- FactorForge package, web UI, CLI, and documentation
- Eijex MCP public access surface
- AgentOS workflow-integrity code and synthetic fixtures
- ValidationHub schemas and registry code
Open evidence infrastructure for plant expression
Design sequences.
Govern workflows.
Preserve evidence.
Eijex develops open-source, human-supervised research software connecting reproducible plant CDS design with reviewable evidence infrastructure.
FactorForge is available now. Shared database integration remains a research checkpoint.
View Eijex on GitHubOpen-source software for reproducible plant bio-design, human-supervised workflow governance, and evidence traceability.
From design to evidence
One workflow. Clear responsibility at every step.
FactorForge creates reviewable design candidates, AgentOS guides human-supervised decisions, and ValidationHub preserves the evidence behind each handoff. The Evidence Toolkit checks a sequence-free package of references across those responsibilities.
- 01
Design
FactorForge
Create reproducible CDS candidates and run deterministic computational checks.
Learn about FactorForge - 02
Review
AgentOS
Apply policies, record approvals, and keep human judgment in control.
Learn about AgentOS - 03
Preserve
ValidationHub
Store reviewed evidence as structured, traceable research records.
Learn about ValidationHub
One identity and provenance trail connects the entire workflow.
The current connection is a development-stage contract over identifiers, provenance, and checksums—not a completed shared runtime or database.
See the Evidence Toolkit contractEijex MCP provides public access to supported tools while scientific source-of-truth records remain with their owning systems.
FactorForge report
A result should tell a researcher what to do next.
The Researcher Decision Report leads with the run's recorded decision, turns unresolved computational checks into bounded review actions, and keeps full evidence available without making the summary read like a regulatory submission.
Available now
Decision report and sequence-free evidence JSON
Planned next
Append-oriented Research Activity Log
Researcher Decision Report
Conditional pass
Review warnings before candidate selection.
0 required fails · 2 warnings
- 01
Decide
See the automated disposition, unresolved checks, and the next review action before reading the full metrics.
- 02
Verify
Compare requested settings with the settings recorded by the run, then inspect checks and provenance.
- 03
Hand off
Download the sequence-bearing report separately from the sequence-free evidence record for downstream review.
The report supports research review. It does not establish expression, biological performance, regulatory acceptance, or experimental validity.
Implementation status
What exists today—and what still has to be proven.
Public software availability and database integration are different milestones. This view keeps them separate.
- Evidence Toolkit v0.1 contract and synthetic package validator
- Captured GCP PostgreSQL 16 research instance
- Separate tested FactorForge SQLite metadata path
- Byte-pinned candidate-asset manifest registration
- Ordered migrations and schema-version policy
- PostgreSQL, SQLite, and specification reconciliation
- Access, restore, state-isolation, and leakage tests
- Public or synthetic end-to-end lineage fixture
These checkpoints show that implementation has begun.
They do not establish a synchronized production database, a completed evidence ledger, or biological performance.
They do not establish a synchronized production database, a completed evidence ledger, or biological performance.
Open-source stack
Public tools with explicit responsibility boundaries.
Public software releases are staged independently from permissioned research data.
FactorForge
AvailableCDS Design Review
Pre-synthesis sequence review for plant CDS workflows, with reproducible candidates, deterministic checks, and citable review artifacts.
pip install factorforge-cdsEijex MCP
AvailablePublic Tool Access
MCP-compatible access to public FactorForge capabilities with explicit tool and data boundaries.
mcp.eijex.comAgentOS
Public codeWorkflow Control Plane
Human-supervised orchestration for policy checks, approval gates, audit trails, and evidence-safe public projection.
Human approval at public boundariesValidationHub
Public codeEvidence Registry
Schema-driven infrastructure for reviewing and curating validation evidence while permissioned records remain outside public repositories.
Open schemas · Permissioned evidenceEvidence Toolkit
DevelopmentReview Package Contract
Sequence-free schemas and deterministic checks for linking design, selection, and validation-evidence references into a reviewable package.
Contract v0.1 · Synthetic fixturesFactorForge available via
pip install factorforge-cdsconda install factorforge-cdssoondocker pull ghcr.io/eijex/factorforge-cds:latestResearch program
Software claims and research evidence advance separately.
Each research track has its own scope, evidence gate, and publication path. Neither substitutes computational results for physical observations.
1Computational case study
Paper 1 — Reproducible CDS design
A frozen N. benthamiana computational study and reproducibility package. It makes no expression, yield, or wet-lab claim.
Current status: Computational scope frozen
2Planning scaffold
Paper 2 — Evidence infrastructure
A research plan for traceable design-to-evidence infrastructure, now recording a partial database implementation checkpoint.
Current status: Evidence gate not yet complete
Research Notes
The decisions, dead ends, and lessons behind the work.
Engineering investigations and computational observations from building reproducible, evidence-aware bio-design software.
Evidence ArchitectureEngineering note
From a result page to a traceable research record
Why FactorForge now separates researcher decisions, sequence-bearing reports, and sequence-free evidence—and what must exist before that history becomes a controlled audit trail.
· 3 minRead note →
Evidence ArchitectureEngineering note
When an integrity framework stops its own best-looking result
A six-arm benchmark prototype produced attractive demonstration values, but provenance and evidence checks showed why they could not yet be treated as scientific results.
· 2 minRead note →
Rule/LM ArchitectureCurrent
Why rule authority still matters in a language-model system
A code-level audit changed the question from whether a language model could replace deterministic design rules to how learned proposals could remain bounded by them.
· 2 minRead note →
Design & ValidationEngineering note
Why every CAI value needs an authority record
Two CAI values for the same sequence were both mathematically correct—and still misleading when shown without their references.
· 2 minRead note →
Research Notes are editorial records, not peer-reviewed publications or biological validation evidence. Incomplete and superseded work remains visible when it helps explain how a decision changed.
Open code · permissioned data
Public software does not mean public research records.
Code, public schemas, documentation, and synthetic fixtures are open. Raw sequences, partner documents, personal information, and permissioned experimental evidence are governed separately and are not published in the software repositories.