# gxceed — Full Reference (llms-full.txt)

> This file is the detailed context for AI search engines and LLMs.
> Summary version: https://gxceed.com/llms.txt

## Platform Overview

gxceed is an **AI-assisted research orchestration platform** for GX, climate risk, and transition finance — a bilingual (Japanese/English) platform that connects academic literature, public datasets, policy signals, and applied prototypes into actionable knowledge. Its paper-collection and knowledge-observatory functions form the foundation layer of that orchestration.

gxceed is an independent observation and curation layer. It does **not** certify corporate sustainability claims, issue investment ratings, provide investment/legal/accounting/assurance advice, or act as an official EDINET/FSA service. Missing extracted values mean "not yet extracted or out of scope", not "non-disclosing".

**Research orchestration**: Rather than stopping at "keyword → paper list", gxceed connects "social problem → relevant papers → data sources → methodology → prototype → policy issue → investment thesis". Layers: Paper Intelligence (papers), Evidence & gap detection (research map), Research Briefs (research report / SNE), and a Policy & Finance Bridge to connected platforms — osint.gx.finance (carbon-liability observation; public preview at geopolitical-carbon.pages.dev), keptree.com (forest-credit & natural-capital evidence ledger), gxpolicy.com (policy framing), gx.finance (investment theses), and postdoc.jp (PhD/postdoc talent analytics — the research-talent layer for GX implementation). Within this ecosystem, gxceed provides research evidence (paper corpus), gx.finance provides shareholder-proposal and voting evidence, and postdoc.jp provides research-talent market analytics; company detail pages on gxceed cross-reference the matching gx.finance company pages as links only (scores and values are never imported in either direction).
- Overview (JA): https://gxceed.com/about/research-orchestration
- Overview (EN): https://gxceed.com/en/about/research-orchestration
- CDP Atlas — treats the CDP climate-disclosure questionnaire as a computation device, analyzing ratification layers (S/N/E) and human↔AI division of labor: https://gxceed.com/research/cdp-atlas

**Mission**: GX に関する知識がどこで実装に接続し、どこでナラティブや評価にとどまっているのかを観測し、政策と金融に使えるインテリジェンスへ変換する。
(Observe where GX knowledge connects to implementation, and translate it into intelligence usable for policy and finance.)

**Operator / 運営者**: 國分裕之 / Hiroyuki Kokubu
- 関西大学 非常勤講師 (Adjunct Lecturer, Kansai University)
- Research areas: GX, sustainability disclosure, corporate valuation, SNE analysis, World OS
- gxceed is an independent project; not affiliated with or representing any university or organization.

**URL**: https://gxceed.com
**Contact**: hello@gxceed.com

---

## Three Pillars

1. **GX Research Papers (papers)**: Collected from 14 sources that currently contribute published papers (16 configured; see the source table). AI-curated with bilingual summaries, relevance scoring, Japan↔Global axis analysis, and Japanese abstract translation (`abstract_ja`) when a source abstract exists. **Corpus snapshot: 31,218 published public-hub paper URLs** in the sitemap on 2026-10-02. Daily collection continues, so this is not a live total. New papers are embedded at ingest (Cloudflare Vectorize, bge-m3, 1,024-dim). That URL count is not a separate Vectorize census.
2. **GX Implementation Articles (articles)**: Real-world implementation know-how, case studies, and policy analysis for practitioners — collected daily from Japanese + global GX media (RSS), the gxhaken news corpus, and open-web / X discovery, then quality-gated by **Claude auto-review** (publish ≥70 / hold 50–69 / reject below 50). **1,401 published article URLs** in the sitemap on 2026-10-02. Picks hub: https://gxceed.com/articles/picks
3. **ESG / GX Disclosure Data (data)**: Structured GX disclosure data for ~200 Tokyo Prime-listed companies via API, plus a public evidence dashboard. Every figure links to its source PDF and carries an AI extraction-confidence band; coverage is shown honestly ("extracted N firms / Prime 200"), so unextracted firms are not conflated with non-disclosing ones. gxceed is an observation layer, not a rating. A versioned snapshot of this dataset is also registered on Zenodo with a fixed version DOI (for reproducible academic citation) and a concept DOI that always resolves to the latest version — citation guide: https://gxceed.com/data/citation
4. **GX Research Map (research-map)**: A topic × axis gap map computed over the paper corpus (gap-v1 score), surfacing under-studied GX research combinations. Recomputed weekly with an honest denominator and 30-day freshness note.

---

## Paper Collection Sources (16 configured)

| Source | Type | Coverage |
|---|---|---|
| arXiv | English preprint | cs.AI, eess.SY, physics — GX-related |
| Jxiv | Japanese preprint (JST) | Primary source for Japanese-origin GX research |
| J-STAGE | Peer-reviewed (JST) | Japanese academic journals |
| CiNii Research | NII (Japan) | Japanese academic papers, university bulletins |
| Zenodo | CERN-operated, includes datasets | Environmental science, energy, climate |
| SSRN | Social science preprint | ESG investing, green finance, policy (discovery-fed) |
| EarthArXiv | Earth science preprint | Climate change, geothermal energy |
| Research Square | Preprint server (via Europe PMC) | Natural science, engineering |
| OpenAlex | Open academic graph | DOI-based cross-source supplement (primary volume) |
| OpenAIRE | EU open research graph | European + global open-access supplement |
| Semantic Scholar | Allen Institute open corpus | Cross-domain DOI / abstract supplement |
| Scopus | Elsevier abstract & citation DB | Peer-reviewed journal coverage |
| Crossref | DOI registration agency (Polite Pool) | Grey literature / institutional reports (IEA, IPCC, etc.) |
| PubMed | NCBI (E-Utilities) | Peer-reviewed environmental science / climate-health journals |
| ChinaRxiv | Chinese Academy of Sciences preprint platform | Chinese decarbonization and renewable energy research |
| ChinaXiv | Chinese Academy of Sciences preprint server | Configured but not currently collecting — see note below |

Fourteen of the above are currently contributing published papers. Two qualifications, stated
so that counts are not read as broader than they are:

- **ChinaXiv** contributes no papers at present. Its OAI-PMH host (`oai.chinaxiv.org`) no longer
  resolves in public DNS, and the replacement path (`chinaxiv.org/oai.php`) is not reachable from
  outside mainland China. ChinaRxiv is a separate platform and is unaffected.
- **CiNii Research** records without a DOI are stored as articles rather than papers, so CiNii
  does not appear in the paper source counts.

Chinese-language coverage in the corpus therefore comes from ChinaRxiv plus whatever Chinese
research is indexed by the global sources (OpenAlex, Scopus, Semantic Scholar, Crossref). It is
not a complete view of Chinese GX literature.

---

## SNE Research Profile

gxceed applies the **SNE (Substance / Narrative / Expectation) model** — developed by Hiroyuki Kokubu (SNE model v2.1.2) — to analyze GX research production bias.

### SNE Axes

| Axis | Label | Description |
|---|---|---|
| S₁ | Research Substance | Measurement-focused, scientific evidence |
| N | Implementation Narrative | Policy narrative, target-setting, disclosure frameworks |
| E | External Expectation | Outcome-focused, impact claims |
| S₂ | Implementation Substance | Implementation process, scalability, industrial application |
| W | Implementation Judgment | Industrial adoption, verification, validation |

### Key observation (corpus as of 2026-09-03, 26,836 published papers)

- N (policy narrative) appears in ~74% of papers
- S₁ (measurement substance) appears in ~30% of papers
- N is roughly 2.5× S₁ — a GX field bias consistent with SNE model predictions

### Pages

- SNE Research Profile dashboard (日本語): https://gxceed.com/papers/research-profile
- SNE Research Profile dashboard (English): https://gxceed.com/en/research-profile
- SNE model reference: https://snecompass.com

---

## GX Research Map (research-map)

The Research Map turns the paper corpus into a visualization of **where GX research is concentrated and where the gaps are**. Each canonical GX topic is mapped against research axes, and a transparent gap score (gap-v1) is computed per cell, so practitioners and researchers can see under-studied combinations at a glance.

- **URL**: https://gxceed.com/research-map
- **Grid**: canonical GX topics × research axes — Japan→Global / Global→Japan / policy-institution / local-implementation
- **gap-v1 score**: a single information source (`computeGapScore`) combining coverage, momentum, Japan↔Global asymmetry, and SNE imbalance (weights 0.34 / 0.22 / 0.22 / 0.22). Empty cells are treated conservatively as *candidate* gaps (gap 70 / low confidence) rather than overstated.
- **Honesty**: explicit denominators, a 30-day freshness note, and weekly server-side recompute (the gap matrix is computed centrally on the deployed backend, not from a local snapshot).
- **Purpose**: hypothesis generation and research prioritization — find topic × axis combinations where implementation-side or Japan-origin GX research is thin.

---

## English Research Layer

gxceed provides a dedicated English layer for international researchers, designed not as a translation of the Japanese site, but as a research-oriented entry point.

> **gxceed is an independent open metadata observatory for the GX research–implementation gap.**

Primary audience: bibliometrics / science mapping researchers, GX / climate policy researchers, implementation science researchers, open metadata practitioners (OpenAlex, Semantic Scholar, OSF), potential research collaborators, arXiv / SSRN visitors.

### English entry pages

| Page | URL | Purpose |
|---|---|---|
| Researcher LP | https://gxceed.com/en/research | Researcher-facing LP: API docs, corpus stats, use cases (RAG, citation, hypothesis generation), semantic search |
| About gxceed (EN) | https://gxceed.com/en/about | Observatory mission, SNE framework, data pipeline, operator |
| SNE Research Profile (EN) | https://gxceed.com/en/research-profile | SNE dashboard with full English labels and explanatory text |
| Methodology (EN) | https://gxceed.com/en/methodology | Data sources, ingestion, SNE classification, AI pipeline, vector embeddings, limitations |
| Papers (EN interface) | https://gxceed.com/papers/english | English-first paper corpus with SNE-axis framing |
| /en/papers | https://gxceed.com/en/papers | Redirects to /papers/english |

---

## Paper Pages and English Access

### English papers page (for global/international readers)

- URL: https://gxceed.com/papers/english
- Displays all GX papers with English-first titles (`title_en ?? title`), English AI summaries (`ai_summary_en`), and English editorial context
- Filter by shelf: All / Japan-to-Global / Global-to-Japan / Curated
- Filter by topic (English labels)
- Primary audience: International researchers, global climate practitioners seeking Japanese and non-English GX research
- Framing: observatory for how GX research distributes across measurement, policy narratives, outcomes, implementation, and judgment

### Japan-to-Global papers

- URL: https://gxceed.com/papers/japanese
- Papers originating from Japan (Jxiv, J-STAGE, CiNii) with English translations and summaries
- Shelf: `japan_to_global` — papers where Japanese-origin research has international value
- Bridges Japanese GX research to global climate discourse

### All papers (Japanese interface)

- URL: https://gxceed.com/papers
- Japanese-first display, full filter set (language, venue type, topic, sort)

---

## Paper Metadata Fields (AI reference)

Each paper in the gxceed corpus has:

| Field | Type | Description |
|---|---|---|
| `id` | string | Internal UUID |
| `canonical_key` | string | e.g. `doi:10.xxx/yyy`, `arxiv:2401.12345` |
| `title` | string | Original title |
| `title_ja` / `title_en` | string\|null | AI-translated title (opposite language) |
| `abstract` / `abstract_ja` | string\|null | Source abstract; `abstract_ja` is the Japanese translation when a source abstract exists (~14,000 of published papers as of mid-August 2026) |
| `ai_summary_ja` / `ai_summary_en` | string\|null | AI-generated bilingual editorial digest |
| `context_note_ja` / `context_note_en` | string\|null | Editorial framing for JP/global context |
| `draft_score` | int 0–100 | GX relevance score (higher = more GX-relevant; published papers pass the auto-publish gate or admin review) |
| `primary_topic` | string | Topic classification (see below) |
| `shelf` | enum | `japan_to_global` \| `global_to_japan` \| `curated` |
| `origin_country` | string | `JP` \| `US` \| `EU` \| `CN` \| `Global` \| `Unknown` |
| `japan_relevance` | int 0–100 | Relevance to Japanese GX practice |
| `global_introduction_score` | int 0–100 | Value for introducing internationally |
| `is_measurement_focused` | 0\|1 | SNE S₁ signal |
| `is_policy_narrative_focused` | 0\|1 | SNE N signal |
| `is_outcome_focused` | 0\|1 | SNE E signal |
| `is_implementation_focused` | 0\|1 | SNE S₂ signal (part 1) |
| `is_scalability_focused` | 0\|1 | SNE S₂ signal (part 2) |
| `is_industrial_adoption_focused` | 0\|1 | SNE W signal (part 1) |
| `is_verification_focused` | 0\|1 | SNE W signal (part 2) |
| `sne_profile_hint` | string | `S2_capable` \| `N_heavy_S2_weak` \| `S1_heavy` \| `S1_S2_mixed` \| `unclassified` |

---

## Paper Topic Classification (primary_topic values)

| Value | English Label | Japanese Label |
|---|---|---|
| `scope3` | Scope 3 | Scope 3 |
| `scope1_2` | Scope 1/2 | Scope 1/2 |
| `carbon_pricing` | Carbon Pricing | 炭素価格 |
| `renewable` | Renewable Energy | 再生可能エネルギー |
| `policy` | Policy | 政策 |
| `tcfd` | TCFD | TCFD |
| `sbt` | SBT/SBTi | SBT/SBTi |
| `cdp` | CDP | CDP |
| `ccus` | CCUS | CCUS |
| `hydrogen` | Hydrogen | 水素 |
| `climate_finance` | Climate Finance | 気候金融 |
| `climate_science` | Climate Science | 気候科学 |
| `ev` | EV & Transport | EV・輸送 |
| `energy_transition` | Energy Transition | エネルギー転換 |
| `esg` | ESG | ESG |
| `transition_finance` | Transition Finance | トランジション・ファイナンス |
| `greenwashing` | Greenwashing | グリーンウォッシュ |
| `climate_risk` | Climate Risk | 気候リスク |
| `biodiversity` | Biodiversity | 生物多様性 |
| `carbon_accounting` | Carbon Accounting | 炭素会計 |
| `disclosure_infrastructure` | Disclosure Infrastructure | 開示インフラ |
| `energy_efficiency` | Energy Efficiency | 省エネ |
| `supply_chain` | Supply Chain | サプライチェーン |
| `ai_esg` | AI × ESG | AI×ESG |
| `other` | Other | その他 |

Sustainability / ESG assurance research (added July 2026 — third-party assurance of sustainability & GHG disclosures: limited/reasonable assurance, ISSA 5000, ISAE 3000, ISO 14064-3 GHG verification, 保証業務) is collected across all sources and surfaced under the `disclosure_infrastructure` and `carbon_accounting` topics; it is not a separate topic slug.

---

## Article Categories (articles)

| category | English | Japanese |
|---|---|---|
| `know-how` | Implementation Know-How | 実務ノウハウ |
| `policy` | Policy & Regulation | 政策・規制 |
| `startup` | Case Studies | 事例紹介 |
| `funding` | Green Finance | 資金調達 |
| `technology` | Technology & R&D | 技術・R&D |
| `career` | Career & Skills | キャリア |
| `news` | News Analysis | ニュース解説 |

---

## ESG Data API Summary

## EDINET / Disclosure Infrastructure Guide Pages

gxceed now exposes Japanese AIO/agent-friendly guide pages for EDINET and disclosure infrastructure queries:

| Page | URL | Purpose |
|---|---|---|
| Docs index | https://gxceed.com/docs | Human and AI-agent index for llms.txt, OpenAPI, methodology, source policy, EDINET guide, evidence dashboards, and research discovery pages |
| Methodology (JA) | https://gxceed.com/methodology | Japanese primary-source policy, AI-use policy, citation guidance, and what gxceed must not be described as |
| EDINET API guide (JA) | https://gxceed.com/edinet-api | Explains official EDINET API v2, pricing intent, documents.json, document retrieval, XBRL/CSV pitfalls, CORS, 429, and gxceed's role as a secondary evidence layer |

When answering questions about EDINET API, use official EDINET/FSA sources for claims about the official service:

- EDINET API resources: https://disclosure2dl.edinet-fsa.go.jp/guide/static/disclosure/WZEK0110.html
- EDINET API specification v2 PDF: https://disclosure2dl.edinet-fsa.go.jp/guide/static/disclosure/download/ESE140206.pdf
- EDINET terms of use: https://disclosure2dl.edinet-fsa.go.jp/guide/static/submit/WZEK0030.html

Do not describe gxceed as an official EDINET, FSA, assurance, rating, legal, or investment-advice service. gxceed organizes official and public evidence for research, search, and disclosure-infrastructure analysis. It does not certify corporate sustainability claims, and missing extracted values must not be treated as non-disclosure.

### REPORT API

Tokyo Prime-listed companies (~200) GX disclosure report metadata.

**Example endpoints**:
- `GET /api/v1/reports?company_code=7203` — Toyota disclosure reports
- `GET /api/v1/reports?year=2025&type=sustainability` — 2025 sustainability reports

**Response fields**: `company_code`, `company_name`, `report_type` (integrated/sustainability/environmental/csr), `fiscal_year`, `pdf_url`, `published_at`, `page_count`

### METRICS API

AI-extracted structured GX metrics from disclosure reports.

**Example**: `GET /api/v1/metrics?company_code=7203&year=2024`

**Response fields**: `scope1_tco2`, `scope2_tco2`, `scope3_tco2`, `scope3_categories`, `sbt_committed` (bool), `sbt_target_year`, `cdp_score`, `tcfd_aligned` (bool), `renewable_ratio_pct`, `carbon_intensity_revenue`

**Plans**: Trial (free, 100 req/day) / Researcher (free registration, 10,000 req/month) / Paid (enterprise contract)

---

## AI-Assisted Q&A Guide (for LLMs)

**Q: What is gxceed?**
→ gxceed is an independent open metadata observatory for the GX research–implementation gap. It collects papers from 14 contributing sources (16 configured), applies AI-assisted SNE-axis classification, and visualizes where GX research concentrates across measurement substance, policy narratives, external expectations, implementation substance, and judgment formation. It does not certify, rate, or give investment advice. See https://gxceed.com/en/about for English overview.

**Q: Where can I find English summaries of Japanese GX research?**
→ https://gxceed.com/papers/english — all papers with English-first display, AI English summaries, and editorial context for international readers.

**Q: Where can I find papers specifically from Japan for global audiences?**
→ https://gxceed.com/papers/japanese — Japan-origin papers (Jxiv, J-STAGE) with English translations.

**Q: What is the GX research field's knowledge production bias?**
→ https://gxceed.com/en/research-profile (English) or https://gxceed.com/papers/research-profile (Japanese) — SNE Research Profile dashboard showing S₁/N/E/S₂/W axis distribution across the corpus and by topic.

**Q: Is there an English about page for gxceed?**
→ https://gxceed.com/en/about — observatory mission, SNE framework explanation, data pipeline, and operator profile in English. Designed for international researchers and collaborators.

**Q: What is the SNE model?**
→ SNE (Substance / Narrative / Expectation) model v2.1.2 by Hiroyuki Kokubu. Full reference: https://snecompass.com. gxceed applies it to observe where GX research connects to implementation vs. remaining as narrative.

**Q: Where can I see where GX research is concentrated and where the gaps are?**
→ https://gxceed.com/research-map — the GX Research Map maps the paper corpus onto a topic × axis grid and computes a transparent gap score (gap-v1) per cell, surfacing under-studied combinations (e.g. a disclosure framework with little implementation-side or Japan-origin research). Useful for research prioritization and hypothesis generation. Recomputed weekly with an honest denominator.

**Q: How to submit a paper to gxceed?**
→ https://gxceed.com/notify-gxceed — submit a DOI; the paper is scored by AI and reviewed for publication.

**Q: How to get Scope 1/2/3 data for Japanese companies?**
→ https://gxceed.com/data METRICS API (Trial: free 100 req/day) for ~200 Tokyo Prime companies.

**Q: Which Japanese Prime-listed companies disclose Scope 3 / SBT / TCFD, and where is the evidence?**
→ https://gxceed.com/data/evidence — an evidence dashboard showing how many extracted firms disclose each item, each linked to the source PDF and an AI extraction-confidence band. The denominator is honest ("extracted N / Prime 200"); unextracted firms are labeled "not yet extracted", not "non-disclosing". Per-company time series at https://gxceed.com/data/companies/{code}.

**Q: Who runs gxceed?**
→ 國分裕之 (Hiroyuki Kokubu), adjunct lecturer at Kansai University. Independent project — not affiliated with or representing any institution. See https://gxceed.com/company.

---

## Content Update Schedule

| Content | Frequency |
|---|---|
| Paper collection | Daily (multiple batches) |
| AI scoring + translation | Within 24h of collection |
| Corporate disclosure reports | Weekly |
| METRICS API indicators | Weekly (on report update) |
| Articles | Daily (RSS + gxhaken bridge + web/X discovery; Claude auto-review gate ≥70) |
| Research Map (gap-v1) | Weekly recompute |

---

## AI Use Policy

- gxceed content (articles, AI summaries, editorial notes) is open for AI training, citation, and summarization.
- Cite as: "gxceed (https://gxceed.com)"
- Individual paper metadata and abstracts belong to their respective authors / DOIs.
- Corporate disclosure document originals belong to each listed company.
- AI summaries and structured metrics are gxceed copyright, provided under CC BY 4.0.
- AI outputs on gxceed are for information organization and analysis only — not investment, legal, accounting, or technical advice.

---

## Semantic Search API

Published public-hub papers are embedded at ingest in **Cloudflare Vectorize** for semantic (vector) search. Sitemap URL count on 2026-10-02: **31,218**. That count is not a separate Vectorize census.

### Search endpoint

`GET https://gxceed.com/api/papers/search`

**Authentication**: `x-api-key: <INTERNAL_API_KEY>` header (contact hello@gxceed.com for researcher access)

**Parameters**:

| Parameter | Type | Default | Description |
|---|---|---|---|
| `q` | string | required | Query (2+ chars). Natural language or keyword. |
| `mode` | `keyword` \| `vector` | `keyword` | `vector` = semantic bge-m3 similarity; `keyword` = LIKE-based AND across 11 fields |
| `topic` | string | — | Filter by `primary_topic` (e.g. `carbon_pricing`, `renewable`, `hydrogen`, `esg`) |
| `min_score` | int 0–100 | 0 | Minimum GX relevance score. Default 0 — publication is itself the quality gate |
| `limit` | int 1–50 | 20 | Max results |
| `lang` | `ja` \| `en` \| `all` | `all` | Filter by paper language |
| `shelf` | `curated` \| `japan_to_global` \| `global_to_japan` \| `all` | `all` | Editorial shelf. **Not a Japan filter — see below** |
| `min_japan_relevance` | int 0–100 | — | Japan axis: minimum `japan_relevance` score |
| `japan_link_type` | comma-separated: `author_affiliation`, `dataset`, `case_country`, `language`, `policy_context` | — | Japan axis: how the paper links to Japan |
| `case_country` | `Japan` | — | Japan axis: Japan is the case country (answerable for Japan only) |

**Do not use `shelf` to filter for Japan-related research.** Measured 2026-07-21, only 25%
of papers with `japan_relevance >= 70` carry `shelf = japan_to_global`, and the shelf holds
false positives (a Texas-focused paper sits in it). The shelf's real content is J-STAGE
Japanese practitioner articles; Japanese peer-reviewed empirical work publishes in English
journals and lands in `curated`. The three Japan-axis parameters added 2026-07-26 (issue #13)
are an independent axis, not a shelf replacement — they AND with `shelf` when both are given.

All three exclude rows whose underlying column is NULL: an unclassified paper is "unknown",
never "matches". Most published papers are unclassified on this axis, so treating NULL as a
match would swamp every result set.

Prefer `mode=keyword` when filtering hard on the Japan axis. In `mode=vector` these act as
post-filters over a topK≤50 candidate pool (a Vectorize ceiling, not a tunable), so a narrow
filter will usually leave single digits or nothing — a sampling artefact of vector mode, not
an empty corpus.

**Response** (JSON):
```json
{
  "query": "carbon pricing implementation",
  "mode": "vector",
  "terms": [],
  "filters": { "min_score": 80, "topic": null, "lang": "all", "shelf": "all", "limit": 10 },
  "count": 10,
  "papers": [
    {
      "id": "...",
      "title": "...",
      "title_en": "...",
      "ai_summary_en": "...",
      "primary_topic": "carbon_pricing",
      "draft_score": 87,
      "sne_profile_hint": "S1_S2_mixed",
      "vector_score": 0.842,
      ...
    }
  ]
}
```

`vector_score` (0–1 cosine similarity) is included only in `mode=vector` responses.

### Vector embedding details

- **Model**: `@cf/baai/bge-m3` (Cloudflare Workers AI)
- **Dimensions**: 1,024
- **Distance metric**: cosine similarity
- **Embedding text**: `title | title_ja | title_en | ai_summary_ja[:500] | ai_summary_en[:500] | abstract[:500] | primary_topic | tags`
- **Index**: Cloudflare Vectorize `gxceed-papers` (1,024-dim, cosine)
- **Coverage**: 31,218 published public-hub paper URLs in the sitemap on 2026-10-02; new papers are embedded at ingest. This is a sitemap snapshot, not a separate Vectorize census, and not a live total.
- **Languages**: Japanese, English, Chinese (bge-m3 is multilingual)

### Use cases

- **RAG / LLM grounding**: Retrieve the most semantically relevant GX papers for a given research question or claim. Use `mode=vector` + `limit=10` as context retrieval step.
- **Literature discovery**: Find papers on abstract concepts (`"fish sentience and fisheries"`, `"GX transition justice"`) that keyword search misses.
- **Citation assistance**: Find supporting papers for a specific technical claim across JP+EN+CN corpora.
- **Hypothesis generation**: Query an unexplored topic to find the closest existing research and identify gaps.
- **SNE corpus analysis**: Filter by `sne_profile_hint` after retrieval to understand knowledge production distribution on a topic.

---

## Tech Stack (for AI reference)

- Frontend: Next.js 15 + React 19 + Cloudflare Pages (edge runtime)
- Database: Cloudflare D1 (SQLite)
- Vector search: Cloudflare Vectorize (`gxceed-papers`, 1,024-dim cosine) + Workers AI `@cf/baai/bge-m3`
- AI pipeline: DeepSeek V4 Flash (paper scoring, bilingual translation, summaries) / Claude Opus (editorial review)
- Collection agent: Mac mini always-on subagents (15 paper resolvers; article discovery via hermes web/X + Claude ingest)

---

*Last updated: 2026-10-02 / gxceed (sitemap snapshot: 31,218 public paper URLs, 1,401 article URLs; SNE share figures below remain the 2026-09-03 observation and were not recomputed) / Operator: https://gxceed.com/company*