Architecture Risks

The primary risk in agentic-proteins is compatibility drift: a historical path appears to work but no longer resolves to the same contract as bijux-proteomics-runtime.

Risk	Failure mode	Control
Duplicate implementation	A forwarding module acquires its own scheduler, provider, state, or API behavior	Keep behavior in runtime and assert canonical ownership
Object substitution	Historical and canonical imports expose look-alike but distinct classes or callables	Verify object identity for public forwarded symbols
Signature drift	Arguments, defaults, return types, or exceptions diverge	Compare public signatures and behavior in compatibility tests
Optional-extra drift	A historical extra installs a different capability set	Map extras directly to matching runtime extras
State fork	Historical paths read or write a separate store or snapshot shape	Forward state contracts and persistence operations to runtime
Error masking	The bridge catches a runtime failure and returns weaker or ambiguous information	Preserve canonical exception and structured error semantics
Documentation fork	Historical docs describe features or authority that runtime does not own	Treat runtime documentation as authoritative for behavior
New adoption	New applications depend on the historical namespace	Lead migration examples with canonical runtime imports

flowchart LR
    H[Historical contract] --> T{Identity, signature, behavior, state}
    T -->|equivalent| C[Compatibility preserved]
    T -->|different| D[Compatibility drift]
    D --> E{Can exact forwarding be restored?}
    E -->|yes| F[Repair bridge]
    E -->|no| M[Declare migration requirement]

Compatibility must not be “mostly equivalent.” Differences in exception type, provider availability, output layout, state identity, or side effects can invalidate automation even when a simple example still runs. If exact forwarding cannot be preserved, the consumer-visible break and canonical migration path must be explicit.

Close a detected risk

Finding	Required disposition	Closure evidence
canonical identity or signature drift	block the affected compatibility release	restored forwarding or a versioned breaking-contract decision with migration guidance
behavior, error, state, or artifact drift	classify as bridge defect unless the consumer contract explicitly permits adaptation	paired parity record over success, failure, side effects, and persisted output
optional-extra mismatch	block the affected installation surface	resolved dependency mapping and clean-environment capability test
undocumented dead namespace	stop promising a replacement	inventory classification, caller search, and removal guidance
new compatibility adoption	route the caller to Runtime or Core	dependency review showing the canonical import or command
incomplete external-consumer inventory	keep retirement blocked	named support boundary and caller-owned migration dispositions

Repairing a bridge defect does not prove caller migration. Migrating every known caller does not prove the inventory is complete. Release and retirement records must preserve those conclusions separately.