[PAC-INT] Integrate Experiment Tracking Spec v1.8.0 registry features

[noahgift]

Context

The entrenar Experiment Tracking Spec v1.8.0 has registry-critical features that pacha MUST integrate.

Epic Reference

• Epic: ENT-EPIC-001
• Spec: entrenar/docs/specifications/experiment-tracking-spec.md
• Tracking: entrenar/docs/work-items/experiment-tracking-epic.yaml

Registry Integration Points

ModelParadigm (§23) - CRITICAL

Pacha must recognize all model paradigms as first-class:

• TraditionalMl (sklearn-style)
• DeepLearning (neural nets)
• FineTuning (LoRA, QLoRA, Adapter, Prefix, IA3)
• Distillation (teacher→student)
• MoE (Mixture of Experts)
• Ensemble

Reference: entrenar#40

OfflineModelRegistry (§30.3) - CRITICAL

Sovereign deployment requires pacha to support:

• mirror_from_hub() for air-gapped seeding
• SHA-256 verification on offline load
• Local registry without network

Reference: entrenar#47

Cost-Performance Metadata (§24)

Registry entries need:

• CostMetrics ($/sample, $/epoch)
• EnergyMetrics (watts, joules, carbon_kg)
• Pareto frontier position
• ComputeDevice compatibility

References: entrenar#38, #39, #41

FAIR Research Artifacts (§31)

Model/dataset entries need:

• CitationMetadata (BibTeX, CFF)
• DOI linkage
• ORCID author attribution
• Upstream citation graph

References: entrenar#49, #50, #55

Recipe Integration

• Recipes should declare ModelParadigm
• Cost estimates in recipe metadata
• Sovereign-compatible recipes (no network)

Acceptance Criteria

• ModelParadigm enum in pacha schema
• OfflineModelRegistry trait/impl
• CostMetrics in model metadata
• CitationMetadata in model/dataset entries
• Recipe paradigm declaration

---

Cross-project tracking for sovereign AI stack

[noahgift]

Triage: P3. Feature request — experiment tracking integration from entrenar spec v1.8.0. Depends on ModelParadigm, OfflineModelRegistry, CostMetrics, and FAIR metadata. Large scope, cross-project coordination required.

[noahgift]

Triage 2026-03-31: Feature request — leave open.