Product Architecture

bijux-proteomics is one bounded product with several owners, not several packages that only happen to sit in one repository.

The end-to-end promise is simple:

1. accept benchmark-backed proteomics evidence
2. preserve shared meaning and compatibility
3. execute a reviewable runtime lane
4. ground the resulting sentence scientifically
5. narrow or refuse the recommendation when pressure appears
6. make downstream lab burden explicit before follow-up is treated as justified

The package split exists because those six moves do not own the same truth.

End-To-End Chain

stage	owner	what that owner contributes	what it must not quietly take over
shared meaning	bijux-proteomics-foundation	identifiers, compatibility, deterministic serialization, stable cross-package contracts	scientific workflow claims, runtime control, recommendation posture
scientific evidence root	bijux-proteomics-core	benchmark assets, sequence and chemistry surfaces, spectra and mzML handling, identification, quantification, PTM and workflow contracts	runtime orchestration, final recommendation strength, assay-worth judgment
runtime proof	bijux-proteomics-runtime	rerun lanes, replay, artifact stability, refusal surfaces, operator entrypoints	benchmark ownership, scientific truth, recommendation authority
grounded review	bijux-proteomics-knowledge	claim grounding, contradiction review, literature pressure, evidence memory	runtime proof, ranking, assay planning
recommendation posture	bijux-proteomics-intelligence	downgrade logic, challenge surfaces, confidence pressure, regret review	evidence storage, benchmark ownership, lab execution
downstream consequence	bijux-proteomics-lab	planning, readiness, refusal, requested-versus-observed outcome loops, follow-up burden	scientific meaning, runtime control, evidence grounding

Two repository-adjacent surfaces sit beside that chain:

• agentic-proteins preserves historical runtime compatibility
• bijux-proteomics-dev owns repository health, docs integrity, release checks, and maintainer tooling

What Is Deeper Since v0.3.7

The product is materially stronger than the older docs implied:

• core now owns visibly broader biology and chemistry surfaces instead of thin workflow wrappers
• runtime now owns replay, rerun kits, artifact verification, and refusal routes rather than one convenient command path
• knowledge and intelligence now expose grounded contradiction, challenge, downgrade, and recommendation pressure explicitly
• lab now owns real consequence, readiness, and outcome-learning loops instead of generic “next step” prose

That deeper substance is exactly why the architecture page matters. More real product depth means the boundary between owners has to become clearer, not blurrier.

What The Architecture Protects

• stronger scientific breadth in core does not silently turn into runtime or recommendation ownership
• stronger runtime proof does not silently widen scientific truth
• grounded claim support does not silently widen recommendation posture
• promising analytical posture does not silently widen downstream lab value

The architecture exists to keep those owner changes legible.

Strongest Reader Route

Read the chain in this order when you want the shortest serious explanation:

1. Benchmark Assets
2. Execution
3. Workflow Claim Grounding
4. Workflow Recommendation Confidence
5. Lab Consequence

If one hop weakens, the public sentence narrows there.

First Boundary Checks

• Cross-Package Ownership
• Workflow Families
• Current Capability Limits

Boundary

This page owns the product chain and the owner split. It should not dissolve into a package inventory once the end-to-end logic is already clear.