i- R d Z ddlmZ ddlZddlmc mZ ddl Z ddl
m Z m Z m
Z
ddlmZ ddlmZ d d dZ d d dZ G d d
Z G d d Z G d
d Z G d d Z G d d Z G d d Z G d d Z G d d Z G d d Z G d d Z y)!u=
Tests for Story 4.06 — TierRouterInterceptor
Black Box (BB) tests: 5
White Box (WB) tests: 5
Total: 10
All router and telemetry dependencies are fully mocked via the constructor so
no Redis, no file I/O, and no actual classification logic is exercised here.
Run:
pytest tests/track_b/test_story_4_06.py -v
) annotationsN) MagicMock AsyncMockpatch)TierRouterInterceptorRoutingDecisionc t | ||| S )N)tiermodel rationale is_cachedr )r r r
cacheds 6/mnt/e/genesis-system/tests/track_b/test_story_4_06.py _decisionr s EYRX YY c t }t }|||j _ n/| | |j _ nt dd |j _ |||j
_ t
|| }|||fS )z{
Build a TierRouterInterceptor with fully injected mocks.
Returns (interceptor, mock_router, mock_telemetry).
T1 gemini-flash)router telemetry)r route side_effect return_valuer recordr )decision
router_raisestelemetry_raises mock_routermock_telemetry interceptors r _make_interceptorr" ! s| +K [N (5 % )1 &)24 )H &#,< )'{nUK ^ 33r c ( e Zd ZdZej
j d Zej
j d Zej
j d Z ej
j d Z
ej
j d Z y) TestBB1_PayloadHasTierAndModelz>BB1: After pre_execute, payload has 'tier' and 'model' fields.c K t t dd \ }}}|j ddi d { }d}||v }|st j d|fd||f t j
| dt
j v st j | rt j
| ndd z }d
d |iz }t t j | d x}}y 7 w) NT0python_functiontype health_checkr inz%(py1)s in %(py3)sresultpy1py3assert %(py5)spy5 r" r pre_execute
@pytest_ar_call_reprcompare _saferepr @py_builtinslocals_should_repr_global_nameAssertionError_format_explanationselfr! _r- @py_assert0 @py_assert2 @py_format4 @py_format6s r #test_tier_present_after_pre_executezBTestBB1_PayloadHasTierAndModel.test_tier_present_after_pre_executeD s -i>O.P Q Q"../GHH v v v I 0C C B+C c K t t dd \ }}}|j ddi d { }d}||v }|st j d|fd||f t j
| dt
j v st j | rt j
| ndd z }d
d |iz }t t j | d x}}y 7 w) Nr& r' r( r) r r* r, r- r. r1 r2 r3 r= s r $test_model_present_after_pre_executezCTestBB1_PayloadHasTierAndModel.test_model_present_after_pre_executeJ s -i>O.P Q Q"../GHH w& w& w & & IrE c K t t ddd \ }}}|j ddi d { }d}||v }|st j d|fd||f t j
| d t
j v st j | rt j
| nd d
z }d d |iz }t t j | d x}}y 7 w)
Nr& r' zmatched T0_TYPESr( r) routing_rationaler* r, r- r. r1 r2 r3 r= s r 0test_routing_rationale_present_after_pre_executezOTestBB1_PayloadHasTierAndModel.test_routing_rationale_present_after_pre_executeP s -i>OQc.d e Q"../GHH","f,,,,"f,,,",,,,,,f,,,f,,,,,,, Is 1C C B+C c x K t t dd \ }}}|j ddi d { }|d }d}||k( }|slt j d|fd||f t j
| t j
| dz }d d
|iz }t
t j | d x}x}}y 7 w) Nr r r( email_draftr ==z%(py1)s == %(py4)sr/ py4assert %(py6)spy6r" r r4 r5 r6 r7 r; r< r> r! r? r- r@ @py_assert3rA @py_format5 @py_format7s r 'test_tier_value_matches_router_decisionzFTestBB1_PayloadHasTierAndModel.test_tier_value_matches_router_decisionV s -in.M N Q"..
/FGGf~%%~%%%%~%%%~%%%%%%%%%% H 0B:B8BB:c x K t t dd \ }}}|j ddi d { }|d }d}||k( }|slt j d|fd||f t j
| t j
| dz }d d
|iz }t
t j | d x}x}}y 7 w) NT2claude-opus-4-6r( complex_reasoningr rM rO rP rR rS rT rU s r (test_model_value_matches_router_decisionzGTestBB1_PayloadHasTierAndModel.test_model_value_matches_router_decision\ s -i>O.P Q Q"..8K/LMMg3"33"33333"3333333"33333333 NrZ N) __name__
__module__ __qualname____doc__pytestmarkasynciorD rG rJ rY r_ r r r$ r$ A s H [[
[[! !
[[- -
[[& &
[[4 4r r$ c e Zd ZdZej
j d Zej
j d Zej
j d Z y) TestBB2_RoutingErrorDefaultsToT2uN BB2: Routing error → payload defaults to tier='T2', model='claude-opus-4-6'.c x K t t d \ }}}|j ddi d { }|d }d}||k( }|slt j d|fd||f t j
| t j
| d z }d
d |iz }t
t j | d x}x}}y 7 w) Nzclassifier downr r( anythingr r\ rM rO rP rR rS r" RuntimeErrorr4 r5 r6 r7 r; r< rU s r "test_router_exception_sets_t2_tierzCTestBB2_RoutingErrorDefaultsToT2.test_router_exception_sets_t2_tierf s -LIZ<[ \ Q"..
/CDDf~%%~%%%%~%%%~%%%%%%%%%% ErZ c x K t t d \ }}}|j ddi d { }|d }d}||k( }|slt j d|fd||f t j
| t j
| d z }d
d |iz }t
t j | d x}x}}y 7 w) Nz bad payloadrk r( rl r r] rM rO rP rR rS )r"
ValueErrorr4 r5 r6 r7 r; r< rU s r %test_router_exception_sets_opus_modelzFTestBB2_RoutingErrorDefaultsToT2.test_router_exception_sets_opus_modell s -J} >r ri c ( e Zd ZdZej
j d Zej
j d Zej
j d Z ej
j d Z
ej
j d Z y) TestBB3_OnErrorReturnsFallbacksz1BB3: on_error returns dict with T2/Opus defaults.c K t \ }}}|j t d ddi d { }t |t }|s ddt j v st j t rt j t nddt j v st j | rt j | nddt j v st j t rt j t ndt j | dz }t t j | d }y 7 &w) Nboomr( xz5assert %(py4)s
{%(py4)s = %(py0)s(%(py1)s, %(py2)s)
}
isinstancer- dict)py0r/ py2rQ ) r" on_errorrn r| r} r8 r9 r5 r: r7 r; r< )r> r! r? r- rV rW s r test_on_error_returns_dictz:TestBB3_OnErrorReturnsFallbacks.test_on_error_returns_dict| s - / Q"++L,@63-PP&$''''''''z'''z''''''&'''&''''''$'''$'''''''''' Qs /EED&Ec v K t \ }}}|j t d ddi d { }|d }d}||k( }|slt j d|fd||f t j
| t j
| dz }d d
|iz }t
t j | d x}x}}y 7 w) Nrz r( r{ r r\ rM rO rP rR rS r" r rn r5 r6 r7 r; r< rU s r test_on_error_tier_is_t2z8TestBB3_OnErrorReturnsFallbacks.test_on_error_tier_is_t2 s - / Q"++L,@63-PPf~%%~%%%%~%%%~%%%%%%%%%% Q /B9B7BB9c v K t \ }}}|j t d ddi d { }|d }d}||k( }|slt j d|fd||f t j
| t j
| dz }d d
|iz }t
t j | d x}x}}y 7 w) Nrz r( r{ r r] rM rO rP rR rS r rU s r test_on_error_model_is_opusz;TestBB3_OnErrorReturnsFallbacks.test_on_error_model_is_opus s - / Q"++L,@63-PPg3"33"33333"3333333"33333333 Qr c x K t \ }}}t d }|j |ddi d { }d}|d }||v }|slt j d|fd||f t j
| t j
| dz }dd |iz } t
t j | d x}x}}y 7 w)
Nzsomething went wrongr( r{ errorr* ru rP rR rS )r" rn r r5 r6 r7 r; r< )
r> r! r? r r- r@ rV rA rW rX s
r #test_on_error_contains_error_stringzCTestBB3_OnErrorReturnsFallbacks.test_on_error_contains_error_string s - / Q 34"++EFC=AA%8 8% 8888% 888%888 8888888 Bs 1B:B8BB:c K t \ }}}ddd}|j t d | d { }|d }||u }|st j d|fd||f t j
| dt
j v st j | rt j
| ndd z }d
d |iz }t t j | d x}}y 7 w) Nr{ abc)r( task_iderr task_payloadis)z%(py1)s is %(py3)spayloadr. r1 r2 ) r" r rn r5 r6 r7 r8 r9 r: r; r< ) r> r! r? r r- r@ rA rB rC s r 'test_on_error_contains_original_payloadzGTestBB3_OnErrorReturnsFallbacks.test_on_error_contains_original_payload s - / Q 51"++L,?IIn%0%0000%000%0000000000000000 Js 2C$C"B.C$N) r` ra rb rc rd re rf r r r r r rg r r rx rx y s ; [[( (
[[& &
[[4 4
[[9 9 [[1 1r rx c e Zd ZdZej
j d Zej
j d Zy) TestBB4_OnCorrectionPassthroughz;BB4: on_correction returns unchanged payload (passthrough).c K t \ }}}ddd}|j | d { }||u }|st j d|fd||f dt j
v st j | rt j | nddt j
v st j | rt j | nddz }d d
|iz }t t j | d }y 7 ӭw) NTr\ )zCORRECTION: retryr r )z%(py0)s is %(py2)sr- r r~ r assert %(py4)srQ )
r"
on_correctionr5 r6 r8 r9 r: r7 r; r< )r> r! r? r r- @py_assert1 @py_format3rW s r $test_on_correction_returns_same_dictzDTestBB4_OnCorrectionPassthrough.test_on_correction_returns_same_dict s - / Q(,d;"0099 v v v :s (D C>CD c K t \ }}}ddd}t | }|j | d { ||k( }|st j d|fd||f dt j v st j | rt j | nddt j v st j | rt j | nddz }d d
|iz }t t j | d }y 7 ԭw) Nvaluer\ )keyr rM z%(py0)s == %(py2)sr
original_copyr r rQ ) r" r} r r5 r6 r8 r9 r: r7 r; r< )r> r! r? r r r r rW s r *test_on_correction_does_not_mutate_payloadzJTestBB4_OnCorrectionPassthrough.test_on_correction_does_not_mutate_payload s - / Q!40 W
''000-''''w-''''''w'''w''''''-'''-''''''' 1s 3D D
CD N) r` ra rb rc rd re rf r r rg r r r r s? E [[! ! [[( (r r c e Zd ZdZej
j d Zej
j d Zej
j d Z y)%TestBB5_PostExecuteRecordsToTelemetryz'BB5: post_execute records to telemetry.c K t dd }t | \ }}}|j ddi d { |j i ddi d { |j j | y 7 <7 "wNr& r' r( r) )r r" r4 post_executer assert_called_once_withr> r r! r? r s r (test_post_execute_calls_telemetry_recordzNTestBB5_PostExecuteRecordsToTelemetry.test_post_execute_calls_telemetry_record sv T#45):8)D& Q%%v~&>???&&rFN+CDDD 55h? @Ds! 2A5A1A5A3!A53A5c K t dd }t |t d \ }}}|j ddi d { |j i ddi d { y 7 !7 w)Nr r z
redis down)r r( rL )r r" rn r4 r )r> r r! r? s r 4test_post_execute_does_not_crash_if_telemetry_raiseszZTestBB5_PostExecuteRecordsToTelemetry.test_post_execute_does_not_crash_if_telemetry_raises sh T>2 -h VbIc d Q%%v}&=>>>&&rFM+BCCC ?Cs" =A%A! A%A# A%#A%c K t \ }}}|j i i d{ |j j y7 w)uN If pre_execute was never called, _last_decision is None — no call to record.N)r" r r assert_not_called)r> r! r? r s r 0test_post_execute_skips_telemetry_if_no_decisionzVTestBB5_PostExecuteRecordsToTelemetry.test_post_execute_skips_telemetry_if_no_decision sD *;)<& Q&&r2... //1 /s $AA AN)
r` ra rb rc rd re rf r r r rg r r r r sa 1 [[@ @ [[D D [[2 2r r c e Zd ZdZd Zy)TestWB1_PriorityIs20z WB1: metadata.priority == 20.c t \ }}}|j }|j }d}||k( }|st j d|fd||f dt j v st j | rt j | ndt j | t j | t j | dz }dd|iz }t t j | d x}x}x}}y )N rM )zN%(py4)s
{%(py4)s = %(py2)s
{%(py2)s = %(py0)s.metadata
}.priority
} == %(py7)sr! r~ r rQ py7assert %(py9)spy9) r" metadatapriorityr5 r6 r8 r9 r: r7 r; r< r> r! r? r rV @py_assert6 @py_assert5 @py_format8 @py_format10s r test_metadata_priorityz+TestWB1_PriorityIs20.test_metadata_priority s - / Q##2#,,22,2222,222222{222{222#222,2222222222r N)r` ra rb rc r rg r r r r s
'3r r c e Zd ZdZej
j d Zej
j d Zy))TestWB2_TelemetryRecordCalledInPostNotPrezHWB2: RoutingTelemetry.record called in post_execute, NOT in pre_execute.c K t dd }t | \ }}}|j ddi d { |j j y 7 wr )r r" r4 r r r s r )test_record_not_called_during_pre_executezSTestWB2_TelemetryRecordCalledInPostNotPre.test_record_not_called_during_pre_execute sU T#45):8)D& Q%%v~&>??? //1 @s 2AA Ac K t dd }t | \ }}}|j ddi d { |j i ddi d { |j j y 7 ;7 !wr )r r" r4 r r assert_called_oncer s r &test_record_called_during_post_executezPTestWB2_TelemetryRecordCalledInPostNotPre.test_record_called_during_post_execute st T#45):8)D& Q%%v~&>???&&rFN+CDDD 002 @Ds! 2A4A0A4A2 A42A4N) r` ra rb rc rd re rf r r rg r r r r s? R [[2 2 [[3 3r r c e Zd ZdZd Zy)TestWB3_MetadataNamez$WB3: metadata.name == 'tier_router'.c t \ }}}|j }|j }d}||k( }|st j d|fd||f dt j v st j | rt j | ndt j | t j | t j | dz }dd|iz }t t j | d x}x}x}}y )N tier_routerrM )zJ%(py4)s
{%(py4)s = %(py2)s
{%(py2)s = %(py0)s.metadata
}.name
} == %(py7)sr! r r r ) r" r namer5 r6 r8 r9 r: r7 r; r< r s r test_metadata_namez'TestWB3_MetadataName.test_metadata_name s - / Q##9#((9M9(M9999(M999999{999{999#999(999M9999999r N)r` ra rb rc r rg r r r r s
.:r r c e Zd ZdZej
j d Zej
j d Zej
j d Z y)*TestWB4_LastDecisionUpdatedAfterPreExecutez1WB4: _last_decision is updated after pre_execute.c K t \ }}}|j }d }||u }|st j d|fd||f dt j
v st j | rt j | ndt j | t j | dz }dd|iz }t t j | d x}x}}y w)Nr )z6%(py2)s
{%(py2)s = %(py0)s._last_decision
} is %(py5)sr! r~ r r2 assert %(py7)sr )
r" _last_decisionr5 r6 r8 r9 r: r7 r; r< )r> r! r? r @py_assert4rV rC r s r *test_last_decision_none_before_pre_executezUTestWB4_LastDecisionUpdatedAfterPreExecute.test_last_decision_none_before_pre_execute s - / Q))1T1)T1111)T111111{111{111)111T1111111s CCc K t ddd }t | \ }}}|j ddi d { |j }d }||u}|st j
d|fd||f dt
j v st j | rt j | ndt j | t j | d z }d
d |iz }t t j | d x}x}}|j }|j }d} || k( }
|
st j
d |
fd
|| f dt
j v st j | rt j | ndt j | t j | t j | dz }dd|iz } t t j | d x}x}x}
} y 7 w)Nr r zmatched T1_TYPESr( rL is notz:%(py2)s
{%(py2)s = %(py0)s._last_decision
} is not %(py5)sr! r r r rM zP%(py4)s
{%(py4)s = %(py2)s
{%(py2)s = %(py0)s._last_decision
}.tier
} == %(py7)sr r r )
r r" r4 r r5 r6 r8 r9 r: r7 r; r< r ) r> r r! r? r r rV rC r r r r s r (test_last_decision_set_after_pre_executezSTestWB4_LastDecisionUpdatedAfterPreExecute.test_last_decision_set_after_pre_execute s@ T>3EF -h 7 Q%%v}&=>>>))55)5555)555555{555{555)5555555555))6)..6$6.$6666.$666666{666{666)666.666$6666666 ?s 3G/G,F7G/c K t t d \ }}}|j ddi d{ |j }d}||u}|st j
d|fd||f dt
j v st j | rt j | ndt j | t j | d z }d
d |iz }t t j | dx}x}}|j }|j }d }||k( } | st j
d
| fd||f dt
j v st j | rt j | ndt j | t j | t j | dz }dd|iz }
t t j |
dx}x}x} }y7 w)zAEven on routing error, _last_decision is populated with fallback.rt rk r( rl Nr r r! r r r r\ rM r r r r )
r" rn r4 r r5 r6 r8 r9 r: r7 r; r< r ) r> r! r? r r rV rC r r r r s r 'test_last_decision_set_on_routing_errorzRTestWB4_LastDecisionUpdatedAfterPreExecute.test_last_decision_set_on_routing_error s8 .L r! r? r- r r r rW s r (test_result_unchanged_after_post_executezOTestWB5_PostExecuteDoesNotMutateResult.test_result_unchanged_after_post_execute s -i>O.P Q Q%%v&?@@@ (D9 V
&&v/HIII &&&&v &&&&&&v&&&v&&&&&& &&& &&&&&&& A Js" 0D8D3+D8 D6 CD86D8c K t t dd \ }}}|j ddi d { |j i i d { }d }||u }|st j
d|fd||f dt
j v st j | rt j | ndt j | dz }d d
|iz }t t j | d x}}y 7 7 w) Nr& r' r( cache_getr )z%(py0)s is %(py3)sret)r~ r0 r1 r2 ) r" r r4 r r5 r6 r8 r9 r: r7 r; r< )r> r! r? r rA r rB rC s r test_post_execute_returns_nonezETestWB5_PostExecuteDoesNotMutateResult.test_post_execute_returns_none s -i>O.P Q Q%%v{&;<<< ,,R44sd{sdssd =4s" 0C;C7C; C9
B+C;9C;N) r` ra rb rc rd re rf r r rg r r r r
s? 3 [[' ' [[ r r )testF)
r strr r r
r r boolreturnr )NNN)r zRoutingDecision | Noner Exception | Noner r r z2tuple[TierRouterInterceptor, MagicMock, MagicMock]) rc
__future__r builtinsr8 _pytest.assertion.rewrite assertionrewriter5 rd
unittest.mockr r r $core.routing.tier_router_interceptorr core.routing.tier_classifierr r r" r$ ri rx r r r r r r r rg r r r s #
5 5 F 8Z
(,&*)-4$4#4 '4 8 4@ 4 4D> >,!1 !1H( (&2 2B3 33 3&: :7 72 r