File size: 22,040 Bytes
e9ce6e9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
"""
tests/test_step.py
==================
Phase 4 rewrite β€” 24 tests covering every CLAUDE.md step() contract point.
"""
import pytest
from pydantic import ValidationError

from unified_gateway import AEPOAction, AEPOObservation, UFRGObservation, UFRGReward, UnifiedFintechEnv


# ---------------------------------------------------------------------------
# Fixtures / helpers
# ---------------------------------------------------------------------------

@pytest.fixture
def env() -> UnifiedFintechEnv:
    e = UnifiedFintechEnv()
    e.reset(options={"task": "easy"})
    return e


def make_action(**kwargs) -> AEPOAction:
    """Build an AEPOAction with safe defaults for unspecified fields."""
    defaults = dict(risk_decision=0, crypto_verify=0, infra_routing=0,
                    db_retry_policy=0, settlement_policy=0, app_priority=2)
    defaults.update(kwargs)
    return AEPOAction(**defaults)


def _force_obs(env: UnifiedFintechEnv, **fields) -> None:
    """Overwrite env._current_obs with specific field values."""
    current = env._current_obs.model_dump()
    current.update(fields)
    env._current_obs = AEPOObservation.model_construct(**current)


# ---------------------------------------------------------------------------
# Test 1 β€” step() returns a 4-tuple
# ---------------------------------------------------------------------------

def test_step_returns_4_tuple(env: UnifiedFintechEnv) -> None:
    """step() must return exactly (obs, reward, done, info)."""
    result = env.step(make_action())
    assert len(result) == 4


# ---------------------------------------------------------------------------
# Test 2 β€” reward always in [0.0, 1.0]
# ---------------------------------------------------------------------------

@pytest.mark.parametrize("task", ["easy", "medium", "hard"])
def test_reward_always_in_range(task: str) -> None:
    """reward.value must never leave [0.0, 1.0] across all tasks."""
    e = UnifiedFintechEnv()
    e.reset(options={"task": task})
    for _ in range(10):
        _, tr, _, _ = e.step(make_action())
        assert 0.0 <= tr.value <= 1.0, f"reward={tr.value} for task={task}"


# ---------------------------------------------------------------------------
# Test 3 β€” done=True when kafka_lag > 4000
# ---------------------------------------------------------------------------

def test_done_true_on_kafka_lag_crash(env: UnifiedFintechEnv) -> None:
    """kafka_lag > CRASH_THRESHOLD (4000) for 2 consecutive steps must set done=True.
    Fix 11.1: single-step grace period β€” one spike is not a crash, sustained overload is.
    """
    _force_obs(env, kafka_lag=4500.0)
    env._rolling_lag = 0.0
    # Step 1: lag > 4000 β†’ grace (streak=1), episode continues
    _, _, done_step1, _ = env.step(make_action())
    assert done_step1 is False, "Grace step should not terminate episode"
    # Step 2: lag still > 4000 β†’ streak=2, crash fires
    _force_obs(env, kafka_lag=4500.0)
    _, tr, done, info = env.step(make_action())
    assert done is True
    assert tr.value == 0.0
    assert info["termination_reason"] == "crash"


# ---------------------------------------------------------------------------
# Test 4 β€” done=True on catastrophic fraud
# ---------------------------------------------------------------------------

def test_done_true_on_fraud_catastrophe(env: UnifiedFintechEnv) -> None:
    """Approve + SkipVerify + risk_score > 80 β†’ reward=0.0, done=True."""
    _force_obs(env, risk_score=90.0)
    _, tr, done, info = env.step(make_action(risk_decision=0, crypto_verify=1))
    assert done is True
    assert tr.value == 0.0
    assert info["termination_reason"] == "fraud"


# ---------------------------------------------------------------------------
# Test 5 β€” done=False before step 100 on easy with valid actions
# ---------------------------------------------------------------------------

def test_done_false_before_step_100_easy() -> None:
    """99 valid steps on easy must not set done=True."""
    e = UnifiedFintechEnv()
    e.reset(options={"task": "easy"})
    for i in range(99):
        _, _, done, _ = e.step(make_action(risk_decision=1, crypto_verify=1, infra_routing=1))
        assert done is False, f"done=True unexpectedly at step {i+1}"


# ---------------------------------------------------------------------------
# Test 6 β€” done=True after exactly 100 steps
# ---------------------------------------------------------------------------

def test_done_true_after_100_steps() -> None:
    """Episode must terminate at exactly 100 steps."""
    e = UnifiedFintechEnv()
    e.reset(options={"task": "easy"})
    done = False
    for _ in range(100):
        _, _, done, _ = e.step(make_action(risk_decision=1, crypto_verify=1, infra_routing=1))
    assert done is True


# ---------------------------------------------------------------------------
# Test 7 β€” catastrophic fraud sets reward=0.0
# ---------------------------------------------------------------------------

def test_fraud_catastrophe_reward_is_zero(env: UnifiedFintechEnv) -> None:
    """Fraud catastrophe must produce reward=0.0, not just low reward."""
    _force_obs(env, risk_score=95.0)
    _, tr, _, _ = env.step(make_action(risk_decision=0, crypto_verify=1))
    assert tr.value == 0.0


# ---------------------------------------------------------------------------
# Test 8 β€” system crash sets reward=0.0
# ---------------------------------------------------------------------------

def test_crash_reward_is_zero(env: UnifiedFintechEnv) -> None:
    """kafka_lag > 4000 for 2 consecutive steps must produce reward=0.0.
    Fix 11.1: grace period means the crash fires on the second sustained step.
    """
    _force_obs(env, kafka_lag=4001.0)
    env._rolling_lag = 0.0
    # Step 1: grace β€” reward is NOT 0.0 yet
    env.step(make_action())
    # Step 2: crash fires β€” reward must be 0.0
    _force_obs(env, kafka_lag=4001.0)
    _, tr, done, _ = env.step(make_action())
    assert done is True
    assert tr.value == 0.0


# ---------------------------------------------------------------------------
# Test 9 β€” SLA breach applies -0.30 penalty
# ---------------------------------------------------------------------------

def test_sla_breach_applies_penalty(env: UnifiedFintechEnv) -> None:
    """rolling_p99 > 800 on a clean step should yield reward β‰ˆ 0.5 (0.8 - 0.30)."""
    _force_obs(env, kafka_lag=0.0, rolling_p99=1000.0, risk_score=10.0)
    env._rolling_lag = 0.0
    env._rolling_p99 = 1000.0
    _, _, _, info = env.step(make_action())
    assert info["reward_breakdown"]["sla_penalty"] == -0.30
    assert abs(info["reward_breakdown"]["final"] - 0.5) < 0.15


# ---------------------------------------------------------------------------
# Test 10 β€” Challenge on high-risk applies +0.05 bonus
# ---------------------------------------------------------------------------

def test_challenge_bonus_on_high_risk(env: UnifiedFintechEnv) -> None:
    """Challenge (risk_decision=2) on risk_score > 80 β†’ bonus += 0.05."""
    _force_obs(env, risk_score=85.0, kafka_lag=0.0, rolling_p99=0.0)
    env._rolling_lag = 0.0
    _, _, _, info = env.step(make_action(risk_decision=2, crypto_verify=0))
    assert info["reward_breakdown"]["bonus"] >= 0.05


# ---------------------------------------------------------------------------
# Test 11 β€” Reject+SkipVerify on high-risk applies +0.04 bonus (blind spot)
# ---------------------------------------------------------------------------

def test_reject_skip_verify_blind_spot_bonus(env: UnifiedFintechEnv) -> None:
    """Reject + SkipVerify on risk > 80 must trigger blind_spot_triggered=True and +0.04."""
    _force_obs(env, risk_score=90.0, kafka_lag=0.0, rolling_p99=0.0)
    env._rolling_lag = 0.0
    _, _, _, info = env.step(make_action(risk_decision=1, crypto_verify=1))
    assert info["blind_spot_triggered"] is True
    assert info["reward_breakdown"]["bonus"] >= 0.04


# ---------------------------------------------------------------------------
# Test 12 β€” CircuitBreaker applies -0.50 penalty
# ---------------------------------------------------------------------------

def test_circuit_breaker_penalty(env: UnifiedFintechEnv) -> None:
    """CircuitBreaker (infra_routing=2) must apply -0.50 infra_penalty."""
    _force_obs(env, kafka_lag=0.0, rolling_p99=0.0, risk_score=10.0)
    env._rolling_lag = 0.0
    _, _, _, info = env.step(make_action(infra_routing=2))
    assert info["reward_breakdown"]["infra_penalty"] <= -0.50
    assert info["circuit_breaker_tripped"] is True


# ---------------------------------------------------------------------------
# Test 13 β€” DeferredAsync during Normal phase applies -0.15 penalty
# ---------------------------------------------------------------------------

def test_deferred_async_normal_phase_penalty(env: UnifiedFintechEnv) -> None:
    """DeferredAsyncFallback (settlement_policy=1) in Normal phase β†’ -0.15."""
    # Force normal conditions: easy task, healthy bank
    _force_obs(env, kafka_lag=0.0, rolling_p99=0.0, risk_score=10.0, bank_api_status=0.0)
    env._rolling_lag = 0.0
    env._last_event_type = "normal"
    _, _, _, info = env.step(make_action(settlement_policy=1))
    assert info["reward_breakdown"]["settlement_penalty"] <= -0.15


# ---------------------------------------------------------------------------
# Test 14 β€” DeferredAsync cumulative > 10 applies extra -0.20
# ---------------------------------------------------------------------------

def test_deferred_async_cumulative_penalty(env: UnifiedFintechEnv) -> None:
    """Cumulative settlement backlog > 10 triggers -0.20 additional penalty."""
    env._cumulative_settlement_backlog = 10   # already at 10; next step makes it 11
    _force_obs(env, kafka_lag=0.0, rolling_p99=0.0, risk_score=10.0, bank_api_status=0.0)
    env._rolling_lag = 0.0
    env._last_event_type = "normal"
    _, _, _, info = env.step(make_action(settlement_policy=1))
    assert info["consecutive_deferred_async"] == 11
    # Both -0.15 (normal phase) and -0.20 (> 10 cumulative) should be applied
    assert info["reward_breakdown"]["settlement_penalty"] <= -0.35


# ---------------------------------------------------------------------------
# Test 15 β€” ExponentialBackoff when db_pool < 20 applies -0.10
# ---------------------------------------------------------------------------

def test_exponential_backoff_low_pool_penalty(env: UnifiedFintechEnv) -> None:
    """ExponentialBackoff (db_retry_policy=1) when db_pool < 20 β†’ -0.10."""
    _force_obs(env, kafka_lag=0.0, rolling_p99=0.0, risk_score=10.0, db_connection_pool=10.0)
    env._rolling_lag = 0.0
    _, _, _, info = env.step(make_action(db_retry_policy=1))
    assert info["reward_breakdown"]["db_penalty"] == -0.10


# ---------------------------------------------------------------------------
# Test 16 β€” ExponentialBackoff when db_pool > 80 applies +0.03 bonus
# ---------------------------------------------------------------------------

def test_exponential_backoff_high_pool_bonus(env: UnifiedFintechEnv) -> None:
    """ExponentialBackoff (db_retry_policy=1) when db_pool > 80 β†’ +0.03 db_penalty."""
    _force_obs(env, kafka_lag=0.0, rolling_p99=0.0, risk_score=10.0, db_connection_pool=85.0)
    env._rolling_lag = 0.0
    _, _, _, info = env.step(make_action(db_retry_policy=1))
    assert info["reward_breakdown"]["db_penalty"] == 0.03


# ---------------------------------------------------------------------------
# Test 17 β€” app_priority=UPI with merchant_tier=Small β†’ +0.02
# ---------------------------------------------------------------------------

def test_app_priority_upi_small_merchant_bonus(env: UnifiedFintechEnv) -> None:
    """UPI priority (app_priority=0) + Small merchant (tier=0) β†’ +0.02 bonus."""
    _force_obs(env, kafka_lag=0.0, rolling_p99=0.0, risk_score=10.0, merchant_tier=0.0)
    env._rolling_lag = 0.0
    _, _, _, info = env.step(make_action(app_priority=0))
    assert info["reward_breakdown"]["bonus"] >= 0.02


# ---------------------------------------------------------------------------
# Test 18 β€” app_priority=Credit with merchant_tier=Enterprise β†’ +0.02
# ---------------------------------------------------------------------------

def test_app_priority_credit_enterprise_bonus() -> None:
    """Credit priority (app_priority=1) + Enterprise merchant (tier=1) β†’ +0.02 bonus."""
    e = UnifiedFintechEnv()
    e.reset(options={"task": "hard"})   # hard task sets merchant_tier=Enterprise=1.0
    obs = e._current_obs
    # Only proceed if the hard obs actually set Enterprise tier
    _force_obs(e, kafka_lag=0.0, rolling_p99=0.0, risk_score=10.0, merchant_tier=1.0)
    e._rolling_lag = 0.0
    _, _, _, info = e.step(make_action(app_priority=1))
    assert info["reward_breakdown"]["bonus"] >= 0.02


# ---------------------------------------------------------------------------
# Test 19 β€” info dict contains all required keys on every step
# ---------------------------------------------------------------------------

def test_info_dict_contains_all_required_keys(env: UnifiedFintechEnv) -> None:
    """Every step() must return an info dict with the full CLAUDE.md key set."""
    _, _, _, info = env.step(make_action())
    required_keys = {
        "phase", "curriculum_level", "step_in_episode", "raw_obs",
        "reward_breakdown", "termination_reason", "adversary_threat_level_raw",
        "blind_spot_triggered", "consecutive_deferred_async",
        # backward-compat keys
        "step", "task", "event_type", "obs_risk_score", "obs_kafka_lag",
        "obs_rolling_p99", "action_risk_decision", "action_infra_routing",
        "action_crypto_verify", "reward_raw", "reward_final",
        "circuit_breaker_tripped", "crashed", "done",
        "internal_rolling_lag", "internal_rolling_latency",
    }
    missing = required_keys - info.keys()
    assert not missing, f"Missing info keys: {missing}"


# ---------------------------------------------------------------------------
# Test 20 β€” info["reward_breakdown"]["final"] matches returned reward
# ---------------------------------------------------------------------------

def test_reward_breakdown_final_matches_return(env: UnifiedFintechEnv) -> None:
    """reward_breakdown['final'] must equal the typed_reward.value returned."""
    _, typed_reward, _, info = env.step(make_action())
    assert info["reward_breakdown"]["final"] == typed_reward.value


# ---------------------------------------------------------------------------
# Test 21 β€” blind_spot_triggered=True on Reject+SkipVerify+high_risk
# ---------------------------------------------------------------------------

def test_blind_spot_triggered_flag(env: UnifiedFintechEnv) -> None:
    """blind_spot_triggered must be True for Reject+SkipVerify on risk > 80."""
    _force_obs(env, risk_score=88.0)
    _, _, _, info = env.step(make_action(risk_decision=1, crypto_verify=1))
    assert info["blind_spot_triggered"] is True


# ---------------------------------------------------------------------------
# Test 22 β€” termination_reason="crash" on lag crash
# ---------------------------------------------------------------------------

def test_termination_reason_crash(env: UnifiedFintechEnv) -> None:
    """kafka_lag > 4000 for 2 consecutive steps β†’ info['termination_reason'] = 'crash'.
    Fix 11.1: single spike gives streak=1 (grace), second spike fires the crash.
    """
    _force_obs(env, kafka_lag=5000.0)
    env._rolling_lag = 0.0
    # Step 1: grace β€” termination_reason is None
    _, _, _, info1 = env.step(make_action())
    assert info1["termination_reason"] is None, "Grace step should not crash"
    assert info1["crash_grace_active"] is True
    # Step 2: sustained overload β€” crash fires
    _force_obs(env, kafka_lag=5000.0)
    _, _, done, info = env.step(make_action())
    assert done is True
    assert info["termination_reason"] == "crash"


# ---------------------------------------------------------------------------
# Test 23 β€” termination_reason="fraud" on fraud catastrophe
# ---------------------------------------------------------------------------

def test_termination_reason_fraud(env: UnifiedFintechEnv) -> None:
    """Fraud catastrophe must set info['termination_reason'] = 'fraud'."""
    _force_obs(env, risk_score=92.0)
    _, _, _, info = env.step(make_action(risk_decision=0, crypto_verify=1))
    assert info["termination_reason"] == "fraud"


# ---------------------------------------------------------------------------
# Test 24 β€” termination_reason=None on a normal step
# ---------------------------------------------------------------------------

def test_termination_reason_none_on_normal_step(env: UnifiedFintechEnv) -> None:
    """A normal step must leave info['termination_reason'] as None."""
    _force_obs(env, kafka_lag=100.0, rolling_p99=50.0, risk_score=10.0)
    env._rolling_lag = 0.0
    _, _, _, info = env.step(make_action(risk_decision=1, crypto_verify=0))
    assert info["termination_reason"] is None


# ---------------------------------------------------------------------------
# Test 25 β€” reject spam penalty (Issue 6 fix)
# ---------------------------------------------------------------------------

def test_reject_spam_penalty_fires_after_5_consecutive_rejects(env: UnifiedFintechEnv) -> None:
    """6+ consecutive Reject decisions must trigger -0.15 infra_penalty (reject_spam_active)."""
    _force_obs(env, kafka_lag=50.0, rolling_p99=50.0, risk_score=30.0)
    env._rolling_lag = 0.0
    env._rolling_p99 = 50.0

    # Step 6 times with Reject β€” reject_spam fires on step 6 (counter > 5)
    for _ in range(5):
        env.step(make_action(risk_decision=1))

    _, _, _, info = env.step(make_action(risk_decision=1))
    assert info["reject_spam_active"] is True
    assert info["reward_breakdown"]["infra_penalty"] <= -0.15, (
        "reject_spam_active must debit -0.15 under infra_penalty"
    )


def test_reject_spam_resets_on_non_reject(env: UnifiedFintechEnv) -> None:
    """A single non-Reject decision must clear the consecutive_rejects counter."""
    _force_obs(env, kafka_lag=50.0, rolling_p99=50.0, risk_score=30.0)
    env._rolling_lag = 0.0

    for _ in range(6):
        env.step(make_action(risk_decision=1))

    # Approve once β€” resets streak
    _force_obs(env, kafka_lag=50.0, rolling_p99=50.0, risk_score=30.0)
    _, _, _, info = env.step(make_action(risk_decision=0))
    assert info["reject_spam_active"] is False
    assert info["consecutive_rejects"] == 0


# ---------------------------------------------------------------------------
# Test 26 β€” throughput bonus (Issue 6 fix)
# ---------------------------------------------------------------------------

def test_throughput_bonus_fires_on_approve_low_risk_healthy_lag(env: UnifiedFintechEnv) -> None:
    """Approve + risk_score < 40 + kafka_lag < 1200 must award +0.03 throughput bonus."""
    _force_obs(env, kafka_lag=100.0, rolling_p99=50.0, risk_score=20.0)
    env._rolling_lag = 0.0
    env._rolling_p99 = 50.0
    env._kafka_lag = 100.0

    _, _, _, info = env.step(make_action(risk_decision=0))  # Approve
    assert info["throughput_bonus_active"] is True
    assert info["reward_breakdown"]["bonus"] >= 0.03, (
        f"Expected throughput bonus β‰₯ 0.03, got {info['reward_breakdown']['bonus']}"
    )


def test_throughput_bonus_does_not_fire_on_high_risk(env: UnifiedFintechEnv) -> None:
    """Approve + risk_score β‰₯ 40 must NOT trigger throughput bonus."""
    _force_obs(env, kafka_lag=100.0, rolling_p99=50.0, risk_score=50.0)
    env._rolling_lag = 0.0
    env._kafka_lag = 100.0

    _, _, _, info = env.step(make_action(risk_decision=0))  # Approve high-risk
    assert info["throughput_bonus_active"] is False


def test_throughput_bonus_does_not_fire_on_high_lag(env: UnifiedFintechEnv) -> None:
    """Approve + risk_score < 40 but kafka_lag β‰₯ 1200 must NOT trigger throughput bonus."""
    _force_obs(env, kafka_lag=1500.0, rolling_p99=50.0, risk_score=20.0)
    env._rolling_lag = 0.0
    env._kafka_lag = 1500.0

    _, _, _, info = env.step(make_action(risk_decision=0))
    assert info["throughput_bonus_active"] is False


# ---------------------------------------------------------------------------
# Test 27 β€” CircuitBreaker half-open mode (uncovered lines 1405-1411)
# ---------------------------------------------------------------------------

def test_circuit_breaker_enters_half_open_after_cb_half_open_after_steps(env: UnifiedFintechEnv) -> None:
    """After CB_HALF_OPEN_AFTER consecutive CircuitBreaker steps, CB enters half-open mode.
    Half-open: reduced penalty (-0.10) and probe checks lag against recovery threshold.
    """
    from unified_gateway import CB_HALF_OPEN_AFTER, CB_HALF_OPEN_PENALTY, CB_LAG_RECOVERY_THRESHOLD

    # Put lag well below recovery threshold so the half-open probe closes the breaker
    _force_obs(env, kafka_lag=500.0, rolling_p99=50.0, risk_score=10.0)
    env._rolling_lag = 0.0
    env._kafka_lag = 500.0

    # Step CB_HALF_OPEN_AFTER times with CircuitBreaker to reach half-open
    for _ in range(CB_HALF_OPEN_AFTER):
        _force_obs(env, kafka_lag=500.0, rolling_p99=50.0, risk_score=10.0)
        env._kafka_lag = 500.0
        env.step(make_action(infra_routing=2, risk_decision=1))  # CircuitBreaker + Reject

    # One more step in half-open with lag < CB_LAG_RECOVERY_THRESHOLD β†’ breaker closes
    _force_obs(env, kafka_lag=500.0, rolling_p99=50.0, risk_score=10.0)
    env._kafka_lag = 500.0
    _, typed_reward, _, info = env.step(make_action(infra_routing=2, risk_decision=1))

    # Breaker should have closed (counter reset to 0) β€” bonus should be awarded
    cb_steps = info.get("circuit_breaker_steps", env._cb_consecutive_steps)
    # After closing, cb_consecutive_steps resets to 0
    assert env._cb_consecutive_steps == 0, (
        f"Breaker should have closed (cb_steps=0), got {env._cb_consecutive_steps}"
    )