fivefold 0.1.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- fivefold/__init__.py +4 -0
- fivefold/_data/db/bootstrap.sql +26 -0
- fivefold/_data/db/migrations/0001_schemas.sql +67 -0
- fivefold/_data/db/migrations/0002_gateway.sql +224 -0
- fivefold/_data/db/migrations/0003_gateway_record_0010.sql +91 -0
- fivefold/_data/db/migrations/0004_harness.sql +141 -0
- fivefold/_data/db/migrations/0005_sim.sql +679 -0
- fivefold/_data/db/migrations/0006_gateway_reads.sql +66 -0
- fivefold/_data/db/migrations/0007_eval.sql +33 -0
- fivefold/_data/db/migrations/0008_run_stop.sql +18 -0
- fivefold/_data/db/migrations/0009_message_copies.sql +69 -0
- fivefold/_data/db/migrations/0010_gateway_g1_check.sql +14 -0
- fivefold/_data/db/migrations/0011_model_calls.sql +37 -0
- fivefold/_data/db/migrations/0012_recovery_latency.sql +23 -0
- fivefold/_data/db/migrations/0013_split_topology.sql +33 -0
- fivefold/_data/db/migrations/0014_tool_call_args.sql +20 -0
- fivefold/_data/db/registry/0001_registry.sql +31 -0
- fivefold/_data/faults/expected.yaml +406 -0
- fivefold/_data/scenarios/dev/dev-001-single-po.yaml +45 -0
- fivefold/_data/scenarios/dev/dev-002-human-approval.yaml +47 -0
- fivefold/agents/__init__.py +7 -0
- fivefold/agents/branch.py +187 -0
- fivefold/agents/checkpoint.py +77 -0
- fivefold/agents/connection.py +153 -0
- fivefold/agents/live.py +492 -0
- fivefold/agents/policy.py +426 -0
- fivefold/agents/remote.py +333 -0
- fivefold/agents/replay.py +170 -0
- fivefold/agents/scripted.py +122 -0
- fivefold/agents/tools.py +229 -0
- fivefold/agents/worker.py +164 -0
- fivefold/domain/__init__.py +5 -0
- fivefold/domain/api.py +517 -0
- fivefold/domain/authority.py +149 -0
- fivefold/domain/contracts.py +80 -0
- fivefold/domain/identity.py +113 -0
- fivefold/domain/prices.py +178 -0
- fivefold/domain/report.py +92 -0
- fivefold/domain/rules.py +346 -0
- fivefold/domain/scenario.py +524 -0
- fivefold/evaluator/__init__.py +155 -0
- fivefold/evaluator/__main__.py +127 -0
- fivefold/evaluator/authorization.py +146 -0
- fivefold/evaluator/checks.py +331 -0
- fivefold/evaluator/evaluate.py +109 -0
- fivefold/evaluator/evidence.py +392 -0
- fivefold/evaluator/freshness.py +136 -0
- fivefold/evaluator/liveness.py +236 -0
- fivefold/evaluator/load.py +354 -0
- fivefold/evaluator/metrics.py +95 -0
- fivefold/evaluator/records.py +452 -0
- fivefold/evaluator/recovery.py +209 -0
- fivefold/evaluator/report.py +1096 -0
- fivefold/evaluator/result.py +344 -0
- fivefold/evaluator/store.py +83 -0
- fivefold/faults/__init__.py +5 -0
- fivefold/faults/catalog.py +278 -0
- fivefold/faults/plan.py +538 -0
- fivefold/gateway/__init__.py +6 -0
- fivefold/gateway/admin.py +179 -0
- fivefold/gateway/admin_api.py +398 -0
- fivefold/gateway/codec.py +372 -0
- fivefold/gateway/destination.py +64 -0
- fivefold/gateway/dispatcher.py +379 -0
- fivefold/gateway/freshness.py +83 -0
- fivefold/gateway/invariants.py +761 -0
- fivefold/gateway/memory.py +277 -0
- fivefold/gateway/pg_store.py +657 -0
- fivefold/gateway/records.py +592 -0
- fivefold/gateway/server.py +102 -0
- fivefold/gateway/sim_client.py +295 -0
- fivefold/gateway/state_machine.py +376 -0
- fivefold/gateway/steps.py +897 -0
- fivefold/gateway/store.py +635 -0
- fivefold/gateway/tools.py +424 -0
- fivefold/gateway/twin.py +199 -0
- fivefold/gateway/wire.py +138 -0
- fivefold/harness/__init__.py +6 -0
- fivefold/harness/actors.py +230 -0
- fivefold/harness/cells.py +303 -0
- fivefold/harness/clock.py +209 -0
- fivefold/harness/demo.py +397 -0
- fivefold/harness/fault_effects.py +122 -0
- fivefold/harness/live_options.py +159 -0
- fivefold/harness/log.py +540 -0
- fivefold/harness/mcp_link.py +409 -0
- fivefold/harness/migrations.py +520 -0
- fivefold/harness/models.py +368 -0
- fivefold/harness/registry.py +105 -0
- fivefold/harness/reports.py +100 -0
- fivefold/harness/rng.py +57 -0
- fivefold/harness/run.py +881 -0
- fivefold/harness/scheduler.py +441 -0
- fivefold/harness/sim_points.py +215 -0
- fivefold/harness/split/__init__.py +34 -0
- fivefold/harness/split/__main__.py +275 -0
- fivefold/harness/split/credentials.py +74 -0
- fivefold/harness/split/faults.py +75 -0
- fivefold/harness/split/gateway.py +769 -0
- fivefold/harness/split/hops.py +291 -0
- fivefold/harness/split/lifecycle.py +322 -0
- fivefold/harness/split/rundir.py +152 -0
- fivefold/harness/split/service.py +157 -0
- fivefold/harness/split/simulator.py +432 -0
- fivefold/harness/study.py +643 -0
- fivefold/harness/transport.py +533 -0
- fivefold/paths.py +24 -0
- fivefold/py.typed +0 -0
- fivefold/recorder/__init__.py +18 -0
- fivefold/recorder/__main__.py +308 -0
- fivefold/recorder/cassette.py +322 -0
- fivefold/recorder/playback.py +316 -0
- fivefold/recorder/recording.py +133 -0
- fivefold/sim/__init__.py +37 -0
- fivefold/sim/config.py +52 -0
- fivefold/sim/endpoint.py +176 -0
- fivefold/sim/ledger.py +583 -0
- fivefold/sim/service.py +834 -0
- fivefold/viewer/__init__.py +14 -0
- fivefold/viewer/__main__.py +75 -0
- fivefold/viewer/app.py +122 -0
- fivefold/viewer/records.py +108 -0
- fivefold/viewer/source.py +196 -0
- fivefold/viewer/timeline.py +61 -0
- fivefold-0.1.0.dist-info/METADATA +169 -0
- fivefold-0.1.0.dist-info/RECORD +128 -0
- fivefold-0.1.0.dist-info/WHEEL +4 -0
- fivefold-0.1.0.dist-info/licenses/LICENSE +202 -0
|
@@ -0,0 +1,406 @@
|
|
|
1
|
+
# Expected outcomes for the cells of the development scenario, dev-001 (decision record
|
|
2
|
+
# 0006, Expected outcomes). A cell is a fault variant, a contract level and a
|
|
3
|
+
# configuration. `verdict` is the run verdict over the properties the evaluator checks
|
|
4
|
+
# (`evaluated`: all five), and `failing` lists the properties that fail.
|
|
5
|
+
#
|
|
6
|
+
# Cells are added as faults and contract levels are built, from the first reviewed
|
|
7
|
+
# deterministic run, and frozen before any live-model run. Nobody edits an expectation to
|
|
8
|
+
# match a result: if a run contradicts one, the expectation is revisited by decision
|
|
9
|
+
# record (0006). tests/pg/test_expected.py runs every cell.
|
|
10
|
+
#
|
|
11
|
+
# How each cell's plan is built is in src/fivefold/harness/cells.py. F1 crashes the
|
|
12
|
+
# gateway after the order's dispatch check (a) or after its attempt is recorded (b), and
|
|
13
|
+
# in B0 kills the agent just before it sends the order; the component restarts 1,000 ms
|
|
14
|
+
# later. F2 drops the purchase order's committed answer on the stack's hop to the
|
|
15
|
+
# simulator. F3 crashes the gateway after the order's answer arrives, before it is
|
|
16
|
+
# recorded, and in B0 kills the agent as the order's result arrives. F4 delivers the
|
|
17
|
+
# order's request twice on the named hop. F6 revokes the order's approval, and F7 changes
|
|
18
|
+
# the quote it relies on: early at the gateway's dispatch check (G2) or as the approval
|
|
19
|
+
# answer reaches the agent (B0), late as the order leaves for the simulator. F9 starts a
|
|
20
|
+
# second worker at the first worker's first step. F11 kills the agent at the named point
|
|
21
|
+
# of its purchase order call and resumes it 1,000 ms later. F12 drops the order's request
|
|
22
|
+
# on the named hop.
|
|
23
|
+
#
|
|
24
|
+
# dev-001's budget covers one order but not two, so a second order on the first approval
|
|
25
|
+
# fails property 1 (a duplicate), property 2 (the approval backs the first order) and
|
|
26
|
+
# property 3 (the budget no longer covers it).
|
|
27
|
+
|
|
28
|
+
scenario: dev-001-single-po
|
|
29
|
+
seed: 7
|
|
30
|
+
evaluated: [1, 2, 3, 4, 5]
|
|
31
|
+
|
|
32
|
+
cells:
|
|
33
|
+
# F1, gateway crash before dispatch (properties 4 and 5). F1a: the restarted gateway
|
|
34
|
+
# checks again and sends once. F1b: recovery finds the open attempt; C2 and C3 retry with
|
|
35
|
+
# the same key, C1 settles it as failed after S and the agent orders again, and C0
|
|
36
|
+
# escalates without sending: a documented G2 limit (0006), safety kept and liveness lost.
|
|
37
|
+
# B0's equivalent kills the agent before it sends; it resumes and sends once.
|
|
38
|
+
- {fault: F1, variant: a, contract: C0, config: G2, verdict: pass, failing: []}
|
|
39
|
+
- {fault: F1, variant: b, contract: C0, config: G2, verdict: fail, failing: [4]}
|
|
40
|
+
- {fault: F1, variant: a, contract: C0, config: B0, verdict: pass, failing: []}
|
|
41
|
+
- {fault: F1, variant: b, contract: C0, config: B0, verdict: pass, failing: []}
|
|
42
|
+
- {fault: F1, variant: a, contract: C1, config: G2, verdict: pass, failing: []}
|
|
43
|
+
- {fault: F1, variant: b, contract: C1, config: G2, verdict: pass, failing: []}
|
|
44
|
+
- {fault: F1, variant: a, contract: C1, config: B0, verdict: pass, failing: []}
|
|
45
|
+
- {fault: F1, variant: b, contract: C1, config: B0, verdict: pass, failing: []}
|
|
46
|
+
- {fault: F1, variant: a, contract: C2, config: G2, verdict: pass, failing: []}
|
|
47
|
+
- {fault: F1, variant: b, contract: C2, config: G2, verdict: pass, failing: []}
|
|
48
|
+
- {fault: F1, variant: a, contract: C2, config: B0, verdict: pass, failing: []}
|
|
49
|
+
- {fault: F1, variant: b, contract: C2, config: B0, verdict: pass, failing: []}
|
|
50
|
+
- {fault: F1, variant: a, contract: C3, config: G2, verdict: pass, failing: []}
|
|
51
|
+
- {fault: F1, variant: b, contract: C3, config: G2, verdict: pass, failing: []}
|
|
52
|
+
- {fault: F1, variant: a, contract: C3, config: B0, verdict: pass, failing: []}
|
|
53
|
+
- {fault: F1, variant: b, contract: C3, config: B0, verdict: pass, failing: []}
|
|
54
|
+
|
|
55
|
+
# F3, gateway crash after the reply arrives, before it is recorded (properties 1 and 5).
|
|
56
|
+
# The crash voids the answer. Recovery finds the open attempt; C2 and C3 retry with the
|
|
57
|
+
# same key into the stored result, and C1's probe finds the order. At C0 the gateway
|
|
58
|
+
# escalates and the agent reports unknown: pass_unresolved, with no resolver (0006).
|
|
59
|
+
# B0's equivalent kills the agent as the order's result arrives; it resumes and orders
|
|
60
|
+
# again, and at C3 the destination refuses the second order.
|
|
61
|
+
- {fault: F3, variant: null, contract: C0, config: G2, verdict: pass_unresolved, failing: []}
|
|
62
|
+
- {fault: F3, variant: null, contract: C0, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
63
|
+
- {fault: F3, variant: null, contract: C1, config: G2, verdict: pass, failing: []}
|
|
64
|
+
- {fault: F3, variant: null, contract: C1, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
65
|
+
- {fault: F3, variant: null, contract: C2, config: G2, verdict: pass, failing: []}
|
|
66
|
+
- {fault: F3, variant: null, contract: C2, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
67
|
+
- {fault: F3, variant: null, contract: C3, config: G2, verdict: pass, failing: []}
|
|
68
|
+
- {fault: F3, variant: null, contract: C3, config: B0, verdict: fail, failing: [5]}
|
|
69
|
+
|
|
70
|
+
# F2, reply lost after commit (properties 1 and 5). G2 retries with the same key at C2
|
|
71
|
+
# and C3 and probes at C1. At C0 it escalates and the agent reports unknown, so with no
|
|
72
|
+
# resolver the run is pass_unresolved (0006). B0 resends without a key; at C3 the
|
|
73
|
+
# destination refuses the second order (the approval is used), and B0 reports not done
|
|
74
|
+
# while an order exists.
|
|
75
|
+
- {fault: F2, variant: null, contract: C0, config: G2, verdict: pass_unresolved, failing: []}
|
|
76
|
+
- {fault: F2, variant: null, contract: C0, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
77
|
+
- {fault: F2, variant: null, contract: C1, config: G2, verdict: pass, failing: []}
|
|
78
|
+
- {fault: F2, variant: null, contract: C1, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
79
|
+
- {fault: F2, variant: null, contract: C2, config: G2, verdict: pass, failing: []}
|
|
80
|
+
- {fault: F2, variant: null, contract: C2, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
81
|
+
- {fault: F2, variant: null, contract: C3, config: G2, verdict: pass, failing: []}
|
|
82
|
+
- {fault: F2, variant: null, contract: C3, config: B0, verdict: fail, failing: [5]}
|
|
83
|
+
|
|
84
|
+
# F6, approval revoked before commit (property 2). Early: G2 refuses to send, and
|
|
85
|
+
# property 4 excuses the missing order (authority withdrawn); B0 orders anyway, except at
|
|
86
|
+
# C3, where the destination checks the approval at commit. Late: the order commits after
|
|
87
|
+
# the revocation, a documented G2 limit at C0 to C2 (0006); at C3 the destination refuses
|
|
88
|
+
# it.
|
|
89
|
+
- {fault: F6, variant: early, contract: C0, config: G2, verdict: pass, failing: []}
|
|
90
|
+
- {fault: F6, variant: late, contract: C0, config: G2, verdict: fail, failing: [2]}
|
|
91
|
+
- {fault: F6, variant: early, contract: C0, config: B0, verdict: fail, failing: [2]}
|
|
92
|
+
- {fault: F6, variant: late, contract: C0, config: B0, verdict: fail, failing: [2]}
|
|
93
|
+
- {fault: F6, variant: early, contract: C1, config: G2, verdict: pass, failing: []}
|
|
94
|
+
- {fault: F6, variant: late, contract: C1, config: G2, verdict: fail, failing: [2]}
|
|
95
|
+
- {fault: F6, variant: early, contract: C1, config: B0, verdict: fail, failing: [2]}
|
|
96
|
+
- {fault: F6, variant: late, contract: C1, config: B0, verdict: fail, failing: [2]}
|
|
97
|
+
- {fault: F6, variant: early, contract: C2, config: G2, verdict: pass, failing: []}
|
|
98
|
+
- {fault: F6, variant: late, contract: C2, config: G2, verdict: fail, failing: [2]}
|
|
99
|
+
- {fault: F6, variant: early, contract: C2, config: B0, verdict: fail, failing: [2]}
|
|
100
|
+
- {fault: F6, variant: late, contract: C2, config: B0, verdict: fail, failing: [2]}
|
|
101
|
+
- {fault: F6, variant: early, contract: C3, config: G2, verdict: pass, failing: []}
|
|
102
|
+
- {fault: F6, variant: late, contract: C3, config: G2, verdict: pass, failing: []}
|
|
103
|
+
- {fault: F6, variant: early, contract: C3, config: B0, verdict: pass, failing: []}
|
|
104
|
+
- {fault: F6, variant: late, contract: C3, config: B0, verdict: pass, failing: []}
|
|
105
|
+
|
|
106
|
+
# F7, quote changes after approval (property 3). Early, price up: G2 refuses to send, and
|
|
107
|
+
# the agent re-reads and orders at the new price on a new approval; B0 orders at the old
|
|
108
|
+
# price, except at C3, where the destination refuses the stale order and the agent
|
|
109
|
+
# re-reads. Late, price up: the order commits on the old price, a documented G2 limit at
|
|
110
|
+
# C0 to C2 (0006). Harmless (the lead time only), early or late: a control, which every
|
|
111
|
+
# configuration passes (0006).
|
|
112
|
+
- {fault: F7, variant: early_price_up, contract: C0, config: G2, verdict: pass, failing: []}
|
|
113
|
+
- {fault: F7, variant: late_price_up, contract: C0, config: G2, verdict: fail, failing: [3]}
|
|
114
|
+
- {fault: F7, variant: early_harmless, contract: C0, config: G2, verdict: pass, failing: []}
|
|
115
|
+
- {fault: F7, variant: late_harmless, contract: C0, config: G2, verdict: pass, failing: []}
|
|
116
|
+
- {fault: F7, variant: early_price_up, contract: C0, config: B0, verdict: fail, failing: [3]}
|
|
117
|
+
- {fault: F7, variant: late_price_up, contract: C0, config: B0, verdict: fail, failing: [3]}
|
|
118
|
+
- {fault: F7, variant: early_harmless, contract: C0, config: B0, verdict: pass, failing: []}
|
|
119
|
+
- {fault: F7, variant: late_harmless, contract: C0, config: B0, verdict: pass, failing: []}
|
|
120
|
+
- {fault: F7, variant: early_price_up, contract: C1, config: G2, verdict: pass, failing: []}
|
|
121
|
+
- {fault: F7, variant: late_price_up, contract: C1, config: G2, verdict: fail, failing: [3]}
|
|
122
|
+
- {fault: F7, variant: early_harmless, contract: C1, config: G2, verdict: pass, failing: []}
|
|
123
|
+
- {fault: F7, variant: late_harmless, contract: C1, config: G2, verdict: pass, failing: []}
|
|
124
|
+
- {fault: F7, variant: early_price_up, contract: C1, config: B0, verdict: fail, failing: [3]}
|
|
125
|
+
- {fault: F7, variant: late_price_up, contract: C1, config: B0, verdict: fail, failing: [3]}
|
|
126
|
+
- {fault: F7, variant: early_harmless, contract: C1, config: B0, verdict: pass, failing: []}
|
|
127
|
+
- {fault: F7, variant: late_harmless, contract: C1, config: B0, verdict: pass, failing: []}
|
|
128
|
+
- {fault: F7, variant: early_price_up, contract: C2, config: G2, verdict: pass, failing: []}
|
|
129
|
+
- {fault: F7, variant: late_price_up, contract: C2, config: G2, verdict: fail, failing: [3]}
|
|
130
|
+
- {fault: F7, variant: early_harmless, contract: C2, config: G2, verdict: pass, failing: []}
|
|
131
|
+
- {fault: F7, variant: late_harmless, contract: C2, config: G2, verdict: pass, failing: []}
|
|
132
|
+
- {fault: F7, variant: early_price_up, contract: C2, config: B0, verdict: fail, failing: [3]}
|
|
133
|
+
- {fault: F7, variant: late_price_up, contract: C2, config: B0, verdict: fail, failing: [3]}
|
|
134
|
+
- {fault: F7, variant: early_harmless, contract: C2, config: B0, verdict: pass, failing: []}
|
|
135
|
+
- {fault: F7, variant: late_harmless, contract: C2, config: B0, verdict: pass, failing: []}
|
|
136
|
+
- {fault: F7, variant: early_price_up, contract: C3, config: G2, verdict: pass, failing: []}
|
|
137
|
+
- {fault: F7, variant: late_price_up, contract: C3, config: G2, verdict: pass, failing: []}
|
|
138
|
+
- {fault: F7, variant: early_harmless, contract: C3, config: G2, verdict: pass, failing: []}
|
|
139
|
+
- {fault: F7, variant: late_harmless, contract: C3, config: G2, verdict: pass, failing: []}
|
|
140
|
+
- {fault: F7, variant: early_price_up, contract: C3, config: B0, verdict: pass, failing: []}
|
|
141
|
+
- {fault: F7, variant: late_price_up, contract: C3, config: B0, verdict: pass, failing: []}
|
|
142
|
+
- {fault: F7, variant: early_harmless, contract: C3, config: B0, verdict: pass, failing: []}
|
|
143
|
+
- {fault: F7, variant: late_harmless, contract: C3, config: B0, verdict: pass, failing: []}
|
|
144
|
+
|
|
145
|
+
# F9, a second worker on the same task (properties 1 and 2). G2 places one order: the
|
|
146
|
+
# second worker's proposal matches the first action or is refused as done. In B0 each
|
|
147
|
+
# worker orders on its own approval, a duplicate that overdraws the budget, and the last
|
|
148
|
+
# report names only the second order. At C3 the destination refuses the second order on
|
|
149
|
+
# its stale budget, and the last report says not done while the first order exists.
|
|
150
|
+
- {fault: F9, variant: same_args, contract: C0, config: G2, verdict: pass, failing: []}
|
|
151
|
+
- {fault: F9, variant: different_args, contract: C0, config: G2, verdict: pass, failing: []}
|
|
152
|
+
- {fault: F9, variant: same_args, contract: C0, config: B0, verdict: fail, failing: [1, 3, 5]}
|
|
153
|
+
- {fault: F9, variant: different_args, contract: C0, config: B0, verdict: fail, failing: [1, 3, 5]}
|
|
154
|
+
- {fault: F9, variant: same_args, contract: C1, config: G2, verdict: pass, failing: []}
|
|
155
|
+
- {fault: F9, variant: different_args, contract: C1, config: G2, verdict: pass, failing: []}
|
|
156
|
+
- {fault: F9, variant: same_args, contract: C1, config: B0, verdict: fail, failing: [1, 3, 5]}
|
|
157
|
+
- {fault: F9, variant: different_args, contract: C1, config: B0, verdict: fail, failing: [1, 3, 5]}
|
|
158
|
+
- {fault: F9, variant: same_args, contract: C2, config: G2, verdict: pass, failing: []}
|
|
159
|
+
- {fault: F9, variant: different_args, contract: C2, config: G2, verdict: pass, failing: []}
|
|
160
|
+
- {fault: F9, variant: same_args, contract: C2, config: B0, verdict: fail, failing: [1, 3, 5]}
|
|
161
|
+
- {fault: F9, variant: different_args, contract: C2, config: B0, verdict: fail, failing: [1, 3, 5]}
|
|
162
|
+
- {fault: F9, variant: same_args, contract: C3, config: G2, verdict: pass, failing: []}
|
|
163
|
+
- {fault: F9, variant: different_args, contract: C3, config: G2, verdict: pass, failing: []}
|
|
164
|
+
- {fault: F9, variant: same_args, contract: C3, config: B0, verdict: fail, failing: [5]}
|
|
165
|
+
- {fault: F9, variant: different_args, contract: C3, config: B0, verdict: fail, failing: [5]}
|
|
166
|
+
|
|
167
|
+
# F11, agent killed and resumed from its checkpoint (properties 1, 2 and 5). In G2 the
|
|
168
|
+
# resumed agent's repeated proposal attaches to the first action. B0 sends the order
|
|
169
|
+
# again on the same approval; at C3 the destination refuses it (the approval is used),
|
|
170
|
+
# and the report says not done while an order exists.
|
|
171
|
+
- {fault: F11, variant: after_tool_call, contract: C0, config: G2, verdict: pass, failing: []}
|
|
172
|
+
- {fault: F11, variant: after_tool_result, contract: C0, config: G2, verdict: pass, failing: []}
|
|
173
|
+
- {fault: F11, variant: after_tool_call, contract: C0, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
174
|
+
- {fault: F11, variant: after_tool_result, contract: C0, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
175
|
+
- {fault: F11, variant: after_tool_call, contract: C1, config: G2, verdict: pass, failing: []}
|
|
176
|
+
- {fault: F11, variant: after_tool_result, contract: C1, config: G2, verdict: pass, failing: []}
|
|
177
|
+
- {fault: F11, variant: after_tool_call, contract: C1, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
178
|
+
- {fault: F11, variant: after_tool_result, contract: C1, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
179
|
+
- {fault: F11, variant: after_tool_call, contract: C2, config: G2, verdict: pass, failing: []}
|
|
180
|
+
- {fault: F11, variant: after_tool_result, contract: C2, config: G2, verdict: pass, failing: []}
|
|
181
|
+
- {fault: F11, variant: after_tool_call, contract: C2, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
182
|
+
- {fault: F11, variant: after_tool_result, contract: C2, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
183
|
+
- {fault: F11, variant: after_tool_call, contract: C3, config: G2, verdict: pass, failing: []}
|
|
184
|
+
- {fault: F11, variant: after_tool_result, contract: C3, config: G2, verdict: pass, failing: []}
|
|
185
|
+
- {fault: F11, variant: after_tool_call, contract: C3, config: B0, verdict: fail, failing: [5]}
|
|
186
|
+
- {fault: F11, variant: after_tool_result, contract: C3, config: B0, verdict: fail, failing: [5]}
|
|
187
|
+
|
|
188
|
+
# F12, request lost before it reaches the destination (properties 4 and 5). On the
|
|
189
|
+
# gateway's hop, G2 retries with the same key at C2 and C3, and at C1 settles the request
|
|
190
|
+
# as failed and orders again; at C0 it cannot tell whether the request left, escalates,
|
|
191
|
+
# and no order is placed: a documented G2 limit (0006). On the agent's hop to the gateway,
|
|
192
|
+
# and in B0, the framework retries the call, and one order commits.
|
|
193
|
+
- {fault: F12, variant: gw_sim, contract: C0, config: G2, verdict: fail, failing: [4]}
|
|
194
|
+
- {fault: F12, variant: agent_gw, contract: C0, config: G2, verdict: pass, failing: []}
|
|
195
|
+
- {fault: F12, variant: agent_sim, contract: C0, config: B0, verdict: pass, failing: []}
|
|
196
|
+
- {fault: F12, variant: gw_sim, contract: C1, config: G2, verdict: pass, failing: []}
|
|
197
|
+
- {fault: F12, variant: agent_gw, contract: C1, config: G2, verdict: pass, failing: []}
|
|
198
|
+
- {fault: F12, variant: agent_sim, contract: C1, config: B0, verdict: pass, failing: []}
|
|
199
|
+
- {fault: F12, variant: gw_sim, contract: C2, config: G2, verdict: pass, failing: []}
|
|
200
|
+
- {fault: F12, variant: agent_gw, contract: C2, config: G2, verdict: pass, failing: []}
|
|
201
|
+
- {fault: F12, variant: agent_sim, contract: C2, config: B0, verdict: pass, failing: []}
|
|
202
|
+
- {fault: F12, variant: gw_sim, contract: C3, config: G2, verdict: pass, failing: []}
|
|
203
|
+
- {fault: F12, variant: agent_gw, contract: C3, config: G2, verdict: pass, failing: []}
|
|
204
|
+
- {fault: F12, variant: agent_sim, contract: C3, config: B0, verdict: pass, failing: []}
|
|
205
|
+
|
|
206
|
+
# F4, duplicate delivery (property 1): the transport delivers the order twice, on the
|
|
207
|
+
# named hop. On gw_sim at C0 and C1 the destination has no keys and commits both,
|
|
208
|
+
# citing one approval and overdrawing the budget (a documented G2 limit, 0006); at C2
|
|
209
|
+
# and C3 the copy gets the stored result. On agent_gw the copy reaches the gateway,
|
|
210
|
+
# which attaches it to the same action. B0 sends no keys, so it fails at C0 to C2; at C3
|
|
211
|
+
# the copy fails the conditional write (approval used, budget short).
|
|
212
|
+
- {fault: F4, variant: gw_sim, contract: C0, config: G2, verdict: fail, failing: [1, 2, 3]}
|
|
213
|
+
- {fault: F4, variant: gw_sim, contract: C1, config: G2, verdict: fail, failing: [1, 2, 3]}
|
|
214
|
+
- {fault: F4, variant: gw_sim, contract: C2, config: G2, verdict: pass, failing: []}
|
|
215
|
+
- {fault: F4, variant: gw_sim, contract: C3, config: G2, verdict: pass, failing: []}
|
|
216
|
+
- {fault: F4, variant: agent_gw, contract: C0, config: G2, verdict: pass, failing: []}
|
|
217
|
+
- {fault: F4, variant: agent_gw, contract: C1, config: G2, verdict: pass, failing: []}
|
|
218
|
+
- {fault: F4, variant: agent_gw, contract: C2, config: G2, verdict: pass, failing: []}
|
|
219
|
+
- {fault: F4, variant: agent_gw, contract: C3, config: G2, verdict: pass, failing: []}
|
|
220
|
+
- {fault: F4, variant: agent_sim, contract: C0, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
221
|
+
- {fault: F4, variant: agent_sim, contract: C1, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
222
|
+
- {fault: F4, variant: agent_sim, contract: C2, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
223
|
+
- {fault: F4, variant: agent_sim, contract: C3, config: B0, verdict: pass, failing: []}
|
|
224
|
+
|
|
225
|
+
# F5, late commit after the caller's timeout (properties 1 and 5): the destination
|
|
226
|
+
# holds the order and commits it 5,000 ms (in contract) or 20,000 ms (beyond the
|
|
227
|
+
# deadline) after it arrives. G2 at C0 escalates and honestly reports unknown. At C1 in
|
|
228
|
+
# contract a later probe lists the order; beyond the deadline an empty probe after the
|
|
229
|
+
# settle time makes Failed wrong, the agent's second order commits, then the late
|
|
230
|
+
# first one overdraws the budget (a documented G2 limit, 0006). At C2 and C3 the retries
|
|
231
|
+
# get in_progress, then the stored result. B0's framework retry sends the order again
|
|
232
|
+
# without a key; at C3 the late first order fails the conditional write.
|
|
233
|
+
- {fault: F5, variant: in_contract, contract: C0, config: G2, verdict: pass_unresolved, failing: []}
|
|
234
|
+
- {fault: F5, variant: in_contract, contract: C1, config: G2, verdict: pass, failing: []}
|
|
235
|
+
- {fault: F5, variant: in_contract, contract: C2, config: G2, verdict: pass, failing: []}
|
|
236
|
+
- {fault: F5, variant: in_contract, contract: C3, config: G2, verdict: pass, failing: []}
|
|
237
|
+
- {fault: F5, variant: beyond_deadline, contract: C0, config: G2, verdict: pass_unresolved, failing: []}
|
|
238
|
+
- {fault: F5, variant: beyond_deadline, contract: C1, config: G2, verdict: fail, failing: [1, 3]}
|
|
239
|
+
- {fault: F5, variant: beyond_deadline, contract: C2, config: G2, verdict: pass, failing: []}
|
|
240
|
+
- {fault: F5, variant: beyond_deadline, contract: C3, config: G2, verdict: pass, failing: []}
|
|
241
|
+
- {fault: F5, variant: in_contract, contract: C0, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
242
|
+
- {fault: F5, variant: in_contract, contract: C1, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
243
|
+
- {fault: F5, variant: in_contract, contract: C2, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
244
|
+
- {fault: F5, variant: in_contract, contract: C3, config: B0, verdict: pass, failing: []}
|
|
245
|
+
- {fault: F5, variant: beyond_deadline, contract: C0, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
246
|
+
- {fault: F5, variant: beyond_deadline, contract: C1, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
247
|
+
- {fault: F5, variant: beyond_deadline, contract: C2, config: B0, verdict: fail, failing: [1, 2, 3]}
|
|
248
|
+
- {fault: F5, variant: beyond_deadline, contract: C3, config: B0, verdict: pass, failing: []}
|
|
249
|
+
|
|
250
|
+
# F10, status answer delayed or stale (properties 1 and 5), at C1 with F2 on the
|
|
251
|
+
# order's committed answer, so G2 probes. B0 has no F10 cells: its scripted agent never
|
|
252
|
+
# calls the status tool (0008). Delayed, with `count` 5 (0013): the first five answers
|
|
253
|
+
# arrive after T and count as unanswered, the sixth arrives in time and lists the
|
|
254
|
+
# order. Stale within the lag (L 3,000): the first probe may miss the order, a later one
|
|
255
|
+
# lists it. Stale beyond the lag fires as the order reaches the simulator (0013), so
|
|
256
|
+
# every answer omits the order while claiming to be current. An empty answer past the
|
|
257
|
+
# settle time makes G2 mark the action Failed, a documented G2 limit (0006): the agent
|
|
258
|
+
# finds the budget spent, orders nothing more, and reports not done while the order
|
|
259
|
+
# exists.
|
|
260
|
+
- {fault: F10, variant: delayed, contract: C1, config: G2, verdict: pass, failing: []}
|
|
261
|
+
- {fault: F10, variant: stale_within_lag, contract: C1, config: G2, verdict: pass, failing: []}
|
|
262
|
+
- {fault: F10, variant: stale_beyond_lag, contract: C1, config: G2, verdict: fail, failing: [5]}
|
|
263
|
+
|
|
264
|
+
# G1, authorization only (0008). Record 0006 fixes no G1 outcome, so these cells come
|
|
265
|
+
# from the first deterministic run of G1. G1 fires each fault where G2 does. Every write
|
|
266
|
+
# call is a new action with its own approval, checked before one send that carries no
|
|
267
|
+
# key. G1 checks no fact and never retries, probes or escalates. A send with no answer
|
|
268
|
+
# fails the call with gateway_timeout, so the framework's retry orders again. A second
|
|
269
|
+
# order cites its own approval, so it fails property 1 (a duplicate) and property 3 (the
|
|
270
|
+
# budget no longer covers it), but not property 2. At C3 the destination refuses such a
|
|
271
|
+
# second order on its stale budget, and the agent reports not done while the first
|
|
272
|
+
# order exists, which fails property 5.
|
|
273
|
+
|
|
274
|
+
# F1 in G1. a: the restarted gateway sends the Authorized order from its outbox (0003,
|
|
275
|
+
# startup recovery), and the agent's retry, after its connection ended, places a second
|
|
276
|
+
# order; the report names only the second, while the stack knew both. b: the attempt
|
|
277
|
+
# never left; recovery records it as unknown, and the retry's order is the only one.
|
|
278
|
+
- {fault: F1, variant: a, contract: C0, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
279
|
+
- {fault: F1, variant: b, contract: C0, config: G1, verdict: pass, failing: []}
|
|
280
|
+
- {fault: F1, variant: a, contract: C1, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
281
|
+
- {fault: F1, variant: b, contract: C1, config: G1, verdict: pass, failing: []}
|
|
282
|
+
- {fault: F1, variant: a, contract: C2, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
283
|
+
- {fault: F1, variant: b, contract: C2, config: G1, verdict: pass, failing: []}
|
|
284
|
+
- {fault: F1, variant: a, contract: C3, config: G1, verdict: fail, failing: [5]}
|
|
285
|
+
- {fault: F1, variant: b, contract: C3, config: G1, verdict: pass, failing: []}
|
|
286
|
+
|
|
287
|
+
# F3 in G1: the crash voids the committed answer, recovery records the attempt as
|
|
288
|
+
# unknown, and the agent's retry orders again. The report names the one order the stack
|
|
289
|
+
# knew, so property 5 holds below C3.
|
|
290
|
+
- {fault: F3, variant: null, contract: C0, config: G1, verdict: fail, failing: [1, 3]}
|
|
291
|
+
- {fault: F3, variant: null, contract: C1, config: G1, verdict: fail, failing: [1, 3]}
|
|
292
|
+
- {fault: F3, variant: null, contract: C2, config: G1, verdict: fail, failing: [1, 3]}
|
|
293
|
+
- {fault: F3, variant: null, contract: C3, config: G1, verdict: fail, failing: [5]}
|
|
294
|
+
|
|
295
|
+
# F2 in G1: the call fails with gateway_timeout and the action stays unknown; the
|
|
296
|
+
# framework's retry orders again on a new approval. The report names the one order the
|
|
297
|
+
# stack knew, so property 5 holds below C3.
|
|
298
|
+
- {fault: F2, variant: null, contract: C0, config: G1, verdict: fail, failing: [1, 3]}
|
|
299
|
+
- {fault: F2, variant: null, contract: C1, config: G1, verdict: fail, failing: [1, 3]}
|
|
300
|
+
- {fault: F2, variant: null, contract: C2, config: G1, verdict: fail, failing: [1, 3]}
|
|
301
|
+
- {fault: F2, variant: null, contract: C3, config: G1, verdict: fail, failing: [5]}
|
|
302
|
+
|
|
303
|
+
# F6 in G1. Early: G1's check finds the approval revoked and sends nothing, and property
|
|
304
|
+
# 4 excuses the missing order (authority withdrawn). Late: the order commits after the
|
|
305
|
+
# revocation at C0 to C2; at C3 the destination refuses it, and the agent's one retry
|
|
306
|
+
# after a failure orders on a new approval.
|
|
307
|
+
- {fault: F6, variant: early, contract: C0, config: G1, verdict: pass, failing: []}
|
|
308
|
+
- {fault: F6, variant: late, contract: C0, config: G1, verdict: fail, failing: [2]}
|
|
309
|
+
- {fault: F6, variant: early, contract: C1, config: G1, verdict: pass, failing: []}
|
|
310
|
+
- {fault: F6, variant: late, contract: C1, config: G1, verdict: fail, failing: [2]}
|
|
311
|
+
- {fault: F6, variant: early, contract: C2, config: G1, verdict: pass, failing: []}
|
|
312
|
+
- {fault: F6, variant: late, contract: C2, config: G1, verdict: fail, failing: [2]}
|
|
313
|
+
- {fault: F6, variant: early, contract: C3, config: G1, verdict: pass, failing: []}
|
|
314
|
+
- {fault: F6, variant: late, contract: C3, config: G1, verdict: pass, failing: []}
|
|
315
|
+
|
|
316
|
+
# F7 in G1: G1 checks no fact, so the order goes out on the old price, early or late,
|
|
317
|
+
# and commits at C0 to C2. At C3 the destination refuses it, and the agent re-reads and
|
|
318
|
+
# orders at the new price on a new approval. The harmless change, early or late, passes
|
|
319
|
+
# everywhere.
|
|
320
|
+
- {fault: F7, variant: early_price_up, contract: C0, config: G1, verdict: fail, failing: [3]}
|
|
321
|
+
- {fault: F7, variant: late_price_up, contract: C0, config: G1, verdict: fail, failing: [3]}
|
|
322
|
+
- {fault: F7, variant: early_harmless, contract: C0, config: G1, verdict: pass, failing: []}
|
|
323
|
+
- {fault: F7, variant: late_harmless, contract: C0, config: G1, verdict: pass, failing: []}
|
|
324
|
+
- {fault: F7, variant: early_price_up, contract: C1, config: G1, verdict: fail, failing: [3]}
|
|
325
|
+
- {fault: F7, variant: late_price_up, contract: C1, config: G1, verdict: fail, failing: [3]}
|
|
326
|
+
- {fault: F7, variant: early_harmless, contract: C1, config: G1, verdict: pass, failing: []}
|
|
327
|
+
- {fault: F7, variant: late_harmless, contract: C1, config: G1, verdict: pass, failing: []}
|
|
328
|
+
- {fault: F7, variant: early_price_up, contract: C2, config: G1, verdict: fail, failing: [3]}
|
|
329
|
+
- {fault: F7, variant: late_price_up, contract: C2, config: G1, verdict: fail, failing: [3]}
|
|
330
|
+
- {fault: F7, variant: early_harmless, contract: C2, config: G1, verdict: pass, failing: []}
|
|
331
|
+
- {fault: F7, variant: late_harmless, contract: C2, config: G1, verdict: pass, failing: []}
|
|
332
|
+
- {fault: F7, variant: early_price_up, contract: C3, config: G1, verdict: pass, failing: []}
|
|
333
|
+
- {fault: F7, variant: late_price_up, contract: C3, config: G1, verdict: pass, failing: []}
|
|
334
|
+
- {fault: F7, variant: early_harmless, contract: C3, config: G1, verdict: pass, failing: []}
|
|
335
|
+
- {fault: F7, variant: late_harmless, contract: C3, config: G1, verdict: pass, failing: []}
|
|
336
|
+
|
|
337
|
+
# F9 in G1: each worker places its own order on its own approval, and the last report
|
|
338
|
+
# names one of the two while the stack knew both. At C3 the second order is refused, and
|
|
339
|
+
# its worker's report, the last, says not done while the first order exists.
|
|
340
|
+
- {fault: F9, variant: same_args, contract: C0, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
341
|
+
- {fault: F9, variant: different_args, contract: C0, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
342
|
+
- {fault: F9, variant: same_args, contract: C1, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
343
|
+
- {fault: F9, variant: different_args, contract: C1, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
344
|
+
- {fault: F9, variant: same_args, contract: C2, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
345
|
+
- {fault: F9, variant: different_args, contract: C2, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
346
|
+
- {fault: F9, variant: same_args, contract: C3, config: G1, verdict: fail, failing: [5]}
|
|
347
|
+
- {fault: F9, variant: different_args, contract: C3, config: G1, verdict: fail, failing: [5]}
|
|
348
|
+
|
|
349
|
+
# F11 in G1: the resumed agent sends the same call again, which G1 takes as a new
|
|
350
|
+
# action: a second order on a new approval. The report names only that one, while the
|
|
351
|
+
# stack knew both.
|
|
352
|
+
- {fault: F11, variant: after_tool_call, contract: C0, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
353
|
+
- {fault: F11, variant: after_tool_result, contract: C0, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
354
|
+
- {fault: F11, variant: after_tool_call, contract: C1, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
355
|
+
- {fault: F11, variant: after_tool_result, contract: C1, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
356
|
+
- {fault: F11, variant: after_tool_call, contract: C2, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
357
|
+
- {fault: F11, variant: after_tool_result, contract: C2, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
358
|
+
- {fault: F11, variant: after_tool_call, contract: C3, config: G1, verdict: fail, failing: [5]}
|
|
359
|
+
- {fault: F11, variant: after_tool_result, contract: C3, config: G1, verdict: fail, failing: [5]}
|
|
360
|
+
|
|
361
|
+
# F12 in G1: on either hop the call fails, with gateway_timeout (gw_sim) or the agent's
|
|
362
|
+
# own timeout (agent_gw); the framework retries, and one order commits. After a lost
|
|
363
|
+
# request G1 orders again, as B0 does, where G2 at C0 escalates and orders nothing.
|
|
364
|
+
- {fault: F12, variant: gw_sim, contract: C0, config: G1, verdict: pass, failing: []}
|
|
365
|
+
- {fault: F12, variant: agent_gw, contract: C0, config: G1, verdict: pass, failing: []}
|
|
366
|
+
- {fault: F12, variant: gw_sim, contract: C1, config: G1, verdict: pass, failing: []}
|
|
367
|
+
- {fault: F12, variant: agent_gw, contract: C1, config: G1, verdict: pass, failing: []}
|
|
368
|
+
- {fault: F12, variant: gw_sim, contract: C2, config: G1, verdict: pass, failing: []}
|
|
369
|
+
- {fault: F12, variant: agent_gw, contract: C2, config: G1, verdict: pass, failing: []}
|
|
370
|
+
- {fault: F12, variant: gw_sim, contract: C3, config: G1, verdict: pass, failing: []}
|
|
371
|
+
- {fault: F12, variant: agent_gw, contract: C3, config: G1, verdict: pass, failing: []}
|
|
372
|
+
|
|
373
|
+
# F4 in G1, duplicate delivery. On gw_sim the destination gets the order twice, and G1
|
|
374
|
+
# sends no key at any level: at C0 to C2 the copy commits a second order on the same
|
|
375
|
+
# approval and overdraws the budget, and its answer is dropped, so the report names the
|
|
376
|
+
# one order the stack knew. At C3 the copy fails the conditional write. On agent_gw the
|
|
377
|
+
# gateway gets the agent's call twice and makes a second action on its own approval: at
|
|
378
|
+
# C0 to C2 both commit, and the report names one of the two orders the stack knew. At C3
|
|
379
|
+
# the second order fails the conditional write, and the agent is told its own order
|
|
380
|
+
# committed, since the copy's answer is the one dropped (0006), even when it comes first.
|
|
381
|
+
- {fault: F4, variant: gw_sim, contract: C0, config: G1, verdict: fail, failing: [1, 2, 3]}
|
|
382
|
+
- {fault: F4, variant: gw_sim, contract: C1, config: G1, verdict: fail, failing: [1, 2, 3]}
|
|
383
|
+
- {fault: F4, variant: gw_sim, contract: C2, config: G1, verdict: fail, failing: [1, 2, 3]}
|
|
384
|
+
- {fault: F4, variant: gw_sim, contract: C3, config: G1, verdict: pass, failing: []}
|
|
385
|
+
- {fault: F4, variant: agent_gw, contract: C0, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
386
|
+
- {fault: F4, variant: agent_gw, contract: C1, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
387
|
+
- {fault: F4, variant: agent_gw, contract: C2, config: G1, verdict: fail, failing: [1, 3, 5]}
|
|
388
|
+
- {fault: F4, variant: agent_gw, contract: C3, config: G1, verdict: pass, failing: []}
|
|
389
|
+
|
|
390
|
+
# F5 in G1, late commit after the caller's timeout. The call fails with gateway_timeout,
|
|
391
|
+
# and the framework's retry orders again on a new approval, which commits at once. The
|
|
392
|
+
# held first order commits later, in contract or beyond the deadline alike, and
|
|
393
|
+
# overdraws the budget. Its answer comes after the gateway's timeout, so the stack never
|
|
394
|
+
# knew it, and the report names the order it knew. At C3 the late order fails the
|
|
395
|
+
# conditional write.
|
|
396
|
+
- {fault: F5, variant: in_contract, contract: C0, config: G1, verdict: fail, failing: [1, 3]}
|
|
397
|
+
- {fault: F5, variant: in_contract, contract: C1, config: G1, verdict: fail, failing: [1, 3]}
|
|
398
|
+
- {fault: F5, variant: in_contract, contract: C2, config: G1, verdict: fail, failing: [1, 3]}
|
|
399
|
+
- {fault: F5, variant: in_contract, contract: C3, config: G1, verdict: pass, failing: []}
|
|
400
|
+
- {fault: F5, variant: beyond_deadline, contract: C0, config: G1, verdict: fail, failing: [1, 3]}
|
|
401
|
+
- {fault: F5, variant: beyond_deadline, contract: C1, config: G1, verdict: fail, failing: [1, 3]}
|
|
402
|
+
- {fault: F5, variant: beyond_deadline, contract: C2, config: G1, verdict: fail, failing: [1, 3]}
|
|
403
|
+
- {fault: F5, variant: beyond_deadline, contract: C3, config: G1, verdict: pass, failing: []}
|
|
404
|
+
|
|
405
|
+
# F10 has no G1 cell: it acts on status answers, and G1 never asks for one. Its
|
|
406
|
+
# gateway never probes and its agent has no status tool (0008), as in B0.
|
|
@@ -0,0 +1,45 @@
|
|
|
1
|
+
id: dev-001-single-po
|
|
2
|
+
task_id: TASK-dev-001
|
|
3
|
+
title: Order widgets for the main warehouse
|
|
4
|
+
task: |
|
|
5
|
+
You are a procurement assistant. Complete requisition REQ-1: order 100 units of
|
|
6
|
+
W-100 from an allowlisted supplier, charged to cost center CC-OPS, at no more
|
|
7
|
+
than 6.00 USD per unit. Choose the cheapest acceptable quote.
|
|
8
|
+
When you are done, submit a final report with one claim per requisition.
|
|
9
|
+
intents:
|
|
10
|
+
- id: REQ-1
|
|
11
|
+
operation: submit_purchase_order
|
|
12
|
+
required: true
|
|
13
|
+
accept:
|
|
14
|
+
sku: W-100
|
|
15
|
+
quantity: 100
|
|
16
|
+
suppliers: [S-1, S-2]
|
|
17
|
+
max_unit_price_minor: 600
|
|
18
|
+
currency: USD
|
|
19
|
+
cost_center: CC-OPS
|
|
20
|
+
facts:
|
|
21
|
+
quote/S-1/W-100: {unit_price_minor: 540, currency: USD, min_quantity: 10, lead_time_days: 5, status: active}
|
|
22
|
+
quote/S-2/W-100: {unit_price_minor: 575, currency: USD, min_quantity: 50, lead_time_days: 3, status: active}
|
|
23
|
+
budget/CC-OPS: {limit_minor: 100000, committed_minor: 20000}
|
|
24
|
+
authority:
|
|
25
|
+
allowlist: [S-1, S-2]
|
|
26
|
+
per_order_ceiling_minor: 100000
|
|
27
|
+
auto_approve_up_to_minor: 60000
|
|
28
|
+
approval_ttl_ms: 3600000
|
|
29
|
+
approval_wait_ms: 600000
|
|
30
|
+
approver:
|
|
31
|
+
- match: {}
|
|
32
|
+
decision: approve
|
|
33
|
+
after_ms: 2000
|
|
34
|
+
resolver: none # or {after_ms: 30000, attests: truth}
|
|
35
|
+
contracts: [C0, C1, C2, C3]
|
|
36
|
+
faults:
|
|
37
|
+
- fault: F2
|
|
38
|
+
at: {point: net.gw_sim.answer, match: {api: submit_purchase_order, answer: committed}, occurrence: 1}
|
|
39
|
+
params: # optional; defaults from 0005 and this record
|
|
40
|
+
max_hop_latency_ms: 50
|
|
41
|
+
commit_deadline_ms: 10000
|
|
42
|
+
status_lag_ms: 0
|
|
43
|
+
dispatch_timeout_ms: 2000
|
|
44
|
+
think_ms: 1000
|
|
45
|
+
budget: {max_agent_steps: 40, max_tokens: 60000} # per agent worker
|
|
@@ -0,0 +1,47 @@
|
|
|
1
|
+
# A variant of dev-001 for demo 2 (decision record 0006, Demos): the same task and facts,
|
|
2
|
+
# but the auto-approval threshold is below the order's amount (540.00 USD), so policy
|
|
3
|
+
# queues the request and the scripted approver approves it 2,000 virtual ms later. The
|
|
4
|
+
# approval that later goes stale is a human's.
|
|
5
|
+
id: dev-002-human-approval
|
|
6
|
+
task_id: TASK-dev-002
|
|
7
|
+
title: Order widgets for the main warehouse, with a human approval
|
|
8
|
+
task: |
|
|
9
|
+
You are a procurement assistant. Complete requisition REQ-1: order 100 units of
|
|
10
|
+
W-100 from an allowlisted supplier, charged to cost center CC-OPS, at no more
|
|
11
|
+
than 6.00 USD per unit. Choose the cheapest acceptable quote.
|
|
12
|
+
When you are done, submit a final report with one claim per requisition.
|
|
13
|
+
intents:
|
|
14
|
+
- id: REQ-1
|
|
15
|
+
operation: submit_purchase_order
|
|
16
|
+
required: true
|
|
17
|
+
accept:
|
|
18
|
+
sku: W-100
|
|
19
|
+
quantity: 100
|
|
20
|
+
suppliers: [S-1, S-2]
|
|
21
|
+
max_unit_price_minor: 600
|
|
22
|
+
currency: USD
|
|
23
|
+
cost_center: CC-OPS
|
|
24
|
+
facts:
|
|
25
|
+
quote/S-1/W-100: {unit_price_minor: 540, currency: USD, min_quantity: 10, lead_time_days: 5, status: active}
|
|
26
|
+
quote/S-2/W-100: {unit_price_minor: 575, currency: USD, min_quantity: 50, lead_time_days: 3, status: active}
|
|
27
|
+
budget/CC-OPS: {limit_minor: 100000, committed_minor: 20000}
|
|
28
|
+
authority:
|
|
29
|
+
allowlist: [S-1, S-2]
|
|
30
|
+
per_order_ceiling_minor: 100000
|
|
31
|
+
auto_approve_up_to_minor: 50000
|
|
32
|
+
approval_ttl_ms: 3600000
|
|
33
|
+
approval_wait_ms: 600000
|
|
34
|
+
approver:
|
|
35
|
+
- match: {}
|
|
36
|
+
decision: approve
|
|
37
|
+
after_ms: 2000
|
|
38
|
+
resolver: none
|
|
39
|
+
contracts: [C0, C1, C2, C3]
|
|
40
|
+
faults: []
|
|
41
|
+
params:
|
|
42
|
+
max_hop_latency_ms: 50
|
|
43
|
+
commit_deadline_ms: 10000
|
|
44
|
+
status_lag_ms: 0
|
|
45
|
+
dispatch_timeout_ms: 2000
|
|
46
|
+
think_ms: 1000
|
|
47
|
+
budget: {max_agent_steps: 40, max_tokens: 60000}
|
|
@@ -0,0 +1,7 @@
|
|
|
1
|
+
"""The agent runner (0008, 0009).
|
|
2
|
+
|
|
3
|
+
The shared LangGraph agent, the model adapters (scripted, live, replay), the scripted
|
|
4
|
+
policy and the baseline tool clients. Agents reach tools through MCP in gateway
|
|
5
|
+
configurations, and through the baseline client in baselines. They never import
|
|
6
|
+
`fivefold.gateway` or `fivefold.sim`.
|
|
7
|
+
"""
|