Company landing page for eijex — open-source research software for reproducible plant CDS design, human-supervised workflow governance, and evidence traceability.
Live: www.eijex.com
- Next.js 16 (App Router)
- Tailwind CSS v4
- next-themes — dark/light mode
- Markdown Research Notes with reviewed public-copy boundaries
- Deployed on Vercel
npm install
npm run devOpen http://localhost:3000.
npm run buildThe public archive is served from content/research-notes/ at
/research-notes. These Markdown files are
curated public copies; the site does not read private or internal workspace notes.
Each note records editorial status, evidence scope, and book-oriented section/order
metadata. The archive also exposes an RSS feed at /research-notes/feed.xml.
Before publishing changes, run:
npm run check:claims
npm run lint
npm run buildPushed to main → auto-deployed to Vercel → served at www.eijex.com.
Naked domain (eijex.com) redirects to www.eijex.com via Cloudflare Redirect Rule.
- FactorForge — CDS design engine
- Eijex MCP — MCP-compatible public tool access
- AgentOS — public human-supervised workflow control-plane code
- ValidationHub — public evidence-registry code and schemas
- Evidence Toolkit — development-stage sequence-free evidence-package contract and validator
- eijex GitHub
FactorForge copy distinguishes sequence-only baseline from experimental local 3DP comparison, labels watermark disabled, and describes ResearchOS/AgentOS/ValidationHub responsibilities. Application version copy follows the current FactorForge package v3.5.5; historical benchmark figures are unchanged. Local synthetic integration is software evidence, not partner approval or wet-lab performance. These edits do not change the deployed website until an authorized deployment.
Job 389 supersedes the Job 388 local-only browser routing: the FactorForge source now implements direct stateless 3DP browser/API testing with uploaded exact-match structure and reviewed positions. No ResearchOS approval or ValidationHub record is collected in this test path. It is unreleased; public MCP tools do not yet expose this execution mode.
/slm separates the research objective, locally implemented foundation, reviewed partial experiments and planned DP-teacher pilot. /validationhub shares the reviewed result panel and labels the initial pilot as historical; counts from different runs are not cumulative totals.
The public aggregate at /evidence/slm-development-2026-10-29.json is manually reviewed static data, not a live registry connection. It excludes raw sequences, individual checkpoint hashes and local storage references. Its rendered copy is src/app/lib/reviewed-research-evidence.json; keep both copies consistent. Reused development-validation NLL does not establish final-test or biological performance.
Run node --test scripts/public-learning-summary.test.mjs scripts/research-evidence.test.mjs as well as the normal build checks after editing these surfaces.