pending — blocked on Phase 4 close
Phase 5 — Cross-era Smoke Test
End-to-end integration test of the full Prime Radiant stack. Five canonical cross-era queries (one per coupling kind from the 10-kind enum) are run through the complete pipeline: HippoRAG-lite retrieval → Gemma 3 4B reasoning → primitive grammar parse → validated card refs. All five must pass before MVP is declared.
← Back to site mapThe five canonical smoke queries
| # | Query | Expected coupling kind | Expected era span |
|---|---|---|---|
| 1 | "How does the Joint Stock Company substrate support AWS cloud infrastructure?" | substrate_provision |
MM → DM |
| 2 | "What zombie dependencies does the Westphalian state system maintain on the Internet?" | zombie_dependency |
MM → DM |
| 3 | "How does the LM mutual aid network resist parasitic extraction by DM platform capitalism?" | parasitic_extraction |
DM → LM |
| 4 | "Trace how the Hanseatic League's adapted inheritance shows up in the Euro clearing system." | adapted_inheritance |
MM → DM |
| 5 | "How does open-source infrastructure sublimates the proprietary software OPP?" | sublimation_coupling |
DM → LM |
Pass criteria
- All five queries retrieve ≥1 correct seed card within top-5 (recall@5 = 1.0)
- Gemma 3 4B emits the expected coupling kind token in its response
prime_radiant/parse.pyvalidates all emitted<|machine|>refs against the card index (no dangling refs)- Any
<|new_machine|>proposals are valid JSON stubs that pass Pydantic schema validation - Total pipeline latency ≤ 60 seconds per query on target hardware
MVP declaration: When Phase 5 passes, the directory is renamed to phase-5-DONE-smoke/ and the project enters the post-MVP iteration cycle (larger corpus, finer grammar, Calvino story mode, Brier calibration loop).