Goal
Sponsors automatically score whether their dependents' work aligns with the sponsor's charter, using the same stance-aware scoring as the alignment pricing function.
Current state
nodes/common/sponsor_agent_manager.py:check_alignment (lines 239–275) computes alignment via keyword overlap with a TODO marker for actual embeddings (line 262). The score is recommendation-only; it doesn't influence reward distribution or sponsor-side budget gating.
Scope
- Replace the keyword-overlap stub with calls into the stance-aware scorer (see related alignment-pricing issue).
- Define the trigger: per-heartbeat? per-dependent-action? per-budget-allocation? Pick one, document it.
- Define the consequence: pure advisory (sponsor sees a score), soft (warning surfaced to user), or hard (budget pull-back). Default: advisory.
- On-chain hook: optional. If hard enforcement is chosen, wire into RPB's training reward distribution.
Acceptance
- A test where a dependent agent operates outside its sponsor's charter shows a meaningfully lower compliance score than an aligned dependent.
- Score appears in the snapshot under each sponsor's view of its dependents.
Depends on
- Stance-aware alignment scorer (separate issue) — this issue should not duplicate that infrastructure.
Goal
Sponsors automatically score whether their dependents' work aligns with the sponsor's charter, using the same stance-aware scoring as the alignment pricing function.
Current state
nodes/common/sponsor_agent_manager.py:check_alignment(lines 239–275) computes alignment via keyword overlap with a TODO marker for actual embeddings (line 262). The score is recommendation-only; it doesn't influence reward distribution or sponsor-side budget gating.Scope
Acceptance
Depends on