← PhenoPaper

Workflow

From paper discovery to AI-verified notebooks.

PhenoPaper separates relevance, publication kind, semantic indexing, paper-linked evidence, and notebook execution. Automated investigation results remain Unverified until reviewed. These jobs can pause independently, so a saved result count does not by itself mean its worker is running now.

Explore API search terms, queries, and providers ↓

1. Discovery and paper classification

These steps identify relevant records and describe the publication type. Relevance and paper kind answer different questions.

01Scholarly APIs

Paper discovery and normalization

No AI model
Input
OpenAlex · Crossref · Europe PMC · bioRxiv · arXiv
Process
Collect metadata, normalize identifiers, and merge duplicate records while preserving source evidence.
Output
One normalized Paper record per work/version group
02OpenAI API

Luna relevance screen

GPT-5.6 Luna · low reasoning
Input
Paper title and available abstract
Process
Estimate whether the work is relevant to plant-phenotyping methods. This is not a code or reproducibility assessment.
Output
Relevant · Insufficient evidence · Out of scope
03TypeSafeAI

JEV publication-kind classifier

jev-latest
Input
Metadata for Luna-relevant papers
Process
Assign a general publication kind; it does not decide plant-phenotyping relevance or approve a Catalog record.
Output
Research · Review · Dataset · Software · Other · Uncertain

2. Independent enrichment and study-material investigation

These jobs enrich relevant records independently. Their results have separate queues, evidence, and completion states.

04Taxonomy-backed classification

Controlled semantic tagging

Canonical vocabulary
Input
Relevant paper title and abstract
Process
Assign evidence-backed crop, trait, modality, task, organ, and environment terms.
Output
Controlled semantic tags; new terms remain review suggestions
05Local paper index

Minisforum Qwen paper embedding

qwen3-embed-1024 · 1,024 dimensions
Input
Relevant paper title and abstract
Process
Create paper vectors for related-paper analysis and the local UMAP/HDBSCAN map.
Output
phenocode-paper-embeddings
06Public search index

Cloudflare Qwen search embedding

@cf/qwen/qwen3-embedding-0.6b · 1,024 dimensions
Input
Public paper text and search query
Process
Represent papers and queries in the public semantic-search index.
Output
phenopaper-search-embeddings; separate from the Minisforum paper index
07OpenCode /analyze

OpenCode study-material investigation

GLM-5.3-Flash · OpenCode Zen
Input
Paper text when obtainable, paper-linked sources, and API metadata
Process
Check full-text routes, resource links and their contents; assess minimal-example readiness and map the study workflow.
Output
Evidence-linked candidates and a full or partial study flow, marked Unverified

3. Colab reproduction and public access

Notebook creation is a separate workflow from /analyze. An investigation result alone is not an AI-verified notebook.

08OpenCode /reproduce

Colab reproduction and publication

Separate notebook project and queue
Input
Eligible paper, confirmed implementation, and available reproduction materials
Process
Author a notebook and run its setup, downloads, conversions, and scientific analysis inside a fresh Google Colab runtime.
Output
Executed notebook published to the dedicated GitHub repository; failed or unavailable runtime paths stay unverified
09PhenoPaper public site

Paper records, statistics, and maps

Database-backed projections
Input
Paper metadata, reviewed records, semantic indices, and published notebook records
Process
Show paper details and associated resources; aggregate the separate classifications and vector projections.
Output
Paper pages · Statistics · Map

Paper, Tool, Repository, and Evidence are separate records. A link check or readiness assessment is not an execution result, and no automated result is promoted directly to the reviewed Catalog.

Sign in / My workspace