20 — MVP Build Checklist
Dataset
- Small evidence corpus prepared.
- Synthetic latent-context dataset generated.
- Train/dev/test split hashes recorded.
- Gold labels isolated from runtime.
Evidence
- Evidence atoms created.
- Stable IDs and hashes.
- Access states attached.
- Citation lookup tested.
SNO extraction
- Candidate SNO schema implemented.
- Claim extraction prompt or model.
- Relation extraction prompt or model.
- Parser tests for malformed output.
Critics
- Citation validator.
- Entailment scorer.
- Logic critic.
- Topology critic.
- Access critic.
- Chirality critic.
Pair selection
- Evidential Entanglement score.
- Graph chirality.
- Evidence-polarity chirality.
- Productive Conflict Score.
- Pair-selection report.
Proof closure
- Rule registry.
- Zero-temperature closure.
- Proof trace generation.
- ZTHR metric.
Predicate invention
- Residual tensor builder.
- Factorization routine.
- Candidate predicate labels.
- Predicate grounding tests.
- PIU metric.
Synthesis
- Synthesizer prompt.
- Synthesized SNO schema.
- Proof/reference preservation.
- Residual contradiction preservation.
Orthesis
- Render/re-ground loop.
- Round-trip residual.
- Stability threshold.
- Orthesis report.
Audit
- strict/likely/hypothesis/unresolved/rejected sections.
- top worlds.
- access gaps.
- proof trace links.
- latent predicate status.
- calibration report.
Stop condition
Do not expand to large multi-agent runtime until the SNO → Antagonist → proof → predicate invention → Synthesizer → orthesis loop works on toy data.