@ictechgy/context-guard 0.4.10 → 0.4.12
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.
- package/CHANGELOG.md +17 -1
- package/README.ko.md +46 -28
- package/README.md +42 -33
- package/docs/benchmark-fixtures/token-savings-12task.evidence.example.jsonl +24 -0
- package/docs/benchmark-workflow-examples.md +3 -0
- package/docs/benchmark-workflows/context-pack-byte-proxy.example.json +278 -137
- package/docs/benchmark-workflows/measured-token-workflow.example.json +279 -138
- package/docs/benchmark-workflows/provider-cache-telemetry.example.json +279 -138
- package/docs/experimental-benchmark-fixtures.md +24 -7
- package/package.json +2 -1
- package/plugins/context-guard/.claude-plugin/plugin.json +1 -1
- package/plugins/context-guard/README.ko.md +14 -11
- package/plugins/context-guard/README.md +15 -14
- package/plugins/context-guard/bin/context-guard +48 -17
- package/plugins/context-guard/bin/context-guard-artifact +342 -33
- package/plugins/context-guard/bin/context-guard-audit +36 -5
- package/plugins/context-guard/bin/context-guard-bench +1675 -44
- package/plugins/context-guard/bin/context-guard-cache-score +347 -35
- package/plugins/context-guard/bin/context-guard-compress +89 -27
- package/plugins/context-guard/bin/context-guard-cost +7 -2
- package/plugins/context-guard/bin/context-guard-experiments +364 -8
- package/plugins/context-guard/bin/context-guard-failed-nudge +6 -2
- package/plugins/context-guard/bin/context-guard-filter +88 -18
- package/plugins/context-guard/bin/context-guard-pack +329 -19
- package/plugins/context-guard/bin/context-guard-read-symbol +27 -0
- package/plugins/context-guard/bin/context-guard-sanitize-output +245 -18
- package/plugins/context-guard/bin/context-guard-setup +21 -5
- package/plugins/context-guard/bin/context-guard-tool-prune +287 -62
- package/plugins/context-guard/bin/context-guard-trim-output +394 -90
- package/plugins/context-guard/brief/README.md +5 -5
- package/plugins/context-guard/lib/context_guard_command_manifest_loader.py +123 -0
- package/plugins/context-guard/lib/context_guard_commands.py +217 -190
|
@@ -1,169 +1,310 @@
|
|
|
1
1
|
{
|
|
2
|
-
"schema": "context-guard-bench-report-v1",
|
|
3
2
|
"baseline_variant": "baseline",
|
|
3
|
+
"caveat": "Proxy byte reductions are reported separately from matched-task token/cost metrics; shifted cost savings require measured primary cost and measured external cost when external tokens are present. Wall time and provider cached-token fields are diagnostic telemetry, not proof of ContextGuard-caused token or cost savings; provider-cache discounts must stay separate from token-reduction claims. Public hosted savings claims must use public_claim_readiness.claim_allowed; unsupported claims are forbidden.",
|
|
4
|
+
"claim_status": "insufficient_paired_data",
|
|
5
|
+
"comparisons": [
|
|
6
|
+
{
|
|
7
|
+
"baseline_corrections_per_successful_task": 0.0,
|
|
8
|
+
"baseline_failure_rate": 0.0,
|
|
9
|
+
"baseline_successful_task_count": 1,
|
|
10
|
+
"baseline_variant": "baseline",
|
|
11
|
+
"corrections_delta_per_successful_task": 0.0,
|
|
12
|
+
"cost_savings_pct_with_shift": null,
|
|
13
|
+
"failure_rate_delta_pp": 0.0,
|
|
14
|
+
"matched_successful_task_count": 1,
|
|
15
|
+
"missing_baseline_success_tasks": [],
|
|
16
|
+
"paired_corrections_task_count": 1,
|
|
17
|
+
"paired_cost_task_count": 0,
|
|
18
|
+
"paired_token_task_count": 0,
|
|
19
|
+
"paired_wall_time_task_count": 1,
|
|
20
|
+
"quality_gate": "pass",
|
|
21
|
+
"token_savings_pct": null,
|
|
22
|
+
"variant": "context_pack_auto",
|
|
23
|
+
"variant_corrections_per_successful_task": 0.0,
|
|
24
|
+
"variant_failure_rate": 0.0,
|
|
25
|
+
"wall_time_change_pct": -8.333333333333332,
|
|
26
|
+
"wall_time_delta_seconds_per_successful_task": -1.0
|
|
27
|
+
}
|
|
28
|
+
],
|
|
29
|
+
"public_claim_readiness": {
|
|
30
|
+
"blocking_gate_ids": [
|
|
31
|
+
"matched_successful_tasks",
|
|
32
|
+
"provider_measured_token_cost",
|
|
33
|
+
"shifted_cost_accounting",
|
|
34
|
+
"confidence_failure_notes",
|
|
35
|
+
"provider_export_provenance"
|
|
36
|
+
],
|
|
37
|
+
"claim_allowed": false,
|
|
38
|
+
"claim_boundary": {
|
|
39
|
+
"claim_allowed_field": "public_claim_readiness.claim_allowed",
|
|
40
|
+
"fixed_percent_savings_claim_without_matched_provider_report_forbidden": true,
|
|
41
|
+
"hosted_api_cost_savings_claim_without_claim_allowed_forbidden": true,
|
|
42
|
+
"hosted_api_token_savings_claim_without_claim_allowed_forbidden": true,
|
|
43
|
+
"id": "public_claim_readiness_authoritative_release_gate",
|
|
44
|
+
"reason": "Public hosted token/cost savings claims are forbidden unless every readiness gate passes and public_claim_readiness.claim_allowed is true.",
|
|
45
|
+
"reporting_only": true,
|
|
46
|
+
"requires_confidence_and_failure_notes": true,
|
|
47
|
+
"requires_matched_successful_tasks": true,
|
|
48
|
+
"requires_provider_export_provenance": true,
|
|
49
|
+
"requires_provider_measured_tokens_and_cost": true,
|
|
50
|
+
"requires_quality_non_inferiority": true,
|
|
51
|
+
"requires_shifted_cost_accounting": true,
|
|
52
|
+
"unsupported_claims_forbidden": true
|
|
53
|
+
},
|
|
54
|
+
"gates": [
|
|
55
|
+
{
|
|
56
|
+
"evidence": {
|
|
57
|
+
"comparison_count": 1,
|
|
58
|
+
"matched_pair_count": 0,
|
|
59
|
+
"min_matched_successful_task_count": 1.0,
|
|
60
|
+
"missing_baseline_success_task_count": 0,
|
|
61
|
+
"variants": [
|
|
62
|
+
"context_pack_auto"
|
|
63
|
+
]
|
|
64
|
+
},
|
|
65
|
+
"id": "matched_successful_tasks",
|
|
66
|
+
"label": "Matched successful tasks",
|
|
67
|
+
"passed": false,
|
|
68
|
+
"reason": "missing_or_regressed_matched_successful_tasks",
|
|
69
|
+
"required": true,
|
|
70
|
+
"status": "fail"
|
|
71
|
+
},
|
|
72
|
+
{
|
|
73
|
+
"evidence": {
|
|
74
|
+
"matched_pair_count": 0,
|
|
75
|
+
"required_fields": [
|
|
76
|
+
"matched_pair_evidence[*].measurements.baseline.primary_tokens.measured",
|
|
77
|
+
"matched_pair_evidence[*].measurements.variant.primary_tokens.measured",
|
|
78
|
+
"matched_pair_evidence[*].measurements.baseline.primary_cost_usd.measured",
|
|
79
|
+
"matched_pair_evidence[*].measurements.variant.primary_cost_usd.measured"
|
|
80
|
+
]
|
|
81
|
+
},
|
|
82
|
+
"id": "provider_measured_token_cost",
|
|
83
|
+
"label": "Provider-measured token and primary cost",
|
|
84
|
+
"passed": false,
|
|
85
|
+
"reason": "missing_provider_measured_primary_tokens_or_cost",
|
|
86
|
+
"required": true,
|
|
87
|
+
"status": "fail"
|
|
88
|
+
},
|
|
89
|
+
{
|
|
90
|
+
"evidence": {
|
|
91
|
+
"max_corrections_delta_per_successful_task": 0.0,
|
|
92
|
+
"max_failure_rate_delta_pp": 0.0,
|
|
93
|
+
"quality_gates": [
|
|
94
|
+
"pass"
|
|
95
|
+
]
|
|
96
|
+
},
|
|
97
|
+
"id": "quality_non_inferiority",
|
|
98
|
+
"label": "Quality non-inferiority",
|
|
99
|
+
"passed": true,
|
|
100
|
+
"reason": "all_quality_gates_pass",
|
|
101
|
+
"required": true,
|
|
102
|
+
"status": "pass"
|
|
103
|
+
},
|
|
104
|
+
{
|
|
105
|
+
"evidence": {
|
|
106
|
+
"matched_pair_count": 0,
|
|
107
|
+
"required_fields": [
|
|
108
|
+
"matched_pair_evidence[*].claim_boundary.shifted_cost_claim_allowed",
|
|
109
|
+
"matched_pair_evidence[*].measurements.baseline.total_cost_with_shift_usd.measured",
|
|
110
|
+
"matched_pair_evidence[*].measurements.variant.total_cost_with_shift_usd.measured"
|
|
111
|
+
]
|
|
112
|
+
},
|
|
113
|
+
"id": "shifted_cost_accounting",
|
|
114
|
+
"label": "Shifted-cost accounting",
|
|
115
|
+
"passed": false,
|
|
116
|
+
"reason": "missing_shifted_cost_claim_accounting",
|
|
117
|
+
"required": true,
|
|
118
|
+
"status": "fail"
|
|
119
|
+
},
|
|
120
|
+
{
|
|
121
|
+
"evidence": {
|
|
122
|
+
"comparison_failure_fields_present": true,
|
|
123
|
+
"explicit_note_count": 0,
|
|
124
|
+
"failed_row_count": 0,
|
|
125
|
+
"failed_rows_with_notes": 0,
|
|
126
|
+
"replay_row_count": 0
|
|
127
|
+
},
|
|
128
|
+
"id": "confidence_failure_notes",
|
|
129
|
+
"label": "Confidence and failure notes",
|
|
130
|
+
"passed": false,
|
|
131
|
+
"reason": "missing_explicit_replay_notes_or_failure_evidence",
|
|
132
|
+
"required": true,
|
|
133
|
+
"status": "unknown"
|
|
134
|
+
},
|
|
135
|
+
{
|
|
136
|
+
"evidence": {
|
|
137
|
+
"mixed_csv": false,
|
|
138
|
+
"provider_names": [],
|
|
139
|
+
"replay_row_count": 0,
|
|
140
|
+
"report_row_count": 2,
|
|
141
|
+
"same_run_complete": false,
|
|
142
|
+
"source_types": []
|
|
143
|
+
},
|
|
144
|
+
"id": "provider_export_provenance",
|
|
145
|
+
"label": "Provider-export provenance",
|
|
146
|
+
"passed": false,
|
|
147
|
+
"reason": "missing_or_mixed_provider_export_provenance",
|
|
148
|
+
"required": true,
|
|
149
|
+
"status": "unknown"
|
|
150
|
+
}
|
|
151
|
+
],
|
|
152
|
+
"generated_from": "matched_pair_evidence_and_replay_provenance",
|
|
153
|
+
"passed_required_gate_count": 1,
|
|
154
|
+
"public_claim_eligible_observed": null,
|
|
155
|
+
"public_claim_status_observed": "csv_provenance_unknown_requires_original_evidence_or_trusted_ledger",
|
|
156
|
+
"raw_metric_claim_status_observed": "insufficient_paired_data",
|
|
157
|
+
"reason": "replay_evidence_required_for_public_claim",
|
|
158
|
+
"required_gate_count": 6,
|
|
159
|
+
"required_gate_ids": [
|
|
160
|
+
"matched_successful_tasks",
|
|
161
|
+
"provider_measured_token_cost",
|
|
162
|
+
"quality_non_inferiority",
|
|
163
|
+
"shifted_cost_accounting",
|
|
164
|
+
"confidence_failure_notes",
|
|
165
|
+
"provider_export_provenance"
|
|
166
|
+
],
|
|
167
|
+
"schema_version": "contextguard.bench.public-claim-readiness.v1",
|
|
168
|
+
"status": "csv_provenance_unknown_requires_original_evidence_or_trusted_ledger"
|
|
169
|
+
},
|
|
4
170
|
"row_count": 2,
|
|
171
|
+
"schema": "context-guard-bench-report-v1",
|
|
5
172
|
"summary_by_variant": {
|
|
6
173
|
"baseline": {
|
|
7
|
-
"
|
|
8
|
-
"
|
|
174
|
+
"artifacts_used_per_successful_task": 0.0,
|
|
175
|
+
"artifacts_used_successful": 0,
|
|
176
|
+
"byte_reduction_ratio": 1.0,
|
|
177
|
+
"byte_savings_pct": 0.0,
|
|
178
|
+
"bytes_after_successful": 24000,
|
|
179
|
+
"bytes_before_successful": 24000,
|
|
180
|
+
"bytes_saved_per_successful_task": 0.0,
|
|
181
|
+
"bytes_saved_successful": 0,
|
|
182
|
+
"compression_strategy": "baseline",
|
|
183
|
+
"corrections_per_successful_task": 0.0,
|
|
184
|
+
"corrections_successful": 0,
|
|
185
|
+
"external_cost_successful_usd": 0.0,
|
|
186
|
+
"external_cost_unknown_successful": 1,
|
|
187
|
+
"external_tokens_measured_successful": 0,
|
|
188
|
+
"external_tokens_per_successful_task": null,
|
|
189
|
+
"external_tokens_successful": 0,
|
|
9
190
|
"failed_runs": 0,
|
|
10
|
-
"
|
|
11
|
-
"
|
|
191
|
+
"failure_rate": 0.0,
|
|
192
|
+
"hook_triggers_successful": 0,
|
|
193
|
+
"is_baseline_strategy": true,
|
|
194
|
+
"observed_telemetry": {
|
|
195
|
+
"byte_savings": "observed",
|
|
196
|
+
"external_tokens": "unavailable",
|
|
197
|
+
"primary_cost": "unavailable",
|
|
198
|
+
"provider_cache": "unavailable",
|
|
199
|
+
"token_proxy": "inferred",
|
|
200
|
+
"tokens": "unavailable",
|
|
201
|
+
"wall_time": "observed"
|
|
202
|
+
},
|
|
12
203
|
"primary_cost_all_runs_usd": 0.0,
|
|
13
204
|
"primary_cost_measured_runs": 0,
|
|
14
|
-
"
|
|
15
|
-
"
|
|
205
|
+
"primary_cost_measured_successful": 0,
|
|
206
|
+
"primary_cost_per_successful_task_usd": null,
|
|
207
|
+
"primary_cost_per_task_including_failures_usd": null,
|
|
208
|
+
"primary_cost_successful_usd": 0.0,
|
|
209
|
+
"primary_tokens_measured_runs": 0,
|
|
210
|
+
"primary_tokens_measured_successful": 0,
|
|
16
211
|
"provider_cached_tokens_all_runs": 0,
|
|
17
212
|
"provider_cached_tokens_measured_runs": 0,
|
|
213
|
+
"provider_cached_tokens_measured_successful": 0,
|
|
214
|
+
"provider_cached_tokens_per_successful_task": 0.0,
|
|
215
|
+
"provider_cached_tokens_per_task_including_failures": 0.0,
|
|
216
|
+
"provider_cached_tokens_successful": 0,
|
|
217
|
+
"runs": 1,
|
|
218
|
+
"successful_runs": 1,
|
|
219
|
+
"successful_task_count": 1,
|
|
220
|
+
"task_count": 1,
|
|
221
|
+
"token_proxy_saved_per_successful_task": 0.0,
|
|
222
|
+
"token_proxy_saved_successful": 0,
|
|
223
|
+
"tokens_per_successful_task": null,
|
|
224
|
+
"tokens_per_task_including_failures": null,
|
|
18
225
|
"total_cost_with_shift_all_runs_usd": 0.0,
|
|
19
226
|
"total_cost_with_shift_measured_runs": 0,
|
|
227
|
+
"total_cost_with_shift_measured_successful": 0,
|
|
228
|
+
"total_cost_with_shift_per_successful_task_usd": null,
|
|
229
|
+
"total_cost_with_shift_per_task_including_failures_usd": null,
|
|
230
|
+
"total_cost_with_shift_successful_usd": 0.0,
|
|
231
|
+
"total_tokens_all_runs": 0,
|
|
20
232
|
"total_tokens_successful": 0,
|
|
21
|
-
"
|
|
22
|
-
"
|
|
23
|
-
"
|
|
24
|
-
"wall_time_seconds_successful": 12.0,
|
|
233
|
+
"turns_successful": 0,
|
|
234
|
+
"wall_time_seconds_all_runs": 12.0,
|
|
235
|
+
"wall_time_seconds_measured_runs": 1,
|
|
25
236
|
"wall_time_seconds_measured_successful": 1,
|
|
26
|
-
"
|
|
27
|
-
"
|
|
237
|
+
"wall_time_seconds_per_successful_task": 12.0,
|
|
238
|
+
"wall_time_seconds_per_task_including_failures": 12.0,
|
|
239
|
+
"wall_time_seconds_successful": 12.0
|
|
240
|
+
},
|
|
241
|
+
"context_pack_auto": {
|
|
242
|
+
"artifacts_used_per_successful_task": 0.0,
|
|
243
|
+
"artifacts_used_successful": 0,
|
|
244
|
+
"byte_reduction_ratio": 0.25,
|
|
245
|
+
"byte_savings_pct": 75.0,
|
|
246
|
+
"bytes_after_successful": 6000,
|
|
247
|
+
"bytes_before_successful": 24000,
|
|
248
|
+
"bytes_saved_per_successful_task": 18000.0,
|
|
249
|
+
"bytes_saved_successful": 18000,
|
|
250
|
+
"compression_strategy": "context_pack_auto",
|
|
251
|
+
"corrections_per_successful_task": 0.0,
|
|
252
|
+
"corrections_successful": 0,
|
|
28
253
|
"external_cost_successful_usd": 0.0,
|
|
29
254
|
"external_cost_unknown_successful": 1,
|
|
30
|
-
"total_cost_with_shift_successful_usd": 0.0,
|
|
31
|
-
"total_cost_with_shift_measured_successful": 0,
|
|
32
|
-
"external_tokens_successful": 0,
|
|
33
255
|
"external_tokens_measured_successful": 0,
|
|
34
|
-
"artifacts_used_successful": 0,
|
|
35
|
-
"corrections_successful": 0,
|
|
36
|
-
"bytes_before_successful": 24000,
|
|
37
|
-
"bytes_after_successful": 24000,
|
|
38
|
-
"turns_successful": 0,
|
|
39
|
-
"hook_triggers_successful": 0,
|
|
40
|
-
"failure_rate": 0.0,
|
|
41
|
-
"task_count": 1,
|
|
42
|
-
"successful_task_count": 1,
|
|
43
|
-
"tokens_per_task_including_failures": null,
|
|
44
|
-
"wall_time_seconds_per_task_including_failures": 12.0,
|
|
45
|
-
"provider_cached_tokens_per_task_including_failures": 0.0,
|
|
46
|
-
"primary_cost_per_task_including_failures_usd": null,
|
|
47
|
-
"total_cost_with_shift_per_task_including_failures_usd": null,
|
|
48
|
-
"tokens_per_successful_task": null,
|
|
49
|
-
"wall_time_seconds_per_successful_task": 12.0,
|
|
50
|
-
"provider_cached_tokens_per_successful_task": 0.0,
|
|
51
|
-
"primary_cost_per_successful_task_usd": null,
|
|
52
|
-
"total_cost_with_shift_per_successful_task_usd": null,
|
|
53
256
|
"external_tokens_per_successful_task": null,
|
|
54
|
-
"
|
|
55
|
-
"
|
|
56
|
-
"
|
|
57
|
-
"
|
|
58
|
-
"is_baseline_strategy":
|
|
59
|
-
"bytes_saved_successful": 0,
|
|
60
|
-
"bytes_saved_per_successful_task": 0.0,
|
|
61
|
-
"byte_savings_pct": 0.0,
|
|
62
|
-
"token_proxy_saved_successful": 0,
|
|
63
|
-
"token_proxy_saved_per_successful_task": 0.0,
|
|
257
|
+
"external_tokens_successful": 0,
|
|
258
|
+
"failed_runs": 0,
|
|
259
|
+
"failure_rate": 0.0,
|
|
260
|
+
"hook_triggers_successful": 0,
|
|
261
|
+
"is_baseline_strategy": false,
|
|
64
262
|
"observed_telemetry": {
|
|
65
|
-
"tokens": "unavailable",
|
|
66
|
-
"primary_cost": "unavailable",
|
|
67
|
-
"external_tokens": "unavailable",
|
|
68
263
|
"byte_savings": "observed",
|
|
264
|
+
"external_tokens": "unavailable",
|
|
265
|
+
"primary_cost": "unavailable",
|
|
266
|
+
"provider_cache": "unavailable",
|
|
69
267
|
"token_proxy": "inferred",
|
|
70
|
-
"
|
|
71
|
-
"
|
|
72
|
-
}
|
|
73
|
-
},
|
|
74
|
-
"context_pack_auto": {
|
|
75
|
-
"runs": 1,
|
|
76
|
-
"successful_runs": 1,
|
|
77
|
-
"failed_runs": 0,
|
|
78
|
-
"total_tokens_all_runs": 0,
|
|
79
|
-
"primary_tokens_measured_runs": 0,
|
|
268
|
+
"tokens": "unavailable",
|
|
269
|
+
"wall_time": "observed"
|
|
270
|
+
},
|
|
80
271
|
"primary_cost_all_runs_usd": 0.0,
|
|
81
272
|
"primary_cost_measured_runs": 0,
|
|
82
|
-
"
|
|
83
|
-
"
|
|
273
|
+
"primary_cost_measured_successful": 0,
|
|
274
|
+
"primary_cost_per_successful_task_usd": null,
|
|
275
|
+
"primary_cost_per_task_including_failures_usd": null,
|
|
276
|
+
"primary_cost_successful_usd": 0.0,
|
|
277
|
+
"primary_tokens_measured_runs": 0,
|
|
278
|
+
"primary_tokens_measured_successful": 0,
|
|
84
279
|
"provider_cached_tokens_all_runs": 0,
|
|
85
280
|
"provider_cached_tokens_measured_runs": 0,
|
|
86
|
-
"total_cost_with_shift_all_runs_usd": 0.0,
|
|
87
|
-
"total_cost_with_shift_measured_runs": 0,
|
|
88
|
-
"total_tokens_successful": 0,
|
|
89
|
-
"primary_tokens_measured_successful": 0,
|
|
90
|
-
"primary_cost_successful_usd": 0.0,
|
|
91
|
-
"primary_cost_measured_successful": 0,
|
|
92
|
-
"wall_time_seconds_successful": 11.0,
|
|
93
|
-
"wall_time_seconds_measured_successful": 1,
|
|
94
|
-
"provider_cached_tokens_successful": 0,
|
|
95
281
|
"provider_cached_tokens_measured_successful": 0,
|
|
96
|
-
"
|
|
97
|
-
"
|
|
98
|
-
"
|
|
99
|
-
"
|
|
100
|
-
"
|
|
101
|
-
"external_tokens_measured_successful": 0,
|
|
102
|
-
"artifacts_used_successful": 0,
|
|
103
|
-
"corrections_successful": 0,
|
|
104
|
-
"bytes_before_successful": 24000,
|
|
105
|
-
"bytes_after_successful": 6000,
|
|
106
|
-
"turns_successful": 0,
|
|
107
|
-
"hook_triggers_successful": 0,
|
|
108
|
-
"failure_rate": 0.0,
|
|
109
|
-
"task_count": 1,
|
|
282
|
+
"provider_cached_tokens_per_successful_task": 0.0,
|
|
283
|
+
"provider_cached_tokens_per_task_including_failures": 0.0,
|
|
284
|
+
"provider_cached_tokens_successful": 0,
|
|
285
|
+
"runs": 1,
|
|
286
|
+
"successful_runs": 1,
|
|
110
287
|
"successful_task_count": 1,
|
|
288
|
+
"task_count": 1,
|
|
289
|
+
"token_proxy_saved_per_successful_task": 4500.0,
|
|
290
|
+
"token_proxy_saved_successful": 4500,
|
|
291
|
+
"tokens_per_successful_task": null,
|
|
111
292
|
"tokens_per_task_including_failures": null,
|
|
112
|
-
"
|
|
113
|
-
"
|
|
114
|
-
"
|
|
293
|
+
"total_cost_with_shift_all_runs_usd": 0.0,
|
|
294
|
+
"total_cost_with_shift_measured_runs": 0,
|
|
295
|
+
"total_cost_with_shift_measured_successful": 0,
|
|
296
|
+
"total_cost_with_shift_per_successful_task_usd": null,
|
|
115
297
|
"total_cost_with_shift_per_task_including_failures_usd": null,
|
|
116
|
-
"
|
|
298
|
+
"total_cost_with_shift_successful_usd": 0.0,
|
|
299
|
+
"total_tokens_all_runs": 0,
|
|
300
|
+
"total_tokens_successful": 0,
|
|
301
|
+
"turns_successful": 0,
|
|
302
|
+
"wall_time_seconds_all_runs": 11.0,
|
|
303
|
+
"wall_time_seconds_measured_runs": 1,
|
|
304
|
+
"wall_time_seconds_measured_successful": 1,
|
|
117
305
|
"wall_time_seconds_per_successful_task": 11.0,
|
|
118
|
-
"
|
|
119
|
-
"
|
|
120
|
-
"total_cost_with_shift_per_successful_task_usd": null,
|
|
121
|
-
"external_tokens_per_successful_task": null,
|
|
122
|
-
"artifacts_used_per_successful_task": 0.0,
|
|
123
|
-
"corrections_per_successful_task": 0.0,
|
|
124
|
-
"byte_reduction_ratio": 0.25,
|
|
125
|
-
"compression_strategy": "context_pack_auto",
|
|
126
|
-
"is_baseline_strategy": false,
|
|
127
|
-
"bytes_saved_successful": 18000,
|
|
128
|
-
"bytes_saved_per_successful_task": 18000.0,
|
|
129
|
-
"byte_savings_pct": 75.0,
|
|
130
|
-
"token_proxy_saved_successful": 4500,
|
|
131
|
-
"token_proxy_saved_per_successful_task": 4500.0,
|
|
132
|
-
"observed_telemetry": {
|
|
133
|
-
"tokens": "unavailable",
|
|
134
|
-
"primary_cost": "unavailable",
|
|
135
|
-
"external_tokens": "unavailable",
|
|
136
|
-
"byte_savings": "observed",
|
|
137
|
-
"token_proxy": "inferred",
|
|
138
|
-
"wall_time": "observed",
|
|
139
|
-
"provider_cache": "unavailable"
|
|
140
|
-
}
|
|
141
|
-
}
|
|
142
|
-
},
|
|
143
|
-
"comparisons": [
|
|
144
|
-
{
|
|
145
|
-
"variant": "context_pack_auto",
|
|
146
|
-
"baseline_variant": "baseline",
|
|
147
|
-
"quality_gate": "pass",
|
|
148
|
-
"baseline_failure_rate": 0.0,
|
|
149
|
-
"variant_failure_rate": 0.0,
|
|
150
|
-
"failure_rate_delta_pp": 0.0,
|
|
151
|
-
"matched_successful_task_count": 1,
|
|
152
|
-
"baseline_successful_task_count": 1,
|
|
153
|
-
"missing_baseline_success_tasks": [],
|
|
154
|
-
"baseline_corrections_per_successful_task": 0.0,
|
|
155
|
-
"variant_corrections_per_successful_task": 0.0,
|
|
156
|
-
"paired_corrections_task_count": 1,
|
|
157
|
-
"corrections_delta_per_successful_task": 0.0,
|
|
158
|
-
"token_savings_pct": null,
|
|
159
|
-
"paired_token_task_count": 0,
|
|
160
|
-
"wall_time_delta_seconds_per_successful_task": -1.0,
|
|
161
|
-
"wall_time_change_pct": -8.333333333333332,
|
|
162
|
-
"paired_wall_time_task_count": 1,
|
|
163
|
-
"cost_savings_pct_with_shift": null,
|
|
164
|
-
"paired_cost_task_count": 0
|
|
306
|
+
"wall_time_seconds_per_task_including_failures": 11.0,
|
|
307
|
+
"wall_time_seconds_successful": 11.0
|
|
165
308
|
}
|
|
166
|
-
|
|
167
|
-
"claim_status": "insufficient_paired_data",
|
|
168
|
-
"caveat": "Proxy byte reductions are reported separately from matched-task token/cost metrics; shifted cost savings require measured primary cost and measured external cost when external tokens are present. Wall time and provider cached-token fields are diagnostic telemetry, not proof of ContextGuard-caused token or cost savings; provider-cache discounts must stay separate from token-reduction claims."
|
|
309
|
+
}
|
|
169
310
|
}
|