T04: deterministic semantic kernel

Alice/Bob/Carol runs end to end, deterministically, replayable from seed.
16 tests pass, no third-party dependencies.

- src/testdriver: intent, provenance, world, actions, drivers, observers,
  oracles, evidence, energy, scenario, runner
- lab/minimal.py: the SUT, exposing the independent observation channel
  required by D-07
- evidence is stratified S1/S2/S3; Runner refuses to attribute S2/S3 to an
  actor; claims are frozen and provenance-checked at construction
- missing evidence yields INCONCLUSIVE, which outranks PASS in the run verdict
- EnergyEvents captured, no scoring (H-005 dormant)

The observation channel records both stored state and an out-of-band
enforcement probe; their disagreement is an invariant and is what detects an
authorization defect that leaves the audit trail intact. A seeded
RevokeIsCosmetic lab fails the run via both the claim and that invariant.

Also closes TD-WP-0001-T02 (stack and commands now exist).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 1629012@bnt-lap001
Assistant-Session: 78d4fb13-8a1e-474b-87a3-9b9261c49a39

tests/test_reference_scenario.py

@@ -0,0 +1,70 @@
+"""The reference scenario must run deterministically and be replayable."""
+
+from __future__ import annotations
+
+import json
+
+import pytest
+
+from testdriver import Runner, Stratum, Verdict
+from scenarios.alice_bob_carol import build
+
+
+def run_once(variant: str = "baseline"):
+    world, driver, observer, asset, oracle = build(variant)
+    return Runner(world, driver, observer, oracle).run(asset), world
+
+
+def test_reference_scenario_passes():
+    result, _ = run_once()
+    assert result.verdict is Verdict.PASS, [
+        j.as_dict() for j in result.judgments if j.verdict is not Verdict.PASS
+    ]
+
+
+def test_every_claim_is_judged():
+    result, _ = run_once()
+    judged = {j.assertion_id for j in result.judgments}
+    assert {"c-bob-reads", "c-carol-denied", "c-bob-revoked"} <= judged
+
+
+def test_invariants_are_evaluated_after_every_step():
+    result, _ = run_once()
+    per_step = [j for j in result.judgments if j.assertion_id == "i-audit-append-only"]
+    assert len(per_step) == 3
+
+
+def test_run_is_replayable_from_known_initial_state():
+    """Two runs from the same seed produce identical judgments."""
+    first, _ = run_once()
+    second, _ = run_once()
+    assert [(j.assertion_id, j.verdict) for j in first.judgments] == [
+        (j.assertion_id, j.verdict) for j in second.judgments
+    ]
+    assert first.run_id != second.run_id
+
+
+def test_evidence_is_stratified_and_serializable():
+    result, _ = run_once()
+    pack = result.evidence
+    assert pack.of_stratum(Stratum.SURFACE)
+    assert pack.of_stratum(Stratum.REALIZATION)
+    assert pack.of_stratum(Stratum.JUDGMENT)
+    parsed = json.loads(pack.to_json())
+    assert parsed["run_id"] == result.run_id
+    assert parsed["sut_version"] == "lab-0.1.0-baseline"
+
+
+def test_evidence_records_claim_provenance():
+    """A verdict must be auditable for the independence of the claim behind it."""
+    result, _ = run_once()
+    assert result.evidence.provenance_index["c-bob-revoked"] == "human"
+
+
+def test_actors_hold_isolated_credentials_and_memory():
+    _, world = run_once()
+    alice, bob = world.cast["alice"], world.cast["bob"]
+    assert alice.credentials["token"] != bob.credentials["token"]
+    alice.remember("secret", "only alice knows this")
+    assert bob.recall("secret") is None
+    assert bob.known_keys() == ()