Publication intake rules
Codified rules from publication discovery batches 1–4 (May 2026). These govern what enters the resource-map publications layer and what may appear in graph exports.
Related: publications-layer-plan.md · publication-data-intake-template.md · publications-quality-review.md
Registry baseline (after batch 4): 34 publications · 39 person–publication links · 6 publication–resource links (4 graph-included).
Purpose
The curated publications layer is not a DePaul bibliography. YAML records exist only when they help document facilities, equipment, labs, technical capability, computing or data practice, or cross-unit hard-tech work tied to map entities. Authorship alone is insufficient for resource linkage.
Harvest of OA Scholars and VIA (authorization: sources/inbox/2026-08-20-library-harvest-authorization/) is in scope as an archive, not as automatic YAML promotion. OpenAlex (public API) and Scopus (institutional key, if issued) are separately scoped. See the 2026-08-20 addendum in publications-layer-plan.md.
Core principles
Person–publication links
Person–publication links are acceptable when authorship is verified against an official DePaul profile, lab page, publisher metadata (Crossref/DOI), or PMC author list with DePaul affiliation.
- One or two included publications per person per batch is the default cap.
- Pre-DePaul work may be retained in the registry when it documents a faculty member’s research program, but mark
pre_depaul_work: trueand do not infer current DePaul resource use from it alone. - Common names require explicit DePaul affiliation on the publication (see publications-quality-review.md §8).
Publication–resource links
Publication–resource links require explicit evidence connecting the publication to a specific resource already in the map. These edges are rare and high-confidence — not default edges created during person intake.
- Target: roughly one publication–resource link per 5–10 included publications at the current evidence bar (batches 1–4: 6 links across 34 publications; 4 remain graph-included after quality review).
- Default for a new publication: person–publication link only;
resource_link_allowed: falseuntil methods/acknowledgment text is reviewed.
What is not sufficient for resource use
| Inference | Rule |
|---|---|
| Profile method lists alone | Not sufficient. Example: PRL-PUB-004 (Grice profile method alignment without paper acknowledgment). |
| Center affiliation alone | Not sufficient. Example: PRL-PUB-003 (STEM Center contact page without proceedings text). |
| Author affiliation alone | Not sufficient. DePaul author on a paper does not link the paper to every lab the author directs. |
| Coauthorship | Not a collaboration edge. Person–publication records authorship; they do not create person–person or cross-unit collaboration edges. |
| Grant abstracts | May guide searches (e.g., GRA-010 manuscript pointer for Kozlowski) but are not resource-use evidence unless the grant or publication text explicitly names a facility. |
| Topical similarity | Not sufficient. Same research theme (e.g., fish physiology vs Poeciliid grant species) does not create a publication–resource link. |
Graph inclusion vs registry retention
| Situation | Action |
|---|---|
| Verified authorship, strong resource-map relevance | Add publication + person–publication link with include_in_graph: true |
| Verified authorship, weak or indirect relevance | Retain registry record if useful for audit trail; set person–publication include_in_graph: false (e.g., PPL-033 Wagner → PUB-028) |
| Verified authorship, medium relevance (scholarly book, bench paper without facility text) | Person–publication include_in_graph: true; no publication–resource link |
| Unclear identity or evidence | confidence: low, needs_review in notes; include_in_graph: false until resolved |
| Clearly wrong record | Prefer needs_review or include_in_graph: false over deletion unless duplicate or fabricated |
Weak or ambiguous records should be retained as registry records only when useful, but excluded from the graph when they would mislead (false hubs, cross-unit implications, or SPARK/instrumentation overclaims).
Hard exclusion rules for publication–resource links
Do not create publication–resource links when the only support is:
- Topical similarity alone — paper topic matches a lab’s research area but does not name the lab, instrument, or facility.
- A person’s general faculty profile — profile lists techniques, equipment, or “research interests” without tying this specific publication to a resource.
- Center membership alone — person directs SPARK, STEM Center, CyberLabs, etc.; paper does not acknowledge or document that center’s infrastructure.
- Author affiliation alone — “DePaul University” or school affiliation in the author block.
- Coauthorship with a lab member — shared authorship does not prove shared facility use.
- Grant abstract keywords — grant title/abstract mentions equipment or species but publication does not acknowledge the facility.
- Weak lab association — “methods align with profile techniques,” “center is coordination hub,” or “faculty lead on resource X.”
Required: Do not create a publication–resource link unless the publication text, abstract, acknowledgment, lab page, project page, or official source explicitly connects the work to the resource.
When in doubt, leave resource_link_allowed: false and document reason_resource_link_not_added in the intake template.
Allowed publication–resource evidence
Acceptable evidence types (with quoted or linked text in evidence_text):
| Evidence type | Example from batches 1–4 |
|---|---|
| Publication explicitly acknowledges a facility, lab, instrument, center, dataset, or project | PRL-PUB-001: APS trial run named in abstract (PUB-002 → EXT-003) |
| Official DePaul news or lab communication cites facility use for the work | PRL-PUB-002: DePaul news quotes APS data collection (PUB-010 → EXT-003) |
| Official lab page lists the publication as output of that lab for this work | PRL-PUB-005/006: Karver/Southern profiles name assay development for PUB-013 |
| Publication documents an instrument, method, software, facility, or deployed system already represented as a resource | Instrument papers where methods section names beamline, detector, or shop |
| Grant/project page explicitly connects the publication to a resource | Rare; quote required — grant abstract alone is insufficient |
| Thesis/project page or thesis methods/data chapter explicitly identifies the facility, lab, instrument, storage method, or compute platform used | Acceptable for student work when the resource or practice is named; harvest metadata/abstract alone is not enough |
Profile pages: A lab profile may support a publication–resource link only when it names this specific publication as output of that lab or quotes facility use for this work — not when it only lists generic methods the PI uses.
Person–publication intake criteria
Include a publication when all apply:
- Hard-tech, computing/data practice, or resource-map utility — instrumentation, sensing, imaging, HPC, fabrication, cyber-physical systems, GIS/analytics tied to mapped resources, bench methods, documented technical capability, or evidenced data-handling/compute practice (see publications-layer-plan.md §3 and the 2026-08-20 addendum).
- DePaul connection — at least one author matches a
people.yamlrecord with verified affiliation on the publication. - Evidence in intake — title, year, venue from allowed sources; DOI required when resolvable (verify via Crossref/ORCID against discovery source).
- Conservative per-person cap — 1–2 includes per person per batch unless quality review expands.
DOI verification and storage (source packets + refactors)
| Step | Rule |
|---|---|
| Discover | Faculty profile, lab page, Google Scholar, ORCID, or candidate link |
| Verify | Crossref (or ORCID/Web of Science) confirms title, year, venue, authorship |
| Store | publications.yaml: citation fields + abstract; evidence_url = https://doi.org/… |
| Do not store | Full-text PDF or publisher HTML in packet (403 publisher snapshots are audit-only) |
| Link | Resource prose and YAML use DOI links or PUB-### registry IDs |
| Person link | Add person_publication_links.yaml when DePaul authorship verified |
| Resource link | Add publication_resource_links.yaml only with explicit facility/instrument evidence |
See also Source Evaluation and Evidence Policy § Publication registry.
ORCID on people.yaml
| Field | Rule |
|---|---|
orcid |
Normalized iD when verified or probable |
orcid_url |
Must be https://orcid.org/{orcid} |
orcid_verification_status |
verified (Crossref on linked pubs), probable (ORCID API), needs_review, unresolved, not_applicable (staff) |
| Enrichment | Run scripts/enrich_scholarly_identifiers.py; review docs/reports/scholarly-identifiers-review.md |
| Scope | Identity and authorship cross-check only — not facility or equipment evidence |
Exclude or defer (candidate_status: maybe | exclude) when:
- No peer-reviewed scholarly output at current bar (e.g., PER-112 Binder — professional credits only).
- Teaching materials, op-eds, documentaries, or policy commentary without instrumentation.
- Pure qualitative health/social-science work with no hard-tech or mapped-resource signal (cohort 4 Wagner case).
- Pre-DePaul work when a DePaul-era paper exists (defer duplicate coverage).
- Bulk API harvest without per-record review.
Batch workflow (batches 1–4 pattern)
| Step | Action |
|---|---|
| 1 | Select 8–10 people from cohort plan; document search in cohort report |
| 2 | Search official DePaul profile / lab page first; Crossref/DOI for verification |
| 3 | Fill publication-data-intake-template.md per candidate |
| 4 | Set candidate_status: include | maybe | exclude |
| 5 | Add publications.yaml + person_publication_links.yaml for includes |
| 6 | Add publication_resource_links.yaml only when allowed evidence exists (usually zero per batch) |
| 7 | Run quality review pass; downgrade overclaimed edges |
| 8 | Run python3 scripts/refresh_all.py |
Expect ~1 weak inclusion per batch at current coverage; plan a quality review after each intake.
Lessons from batches 1–4
| Batch | People searched | Included | Pub–resource added | Key lesson |
|---|---|---|---|---|
| 1 (seed) | 9 | 8 | 1 | Official-page-first works; most papers get person–pub only |
| 2 (ISE/CSH/ID) | ~14 | 7 | 2 | APS/facility text in abstract or news is strong; profile inference is weak |
| 3 (hard-tech 10) | 10 | 9 | 0 | Grant-adjacent faculty yield person–pub without facility edges — correct |
| 4 (directors/grants) | 10 | 9 | 0 | Center directors need instrumentation/methods filter; zero pub–resource is success |
Downgraded in quality review: PPL-033 (weak person–pub), PRL-PUB-003 and PRL-PUB-004 (profile/affiliation inference for resource links).
Validation checks (automated)
scripts/refresh_all.py reports:
- Publication–resource links with
include_in_graph: trueand confidence belowhigh - Publication–resource links with vague
evidence_text - Person–publication links excluded from graph (
include_in_graph: false) - Publications with no resource links (informational count — not an error)
Full-text facility-evidence pass (opt-in, gated)
Metadata (title/abstract) rarely names the specific facility a paper used. This
pass optionally downloads full text for a narrow set of candidates and
extracts facility-use evidence, so that edges like PUB → APS (uses) are backed
by a quoted Methods/Acknowledgment line rather than co-authorship inference.
This pass is the only sanctioned exception to the "do not store full text" rule in source-evaluation-policy.md § Publication registry. It is scoped, human-gated, and never runs corpus-wide by default.
Script: scripts/fulltext_facility_pass.py (accepts an explicit DOI list or a
single --person; there is no "all publications" mode).
The funnel (each gate cheaper than the next)
| Gate | Input | Action | Advances when |
|---|---|---|---|
| 1 Resource-relevant/author | OpenAlex/ORCID/Crossref metadata | existing discovery filter | DePaul map author + documented resource relevance |
| 2 Facility candidacy | title + abstract only | score named vs generic vs affiliation-only | named or ambiguous (see below) |
| 3 Acquisition | DOI | tiered full-text fetch (OA → repo/PMC/arXiv → publisher → library) | any text retrieved (success/partial) |
| 4 Extraction | Methods / Experimental / Acknowledgments / Data Availability | gazetteer match + use-verb classifier | named facility and use language |
Gate 2 — facility candidacy (metadata only)
| Signal | Candidacy | Download? |
|---|---|---|
| Named facility in title/abstract ("Advanced Photon Source", "Pohang Light Source", "PAL-XFEL", "beamline 7-ID") | named |
Yes |
| Generic capability, no named facility ("X-ray free-electron laser diffraction", "synchrotron", "EXAFS") | ambiguous |
Yes — full text needed to disambiguate |
| Facility/partner appears only as a co-author affiliation | affiliation_only |
No — never triggers download or an edge |
Partner-institution co-authorship alone is never sufficient to download a paper or to create a facility edge. This is the standing lesson from the OpenAlex pass-2 trap (a co-author affiliated with a facility ≠ the work used that facility).
Gate 3 — tiered acquisition
Cheapest and most license-safe first; stop at first success:
| Tier | Source | Automation | Notes |
|---|---|---|---|
| 1 | Unpaywall / OpenAlex best_oa_location |
automated | free, legal |
| 2 | arXiv / PMC / institutional repository | automated | common in physics |
| 3 | Publisher via DOI | automated, best-effort | HTTP 403 stays partial; do not scrape around paywalls |
| 4 | DePaul library | curator-mediated | authenticated; drop retrieved file into packet uploads/ |
Every attempt records capture_status (success / partial / failed) and a
SHA-256, matching existing capture provenance. Full-text bodies are audit
evidence under the source packet (fulltext/ or uploads/), not registry
content, and are not committed to the public repo when the license forbids
redistribution.
Gate 4 — extraction and classification
Only evidence-bearing sections are read (Methods, Experimental, Acknowledgments, Data Availability). Each facility mention is classified:
| Text pattern | Classification | Edge |
|---|---|---|
| Named facility + use verb ("data collected at APS beamline 7-ID", "experiments at the Pohang Light Source", "beamtime at PAL-XFEL") | facility_use | propose uses / produced_by |
| Named facility with no use verb; generic technique only | ambiguous | no edge; flag for human |
| Facility name appears only in the author affiliation block | affiliation_only | no edge (hard exclusion) |
The classifier keys on named facility + use language, not co-occurrence. Prefer the specific sub-facility resource (e.g. EXT-003 APS) over the parent organization (EXT-015 Argonne) when the beamline/facility is named.
Output and review gate
Extraction emits proposed publication_resource_links rows with the quote
pre-filled and:
confidence: highonly forfacility_useextraction_method: fulltext_v1include_in_graph: false,needs_review: trueuntil a curator verifies the quote against the stored text
Proposed rows are written to a report / candidate file, never directly into
data/publication_resource_links.yaml.
Person-level rollup
Person→facility "uses" is derived from confirmed pub→facility edges, so a researcher's true facility portfolio emerges (e.g. Landahl → APS, PLS, PAL-XFEL) while affiliation-only facilities (e.g. European XFEL when never used) never appear.
Facility gazetteer and missing resources
The extractor maps facility names → resource IDs via data/facility_gazetteer.yaml.
When a named, used facility has no resource record yet (e.g. Pohang Light
Source, PAL-XFEL), the pass emits a "proposed new EXT resource" rather than
silently dropping the evidence — it must not be lost.
Regenerate counts: python3 scripts/refresh_all.py