capstone/reference.py

"""Reference capstone evidence. All work remains synthetic and local."""
from pathlib import Path
import sys
sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "project"))
from engine import BaselinePlanner, Checkpoints, Engine, SimulatedCrash
from evaluate import evaluate
from workflow import Request


def build_evidence():
    store = Checkpoints()
    try:
        engine = Engine(store)
        engine.start(Request("CAP-RECOVERY", "EX-001", 1927))
        engine.plan("CAP-RECOVERY", BaselinePlanner())
        try:
            engine.invoke("CAP-RECOVERY", crash_after_read=True)
        except SimulatedCrash:
            pass
        before = store.get("CAP-RECOVERY")
        exported = store.export()
    finally:
        store.close()
    restored = Checkpoints.restore(exported)
    try:
        engine = Engine(restored)
        engine.recover("CAP-RECOVERY", abandon_read=True)
        reviewed = engine.run_until_pause("CAP-RECOVERY", now_tick=5)
        approved = engine.review("CAP-RECOVERY", reviewed["revision"], reviewed["draft_digest"],
                                  "synthetic-reviewer-a", True, now_tick=6)
        final = engine.finalize_simulation("CAP-RECOVERY", now_tick=6)
        replay = engine.finalize_simulation("CAP-RECOVERY", now_tick=99)
        # Independent added case: upper accepted year boundary on EX-002, then reject.
        engine.start(Request("CAP-BOUNDARY", "EX-002", 2100))
        boundary = engine.run_until_pause("CAP-BOUNDARY", now_tick=5)
        rejected = engine.review("CAP-BOUNDARY", boundary["revision"], boundary["draft_digest"],
                                  "synthetic-reviewer-a", False, now_tick=6)
        return {"evidence_kind": "synthetic_local_capstone_no_llm", "baseline": evaluate("baseline"),
                "candidate": evaluate("repeat"),
                "recovery": {"before_phase": before["phase"], "before_reservations": before["tool_calls"],
                             "after_phase": reviewed["phase"], "after_reservations": reviewed["tool_calls"]},
                "review": {"approved_phase": approved["phase"], "final_phase": final["phase"],
                           "receipt": final["receipt"], "saved_receipt_replayed": replay == final,
                           "rejected_phase": rejected["phase"], "rejected_receipt": rejected["receipt"]},
                "added_case": {"observed_year": 2100, "catalogue_year": boundary["draft"]["catalogue_year"],
                               "finding": boundary["draft"]["finding"], "phase": rejected["phase"]},
                "release_recommendation": "local_synthetic_demo_only",
                "limitations": ["no_live_model", "no_real_auth_or_clock", "no_disk_durability",
                                "no_production_or_learner_validation"]}
    finally:
        restored.close()


if __name__ == "__main__":
    import json
    print(json.dumps(build_evidence(), indent=2))