; Crossref warns
that not depositing your own references under-reports counts **by at least 20%**. Crossref only
matches Crossref-registered → Crossref-registered works.
### 4.4 OpenCitations
CC0 open citation index (COCI, OpenCitations Meta). Good as a third source and as the ideologically
consistent one for an open platform. REST APIs + full dumps.
### 4.5 Summary
| Source | Cost | Author metrics | Coverage | Use it for |
|---|---|---|---|---|
| **OpenAlex** | free (key, $1/day) | **h-index, i10-index, works, citations** | broadest | **primary** — profiles, metrics, affiliations, ROR |
| Semantic Scholar | free (key advised) | hIndex, paperCount, citationCount | strong in CS/bio | cross-check, influential citations, TLDRs, embeddings |
| Crossref | free | none | DOI-registered only | per-paper counts, reference deposit, our own DOIs |
| OpenCitations | free | none | DOI-based | open citation graph, third opinion |
| Google Scholar | ❌ no API | — | broadest | **link out only** |
| Scopus / WoS | paid | yes | curated | not worth it |
**Design rule:** merge sources by DOI → arXiv id → normalised title+year, and **always display which
source a number came from and when it was fetched.** Nothing destroys trust in an academic product
faster than an unattributed citation count that disagrees with Scholar.
---
## 5. Making *our* papers citable and indexable
Six things, in dependency order. All are cheap; none are optional if profiles are to mean anything.
### 5.1 Persistent identifiers
Mint a DOI per paper and per version. **Use Crossref, not DataCite** — see the pricing comparison in
[preprint-design-ingest-citations-profiles.md](preprint-design-ingest-citations-profiles.md) §5.1.
(arXiv uses DataCite — the OpenAlex record above shows `indexed_in: ["arxiv","datacite"]` — but that
only pays off above ~12,000 papers/year, and arXiv posts ~200k.) Register the **`is-preprint-of` /
`is-previous-version-of` relationships** so aggregators can cluster our copy with the published
version (§3).
Version policy to copy from bioRxiv: **all versions share one DOI**, with version-specific URLs
(`/abs/XXXX.NNNNNv2`) for citing a specific one.
### 5.2 An OAI-PMH endpoint
This is how the aggregators find us — it is the single highest-leverage item on this list.
arXiv's implementation is the template:
- OAI-PMH **v2.0**, one stable base URL.
- **Item = article**, latest version exposed (with version history inside the richer format).
- Metadata formats: **`oai_dc`** (Simple Dublin Core — the universal one) plus a native format with
authors split out, categories, and license. arXiv offers `oai_dc`, `arXiv`, `arXivRaw`.
- **Sets** for selective harvesting, hierarchical: `group:archive:CATEGORY`.
- Datestamps = last modification, supporting incremental harvest via `from`.
- Update nightly, right after new papers go live.
Harvesters that will pick us up: OpenAlex, CORE, BASE, Semantic Scholar, OpenAIRE, Unpaywall.
### 5.3 Landing pages that satisfy §2
SSR'd HTML, unique URL per paper and per version, abstract visible with no interstitial, full
Highwire `citation_*` tags, absolute `citation_pdf_url`, and a browse interface within 10 clicks
(by date, by category, by author) plus a "recent" feed. Sitemap. Correct 301s forever.
### 5.4 Full text in both HTML and PDF
PDF ≤5 MB with a real text layer and Type 1 fonts; HTML for accessibility, mobile, and better
parsing. Tier-1 (LaTeX) submissions can produce both from one source — this is exactly why §1
recommends requiring source.
### 5.5 A conventionally formatted reference list
`References
` + `` in HTML; numbered entries under a "References" heading in the PDF;
no commentary interleaved. See §2.5.
### 5.6 Deposit our references
If we register DOIs and deposit reference lists, we join the citation graph as a *citing* party, not
just a cited one — and Crossref Cited-by starts returning real data for our papers.
---
## 6. The researcher profile — design
### 6.1 What Google Scholar's profile actually is (the thing to match, then beat)
**Creation:** sign in with a Google account → four steps — confirm name spelling; enter affiliation
and research interests; **select your articles from clusters of similarly-named authors**; choose how
future updates are handled.
**Fields:** name, affiliation, research interests, photo, homepage URL, and a **verified
institutional email**. The email is *hidden from the public* but is **required for the profile to
appear in Scholar search results** — profile must be public **and** email-verified.
**Adding works, three ways:** search by title/keyword; enter bibliographic data manually if search
fails; or **"add article groups"** for people who publish under several name forms.
**Merging:** duplicates are consolidated with a **Merge** button that combines counts **without
double-counting** papers that cite both versions.
**Update mode:** *automatic* (matching articles are added without approval) or *notify me* (email
review first). This applies only to the article list — **citation counts always update automatically.**
**Metrics shown**, each in an **All** and a **Recent (last 5 years)** column:
- **Citations** — total.
- **h-index** — "the largest number *h* such that at least *h* articles were cited at least *h* times
each." (Example from their own docs: citations of 17, 9, 6, 3, 2 → h = 3.)
- **i10-index** — number of articles with **≥10 citations**. Scholar-specific; nobody else uses it.
- An **asterisk** marks counts that include possibly-mismatched citations.
**Also:** public/private toggle; a stable public URL `scholar.google.com/citations?user=`;
co-author list; a per-article citation-over-time graph; "follow" alerts for new articles by an author
or new citations to a paper.
*(For venues rather than people, Scholar Metrics uses **h5-index** and **h5-median** — the h-index and
h-core median restricted to the last five complete calendar years. If we ever want to publish a
"ScholarXIV h5-index", that's the definition.)*
### 6.2 Proposed data model
```
ResearcherProfile
userId, handle (public slug), visibility: private | public
displayName, nameVariants[] ← seed from OpenAlex raw_author_names
photoUrl, pronouns?, bio, homepageUrl, socials[]
affiliations[] { institutionName, rorId, department, role, startYear, endYear, isCurrent }
verifiedEmails[] { address, domain, verifiedAt, isInstitutional, public: false }
interests[] / topics[] ← controlled vocabulary + free text
identifiers { openAlexId, semanticScholarId, arxivAuthorId, dblp, googleScholarUrl }
metricsCache { source, fetchedAt, citations, hIndex, i10Index,
citationsRecent5y, hIndexRecent5y, i10IndexRecent5y,
worksCount, perYear: [{year, count}] }
settings { autoAddWorks: bool, publicMetrics: bool, showEmail: false }
ProfileWork
profileId, workId
origin: native | claimed_external
claimState: auto_matched | confirmed | rejected
identifiers { doi, arxivId, openAlexId, s2PaperId }
authorPosition, isCorresponding
citationCounts[] { source, count, fetchedAt }
mergedInto? ← duplicate consolidation, Scholar-style
```
### 6.3 Claiming and disambiguation (the genuinely hard part)
Author name disambiguation is the unsolved problem of this domain. Ladder, strongest first:
1. **Native works.** Anything submitted through ScholarXIV by that account is auto-attached.
2. **Email match** against the corresponding-author email in the metadata.
3. **Candidate matching** — OpenAlex author entities whose `raw_author_names` overlap plus
affiliation/topic overlap — presented as *suggestions the user confirms*, never auto-attached.
This is exactly Scholar's step 3 (pick your articles from clusters) and it works because the human
does the last mile.
4. **Manual add** by DOI / arXiv ID / free-form citation.
5. **Merge / split** controls, with a rule that merging never double-counts a citing paper.
Store `claimState` so an unconfirmed auto-match can never inflate public metrics.
### 6.5 Metrics: compute, cache, attribute
- Pull `summary_stats` from OpenAlex as the headline; cross-check totals against Semantic Scholar and
Crossref; store all three with `fetchedAt`.
- Compute **h-index** and **i10-index** ourselves too, from the merged deduplicated work list, so a
claimed-works profile stays correct when a user adds a paper OpenAlex hasn't linked to them.
- Offer the **All / Recent(5y)** split Scholar users expect.
- Show a **per-year citation histogram** and a **per-year publication count**.
- **Always label the source and the as-of date.** "Citations 1,204 · via OpenAlex, 2 hours ago."
- Refresh on a schedule (daily for active profiles, weekly otherwise), never on page load.
### 6.6 Where we can beat Scholar
Scholar's profile is a 2011 web page. Cheap differentiators, roughly in order of value/effort:
1. **Platform-native metrics Scholar can't have** — views, downloads, and version history for papers
hosted here (arXiv-style, and honest: label them as engagement, not impact).
2. **Co-author graph** — an actual visualisation; OpenAlex `authorships` gives us the edges free.
3. **Topic map** over the author's corpus (we already do classification in this product).
4. **"Cited by" reading list** with our own summarisation on top — this is the natural bridge to the
rest of ScholarXIV.
5. **One-click claim at onboarding**: paste an arXiv author ID, get a populated profile in
seconds. First-run experience decides whether anyone ever comes back.
7. **Public API + machine-readable profile** (JSON-LD `ScholarlyArticle` / `Person`), because Scholar
has none and researchers want their data out.
8. **DORA-consistent framing** — show distributions, not just a single h-index. It costs nothing and
signals we understand the field's own critique of these metrics.
---
## 7. Build order for this half
1. **Landing pages + Highwire meta tags + sitemap + SSR browse** — nothing else works without these.
2. **PDF hygiene in the compile pipeline** — <5 MB warning, text layer assertion, Type 1 font lint.
3. **DOIs via DataCite**, with version + `is-preprint-of` relationships.
4. **OAI-PMH endpoint** (`oai_dc` + a native format, hierarchical sets, nightly).
5. **Profile v1** — identity, affiliations, and native works.
6. **Metrics v1** — OpenAlex ingest, cached, attributed, with All/Recent split.
7. **Claim & merge UI** — candidate matching, confirm/reject, merge.
8. **Differentiators** — co-author graph, topic map, cited-by reading list, public profile API.
---
## Sources
Google Scholar: [technical inclusion guidelines](https://scholar.google.com/intl/en/scholar/inclusion.html) ·
[Scholar Citations / profiles](https://scholar.google.com/intl/en/scholar/citations.html) ·
[how Scholar works](https://scholar.google.com/intl/en/scholar/help.html) ·
[Scholar Metrics](https://scholar.google.com/intl/en/scholar/metrics.html) —
[OpenAlex API docs](https://docs.openalex.org/) (author/work shapes verified live against
`api.openalex.org`) —
[Semantic Scholar Academic Graph API](https://api.semanticscholar.org/api-docs/graph) (verified live) —
[Crossref Cited-by](https://www.crossref.org/documentation/cited-by/) (verified live via
`api.crossref.org`) —
[OpenCitations](https://opencitations.net/) —
[arXiv OAI-PMH interface](https://info.arxiv.org/help/oa/index.html) —
[arXiv submission formats](https://info.arxiv.org/help/submit/index.html) —
## Sitemap
See the full [sitemap](/sitemap.md) for all pages.
Docs-scoped sitemap: [/docs/sitemap.md](/docs/sitemap.md).
Well-known sitemap: [/.well-known/sitemap.md](/.well-known/sitemap.md).