Skip to content

SWOT Analysis — HeritageGraph for npj Heritage Science (Nature portfolio)

Frame: Reviewer + author dual-perspective. Evidence cited from README, doc filenames, top-level directory structure, and recent commit history only.

1. Strengths

S1 — Full-stack, standards-aligned semantic infrastructure (CRITICAL)

Evidence: Co-presence of ONTOLOGY.md, doc_schema.owl, final_schema.yaml, oxigraph_db/, docker-compose.fuseki.yml, FUSEKI.md, and the commit "add tool to emit minimal SHACL and verify intent routes" (89f42850f) indicates a real CIDOC-CRM/OWL-DL/SHACL/PROV-O pipeline backed by both an embedded triplestore (Oxigraph) and a federated SPARQL endpoint (Fuseki). Few heritage-informatics submissions ship two graph stores plus a SHACL emitter. Leverage: Position the paper as an implemented reference architecture, not a model. Cite Oxigraph + Fuseki as evidence of openness to both embedded and federated deployment topologies.

S2 — End-to-end ingestion → curation → publish lifecycle (CRITICAL)

Evidence: Commits in the last cycle cover OCR (OCR_INTEGRATION_SUMMARY.md, feat(ingestion): … OCR preview), supervised document/tabular ingestion wizards, an entity & relationship proposal workflow (3a6cf529b, 685a42b07), reviewer triage spec (specs/006-reviewer-triage-and-approval), RDF sync signals (rdf_signals.py in dirty state), and a Cesium-based Heritage Atlas (feat(ui): add Heritage Atlas globe workspace). This is rare — most CH-KG papers ship only the schema. Leverage: Frame contribution as socio-technical pipeline, not ontology. Quantify human-in-the-loop curation throughput and reviewer SLA — this is the differentiator vs ResearchSpace/ArCo.

S3 — Domain depth on an under-represented corpus (CRITICAL)

Evidence: Project framing as Newar/Nepali heritage + TRANSLATION.md + multilingual UI artifacts (site/ne/) + CAIR-Nepal authorship. South Asian / Himalayan intangible heritage is critically under-represented in LOD literature dominated by EU/US collections. Leverage: Lead the abstract with the gap. Pair with FAIR + CARE narrative; Nature portfolio actively solicits Global South digital infrastructure work.

S4 — Reproducibility-by-construction

Evidence: Dockerfile.backend, Dockerfile.frontend, docker-compose.{prod,dokploy,coolify,fuseki}.yml, DEPLOYMENT.md, DOKPLOY.md, Makefile, entrypoint.sh, mkdocs.yml + published site/. A reviewer can run make setup and replicate. Leverage: Submit a Zenodo-archived v1 tag as supplementary; include make verify target that runs SHACL + ontology consistency.

Top 3 prioritized: S1, S2, S3.

2. Weaknesses

W1 — No declared license (CRITICAL, BLOCKING)

Evidence: README literally states "The license for this project is yet to be finalized." npj Heritage Science is fully OA and requires FAIR-compliant code + data with a declared open license. Without this, the paper is desk-rejectable. Mitigation: Adopt CC-BY-4.0 for data + Apache-2.0/MIT for code before submission; commit LICENSE and LICENSE-DATA files; add license metadata to ontology IRI; cite Zenodo DOI in the paper.

W2 — Documentation drift signalling immature research artifact (CRITICAL)

Evidence: README says backend uses "Keycloak (JWT via OIDC)" while CLAUDE.md says "NextAuth v4 + Google OAuth". Three overlapping auth docs (AUTH.md, AUTH_GUIDE.md, AUTH_ROLES_DEVELOPER_GUIDE.md). Three schema files (schema.yaml, new_schema.yaml, final_schema.yaml) co-existing. Empty files: new.owl (0 bytes), Dockerfile.keycloak (0 bytes). deleted/ directory present. Multiple cache strategies (CACHE.md + CACHING_STRATEGY_NO_REDIS.md). Mitigation: Freeze a paper-release branch; delete or quarantine stale files; collapse auth docs; pick one schema and label it canonical with a version IRI. Reviewers read repos in 2026.

W3 — Engineering scale ≠ scientific evaluation (CRITICAL)

Evidence: The 17,894-inferred-triples figure is operational throughput, not a benchmark. There is no evaluation/, no benchmarks/, no published precision/recall on Getty AAT/TGN/ULAN alignment, no inter-annotator agreement artefact in specs/006-reviewer-triage-and-approval, no SHACL conformance report committed. PLATFORM_ISSUES_CHECKLIST.md and UI_UX_AUDIT_ALL_PAGES.md suggest a product audit posture, not a research-evaluation posture. Mitigation: Add a §Evaluation with: (a) entity-linking F1 against a held-out Getty alignment gold set, (b) SHACL conformance pass-rate over the corpus, (c) reasoner-derived-triple novelty rate (how many are non-tautological), (d) reviewer agreement κ on the proposal workflow.

W4 — Active churn during submission window

Evidence: Recent commits (feat(cidoc): relationship predicates and binary HeritageAssertion fields, "Restructuring oxigraph components", "Final testing of oxigraph") indicate ontology + triplestore are still moving. specs/004-yaml-driven-schema, specs/007-entity-relationship-proposals are open spec directories. Mitigation: Cut a tagged paper-v1.0 release; freeze ontology IRI version; document any post-submission changes in a CHANGELOG referenced in the paper.

W5 — Newar-specific scope undermines generality claims

Evidence: README + commit history are entirely Newar/Nepali. No multi-region pilot, no cross-collection federation demonstration in commits. Mitigation: Reframe scope as exemplar deployment of a generalizable architecture, not a universal platform. Add a §Transferability subsection with one swappable component diagram.

Top 3 prioritized: W1, W2, W3.

3. Opportunities

O1 — Global-South digital heritage gap (CRITICAL)

South Asian / Newar / Himalayan heritage is structurally absent from LOD ecosystems (Europeana, ResearchSpace, ArCo, WissKI). npj Heritage Science explicitly solicits non-EU/non-US heritage infrastructure work. Leverage: Lead the framing. Cite the gap quantitatively (e.g., proportion of CIDOC-CRM-aligned datasets covering Nepal/South Asia in LOD Cloud).

O2 — LLM-assisted curation is an emerging publishable angle (CRITICAL)

Evidence: AGENTS.md, SKILLS.md, .cursor/, .specify/ indicate substantial AI-agent-driven workflow. This is a hot venue topic in 2025–26 (LLM + KG construction for cultural heritage). Leverage: Add a small ablation: ingestion throughput with vs without agent-assisted curation; F1 of LLM-suggested CIDOC class assignments.

O3 — Federation/Linked-Data-Fragments demo (CRITICAL)

Evidence: Co-presence of Fuseki + Oxigraph + Getty alignments + Cesium Atlas. A small SPARQL federation demo (HG ↔ Wikidata ↔ Getty AAT) would directly answer the "so what?" reviewer question. Leverage: Build one published SPARQL query that traverses 3 endpoints and resolves a Newar entity end-to-end. Include the URL in the paper.

O4 — FAIR + CARE alignment for indigenous/community heritage

The CARE principles (Collective benefit, Authority to control, Responsibility, Ethics) for Indigenous data are gaining traction at Nature portfolio. Newar community heritage is a natural fit. Leverage: Add a §FAIR+CARE compliance table; cite TK Labels.

Top 3 prioritized: O1, O2, O3.

4. Threats

T1 — Established competitor platforms set the bar (CRITICAL)

ResearchSpace (British Museum, CIDOC-CRM-native), ArCo (Italian MiBACT, ~169M triples), WissKI (German CH consortium), Linked.art. Reviewers will demand a head-to-head table. Mitigation: Include a comparison matrix (ontology depth, reasoning, license, deployment model, languages, OCR, agentic curation). Concede where they win (scale); claim where you differ (agent-assisted curation + Global South + reproducibility).

T2 — Reasoning/SHACL claim is fragile under change (CRITICAL)

HermiT over an evolving OWL-DL ontology + SHACL shapes can quietly drift to inconsistency. With three schema files committed and active ontology commits, a reviewer running the consistency check at review time may get a different result than reported. Mitigation: Freeze ontology version IRI; commit reasoner logs and SHACL conformance reports as supplementary data; add CI that re-runs reasoning on every commit and refuses merges that change inferred-triple count without explanation.

T3 — Ethics, FPIC, and repatriation sensitivity (CRITICAL)

Publishing identifying/location data on Newar heritage objects (especially given Nepal's contested artifact-repatriation context) without documented community consent and TK Labels invites rejection on ethics grounds — increasingly enforced at Nature portfolio. Mitigation: Document the FPIC / community-engagement process; apply Local Contexts TK Labels; redact sensitive geocoordinates by default; cite community partners (e.g., Department of Archaeology Nepal, Newar heritage trusts) as co-authors or acknowledged stakeholders.

T4 — Sustainability question

CAIR-Nepal is a small organization. Reviewers ask: who maintains the IRIs in 10 years? Persistent identifier strategy? Mitigation: Register a w3id.org or PURL prefix; mirror to Zenodo; name an institutional steward.

Top 3 prioritized: T1, T2, T3.

5. Reviewer Attack Vectors (npj HS-flavored) & Preemption

┌──────┬────────────────────────────────────────────────────────────────────────────────────┬───────────────────────────────────────────────────────────────────────────────────────────┐ │ # │ Likely objection │ Preempt in paper │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA1 │ "What is the scientific contribution beyond engineering?" │ Lead Section 1 with the gap claim (Global South + agent-assisted curation), not the stack │ │ │ │ list. Move infrastructure description to §3. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA2 │ "How does this differ from ResearchSpace, ArCo, WissKI, Linked.art?" │ Mandatory §Related Systems with a quantitative comparison table. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA3 │ "17,894 inferred triples — what is novel vs tautological CRM expansion?" │ Report a novelty rate: % of inferred triples not derivable by single-step CRM property │ │ │ │ chains. Show the histogram. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA4 │ "Where is the entity-linking evaluation against Getty AAT?" │ Provide a gold-standard 200-entity sample with κ and F1. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA5 │ "FAIR compliance: PIDs, license, machine-readable metadata?" │ Add §FAIR self-assessment table; resolve W1 (license) and add a w3id namespace. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA6 │ "Reproducibility: docker image hashes, ontology version IRI, Zenodo DOI?" │ Cut a paper-v1.0 tag; archive on Zenodo; cite DOI in §Data Availability. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA7 │ "Ethics / FPIC / TK Labels for community heritage?" │ Add §Ethics with community-partner statement and TK label coverage. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA8 │ "Sample size and coverage of the deployed instance?" │ Report counts per CRM class, geographic coverage map, time-period histogram, growth │ │ │ │ curve. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA9 │ "Why CIDOC-CRM + LinkML + OWL-DL — isn't this redundant?" │ Justify the layering: LinkML for forms/validation, OWL-DL for reasoning, SHACL for │ │ │ │ closed-world conformance. Cite Linked.art's similar three-layer choice. │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA10 │ "Multiple schema files in the repo (new_schema.yaml, final_schema.yaml) — which is │ Resolve W2 before review; pin one schema; add a SCHEMA.md declaring the canonical IRI. │ │ │ canonical?" │ │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA11 │ "NextAuth + Google OAuth tied to a US-vendor identity provider for a heritage │ Address sovereignty concern; document that institutional SSO / Keycloak is supported │ │ │ repository?" │ (already in keycloak/ directory). │ ├──────┼────────────────────────────────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────────────────────────────┤ │ RA12 │ "Reasoner = HermiT but the project also ships Oxigraph (no OWL reasoning) and │ Diagram the reasoning pipeline; state whether inference is materialized in the │ │ │ Fuseki — where is reasoning actually performed in production?" │ triplestore or computed offline. │ └──────┴────────────────────────────────────────────────────────────────────────────────────┴───────────────────────────────────────────────────────────────────────────────────────────┘

6. Weakness → Threat Exposure Map

┌─────────────────┬──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┬─────────────────────────────────────────────────┐ │ Weakness │ Exposes us to threats │ Why │ ├─────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────┤ │ W1 No license │ T1 (competitor comparison: they all have licenses), T3 (cannot make CARE/TK Label claims without an IP │ License is foundational; absence amplifies │ │ │ framework), T4 (no license = no sustainable downstream reuse) │ every other criticism. │ ├─────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────┤ │ W2 Doc/schema │ T2 (reviewers run the code and get a different graph than reported), T1 (competitors look mature by contrast) │ Drift makes the reasoning claim unverifiable. │ │ drift │ │ │ ├─────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────┤ │ W3 No │ T1 (no quantitative basis to claim differentiation), T2 (no SHACL conformance baseline to anchor reasoning │ Without evaluation, every comparison reduces to │ │ evaluation │ claim) │ opinion. │ ├─────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────┤ │ W4 Active churn │ T2 (reasoning brittleness), T1 (moving target undermines comparison) │ A paper describes a moving artifact. │ ├─────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────┤ │ W5 Newar-only │ T1 (competitors will be framed as general; you as narrow) │ Without a transferability section, scope is a │ │ scope │ │ liability. │ └─────────────────┴──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┴─────────────────────────────────────────────────┘

7. Strength → Threat-to-Opportunity Conversion

┌────────────────────────────────────┬────────────────────────┬────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐ │ Strength │ Converts threat │ Into opportunity │ ├────────────────────────────────────┼────────────────────────┼────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤ │ S1 Standards-aligned semantic │ T1 "what's novel?" │ O3 SPARQL federation demo: only a CIDOC-CRM + Getty + Fuseki+Oxigraph stack supports a 3-endpoint federation query │ │ stack │ │ end-to-end. │ ├────────────────────────────────────┼────────────────────────┼────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤ │ S2 End-to-end pipeline │ T1 competitor │ O2 LLM-assisted curation paper: existing platforms ship the schema, not the agent-assisted ingestion workflow. │ │ │ comparison │ │ ├────────────────────────────────────┼────────────────────────┼────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤ │ S3 Newar/Nepali corpus │ T3 ethics scrutiny │ O1 + O4 FAIR+CARE Global South exemplar: turn ethics from threat into the paper's central thesis. │ ├────────────────────────────────────┼────────────────────────┼────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤ │ S4 Reproducibility-by-construction │ T2 reasoning fragility │ A make verify target + CI gate operationalizes the reasoning claim, turning a fragility threat into a methodological │ │ │ │ contribution. │ └────────────────────────────────────┴────────────────────────┴────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

8. SWOT 4×4 Strategy Matrix

Column 1: Strengths (S1–S4) Opportunities (O1 Gap • O2 LLM-curation • O3 Federation • O4 CARE): SO — Aggressive moves • S1+S3 → O1: lead with "first CIDOC-CRM-aligned, OWL-DL-reasoned KG for Newar heritage" • S2 → O2: report agent-curation ablation as a contribution • S1 → O3: ship one federated SPARQL query (HG ↔ Wikidata ↔ Getty) in the paper • S4 → O4: a make audit target that prints FAIR+CARE compliance evidence Threats (T1 Competitors • T2 Reasoning fragility • T3 Ethics • T4 Sustainability): ST — Defensive moves • S1 vs T1: comparison table where standards-completeness is your axis • S4 vs T2: pinned ontology IRI + CI reasoner gate; ship reasoner-report.ttl • S2+S3 vs T3: document the community-in-the-loop proposal workflow as the ethics mechanism • S4 vs T4: Zenodo DOI + w3id

PID + named institutional steward

──────────────────────────────────────── Column 1: Weaknesses (W1–W5) Opportunities (O1 Gap • O2 LLM-curation • O3 Federation • O4 CARE): WO — Catch-up moves • W1 → O1/O4: pick CC-BY + CARE-compatible TK labels in one stroke (license decision becomes a contribution) • W3 → O2: the missing evaluation can be the LLM-curation evaluation (reuse the work) • W5 → O3: federation demo itself widens scope without writing more Newar data Threats (T1 Competitors • T2 Reasoning fragility • T3 Ethics • T4 Sustainability): WT — Survival moves • W1+W2 vs T1: a 1-week "submission-hardening sprint": freeze branch, finalize license, collapse auth docs, delete empty files, pin one schema • W3 vs T2: minimal acceptable evaluation = SHACL conformance % + 200-entity Getty alignment F1 + reasoner novelty rate • W4 vs T2: cut paper-v1.0 tag, declare it the artifact-of-record; all later commits go in a CHANGELOG • W2+W3 vs T1: do not submit to npj HS until W1+W2 are resolved; soft-launch as a preprint first

9. Critical Path Before Submission

In order, no skipping:

  1. Choose and commit license (W1). Resolves T1, T3, T4 partial.
  2. Cut paper-v1.0 release branch and archive on Zenodo (W4, T2, T4).
  3. Collapse documentation drift: one auth doc, one canonical schema, delete new.owl/Dockerfile.keycloak, retire deleted/ (W2).
  4. Build the minimal evaluation: SHACL pass-rate, Getty alignment F1, reasoner novelty rate, inter-annotator κ (W3, T1, T2).
  5. Run one 3-endpoint SPARQL federation query and screenshot/document it (O3, ST move).
  6. Draft §Ethics with FPIC + TK Labels + community-partner acknowledgment (T3, O4).
  7. Write §Related Systems comparison table vs ResearchSpace/ArCo/WissKI/Linked.art (T1).

Generated from README, doc filenames, top-level structure, and recent commit history only. No source files inspected.