audrey 1.1.2 → 1.1.3

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.
Files changed (45) hide show
  1. package/CHANGELOG.md +8 -0
  2. package/README.md +1 -1
  3. package/benchmarks/output/adapter-self-test/guardbench-adapter-self-test.json +4 -4
  4. package/benchmarks/output/external/guardbench-external-dry-run.json +1 -1
  5. package/benchmarks/output/external/guardbench-external-evidence.json +1 -1
  6. package/benchmarks/output/guardbench-conformance-card.json +9 -9
  7. package/benchmarks/output/guardbench-raw.json +108 -108
  8. package/benchmarks/output/guardbench-summary.json +174 -174
  9. package/benchmarks/output/leaderboard/guardbench-leaderboard.json +5 -5
  10. package/benchmarks/output/leaderboard/guardbench-leaderboard.md +2 -2
  11. package/benchmarks/output/submission-bundle/guardbench-conformance-card.json +9 -9
  12. package/benchmarks/output/submission-bundle/guardbench-raw.json +108 -108
  13. package/benchmarks/output/submission-bundle/guardbench-summary.json +174 -174
  14. package/benchmarks/output/submission-bundle/submission-manifest.json +10 -10
  15. package/benchmarks/output/submission-bundle/validation-report.json +1 -1
  16. package/benchmarks/output/summary.json +54 -54
  17. package/dist/mcp-server/config.d.ts +1 -1
  18. package/dist/mcp-server/config.js +1 -1
  19. package/docs/paper/07-evaluation.md +6 -6
  20. package/docs/paper/audrey-paper-v1.md +6 -6
  21. package/docs/paper/evidence-ledger.md +1 -1
  22. package/docs/paper/output/arxiv/arxiv-manifest.json +4 -4
  23. package/docs/paper/output/arxiv/main.tex +6 -6
  24. package/docs/paper/output/arxiv-compile-report.json +3 -3
  25. package/docs/paper/output/submission-bundle/README.md +1 -1
  26. package/docs/paper/output/submission-bundle/benchmarks/output/adapter-self-test/guardbench-adapter-self-test.json +4 -4
  27. package/docs/paper/output/submission-bundle/benchmarks/output/external/guardbench-external-dry-run.json +1 -1
  28. package/docs/paper/output/submission-bundle/benchmarks/output/external/guardbench-external-evidence.json +1 -1
  29. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-conformance-card.json +9 -9
  30. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-raw.json +108 -108
  31. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-summary.json +174 -174
  32. package/docs/paper/output/submission-bundle/benchmarks/output/leaderboard/guardbench-leaderboard.json +5 -5
  33. package/docs/paper/output/submission-bundle/benchmarks/output/leaderboard/guardbench-leaderboard.md +2 -2
  34. package/docs/paper/output/submission-bundle/benchmarks/output/submission-bundle/submission-manifest.json +10 -10
  35. package/docs/paper/output/submission-bundle/benchmarks/output/submission-bundle/validation-report.json +1 -1
  36. package/docs/paper/output/submission-bundle/benchmarks/output/summary.json +60 -60
  37. package/docs/paper/output/submission-bundle/docs/paper/07-evaluation.md +6 -6
  38. package/docs/paper/output/submission-bundle/docs/paper/audrey-paper-v1.md +6 -6
  39. package/docs/paper/output/submission-bundle/docs/paper/evidence-ledger.md +1 -1
  40. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/arxiv-manifest.json +4 -4
  41. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/main.tex +6 -6
  42. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv-compile-report.json +3 -3
  43. package/docs/paper/output/submission-bundle/package.json +2 -2
  44. package/docs/paper/output/submission-bundle/paper-submission-manifest.json +28 -28
  45. package/package.json +2 -2
package/benchmarks/output/guardbench-raw.json CHANGED
@@ -1,17 +1,17 @@
1
1
  {
2
2
  "suite": "GuardBench comparative",
3
- "generatedAt": "2026-07-13T14:48:49.312Z",
3
+ "generatedAt": "2026-07-13T15:07:31.178Z",
4
4
  "manifestVersion": "0.2.0",
5
5
  "provenance": {
6
- "generatedAt": "2026-07-13T14:48:49.313Z",
7
- "gitSha": "7111c630be69d3ab96439018f094dc4fac826aca",
6
+ "generatedAt": "2026-07-13T15:07:31.179Z",
7
+ "gitSha": "36842cfb0cad7ca5da864231a7e265229ba56fbd",
8
8
  "gitDirty": false,
9
9
  "node": "v24.18.0",
10
10
  "v8": "13.6.233.17-node.50",
11
11
  "platform": "linux",
12
12
  "arch": "x64",
13
13
  "osRelease": "6.17.0-1018-azure",
14
- "cpuModel": "AMD EPYC 9V74 80-Core Processor",
14
+ "cpuModel": "AMD EPYC 7763 64-Core Processor",
15
15
  "cpuCount": 4,
16
16
  "totalMemoryGb": 15.61,
17
17
  "embeddingProvider": "mock",
@@ -33,12 +33,12 @@
33
33
  "decisionCorrect": true,
34
34
  "riskScore": 0.9,
35
35
  "passed": true,
36
- "latencyMs": 11.267,
36
+ "latencyMs": 12.274,
37
37
  "evidenceCount": 3,
38
38
  "evidenceIds": [
39
- "01KXDZ7CQ6VDPVF6JH0HH1KM2A",
40
- "failure:Bash:2026-07-13T14:48:45.542Z",
41
- "01KXDZ7CQ73T2NW3XX0033D6F7"
39
+ "01KXE09KRM3F7XNRBG7KRBC2PP",
40
+ "failure:Bash:2026-07-13T15:07:26.869Z",
41
+ "01KXE09KRP13YT2CWDQSQDF6KB"
42
42
  ],
43
43
  "recommendedActions": [
44
44
  "Do not repeat the exact failed action until the prior error is understood or the command is changed.",
@@ -61,7 +61,7 @@
61
61
  "decisionCorrect": false,
62
62
  "riskScore": 0,
63
63
  "passed": false,
64
- "latencyMs": 0.036,
64
+ "latencyMs": 0.077,
65
65
  "evidenceCount": 0,
66
66
  "evidenceIds": [],
67
67
  "recommendedActions": [],
@@ -81,10 +81,10 @@
81
81
  "decisionCorrect": false,
82
82
  "riskScore": 0.55,
83
83
  "passed": false,
84
- "latencyMs": 0.225,
84
+ "latencyMs": 0.305,
85
85
  "evidenceCount": 1,
86
86
  "evidenceIds": [
87
- "01KXDZ7CRE1TZGVGS5FYAB55J5"
87
+ "01KXE09KT14FCCQ17VECTYQ62H"
88
88
  ],
89
89
  "recommendedActions": [
90
90
  "Check the recent failed event before repeating a similar action."
@@ -105,10 +105,10 @@
105
105
  "decisionCorrect": false,
106
106
  "riskScore": 0.35,
107
107
  "passed": false,
108
- "latencyMs": 0.936,
108
+ "latencyMs": 1.087,
109
109
  "evidenceCount": 1,
110
110
  "evidenceIds": [
111
- "01KXDZ7CRVXW50THCG4JGFFGVZ"
111
+ "01KXE09KTHDVHXH9QY977Y3VBQ"
112
112
  ],
113
113
  "recommendedActions": [
114
114
  "Treat retrieved memory as advisory context."
@@ -129,7 +129,7 @@
129
129
  "decisionCorrect": false,
130
130
  "riskScore": 0,
131
131
  "passed": false,
132
- "latencyMs": 0.462,
132
+ "latencyMs": 0.615,
133
133
  "evidenceCount": 0,
134
134
  "evidenceIds": [],
135
135
  "recommendedActions": [],
@@ -156,10 +156,10 @@
156
156
  "decisionCorrect": true,
157
157
  "riskScore": 0.85,
158
158
  "passed": true,
159
- "latencyMs": 3.735,
159
+ "latencyMs": 3.943,
160
160
  "evidenceCount": 1,
161
161
  "evidenceIds": [
162
- "01KXDZ7CSN0BGBJM8JA5SMHZWT"
162
+ "01KXE09KVH81YSF0V7QA4ZDERQ"
163
163
  ],
164
164
  "recommendedActions": [
165
165
  "Do not proceed until the high-severity memory warning is addressed.",
@@ -181,7 +181,7 @@
181
181
  "decisionCorrect": false,
182
182
  "riskScore": 0,
183
183
  "passed": false,
184
- "latencyMs": 0.007,
184
+ "latencyMs": 0.006,
185
185
  "evidenceCount": 0,
186
186
  "evidenceIds": [],
187
187
  "recommendedActions": [],
@@ -201,10 +201,10 @@
201
201
  "decisionCorrect": true,
202
202
  "riskScore": 0.85,
203
203
  "passed": true,
204
- "latencyMs": 0.34,
204
+ "latencyMs": 0.43,
205
205
  "evidenceCount": 1,
206
206
  "evidenceIds": [
207
- "01KXDZ7CTHTZ78ND87V3MAJ2NG"
207
+ "01KXE09KWN7GCDW06T5K69GYMG"
208
208
  ],
209
209
  "recommendedActions": [
210
210
  "Review retrieved memory before acting."
@@ -225,10 +225,10 @@
225
225
  "decisionCorrect": true,
226
226
  "riskScore": 0.85,
227
227
  "passed": true,
228
- "latencyMs": 0.621,
228
+ "latencyMs": 0.633,
229
229
  "evidenceCount": 1,
230
230
  "evidenceIds": [
231
- "01KXDZ7CTXJT1PW9SH4VBT0336"
231
+ "01KXE09KX3RM8CFMH2W52JXVR5"
232
232
  ],
233
233
  "recommendedActions": [
234
234
  "Review retrieved memory before acting."
@@ -249,7 +249,7 @@
249
249
  "decisionCorrect": false,
250
250
  "riskScore": 0,
251
251
  "passed": false,
252
- "latencyMs": 0.45,
252
+ "latencyMs": 0.528,
253
253
  "evidenceCount": 0,
254
254
  "evidenceIds": [],
255
255
  "recommendedActions": [],
@@ -276,11 +276,11 @@
276
276
  "decisionCorrect": true,
277
277
  "riskScore": 0.55,
278
278
  "passed": true,
279
- "latencyMs": 4.478,
279
+ "latencyMs": 4.913,
280
280
  "evidenceCount": 2,
281
281
  "evidenceIds": [
282
- "failure:Bash:2026-07-13T14:48:45.688Z",
283
- "01KXDZ7CVSXNTMNMP89QBMCXYA"
282
+ "failure:Bash:2026-07-13T15:07:27.044Z",
283
+ "01KXE09KY55M1KXB4ET1PRF571"
284
284
  ],
285
285
  "recommendedActions": [
286
286
  "Before re-running Bash, check what changed since the last failure.",
@@ -302,7 +302,7 @@
302
302
  "decisionCorrect": false,
303
303
  "riskScore": 0,
304
304
  "passed": false,
305
- "latencyMs": 0.005,
305
+ "latencyMs": 0.007,
306
306
  "evidenceCount": 0,
307
307
  "evidenceIds": [],
308
308
  "recommendedActions": [],
@@ -322,10 +322,10 @@
322
322
  "decisionCorrect": true,
323
323
  "riskScore": 0.55,
324
324
  "passed": true,
325
- "latencyMs": 0.063,
325
+ "latencyMs": 0.102,
326
326
  "evidenceCount": 1,
327
327
  "evidenceIds": [
328
- "01KXDZ7CWPAR45TR9BEVEERQT1"
328
+ "01KXE09KZB2M0RM2W1H16XTN15"
329
329
  ],
330
330
  "recommendedActions": [
331
331
  "Check the recent failed event before repeating a similar action."
@@ -346,10 +346,10 @@
346
346
  "decisionCorrect": true,
347
347
  "riskScore": 0.35,
348
348
  "passed": true,
349
- "latencyMs": 0.464,
349
+ "latencyMs": 0.542,
350
350
  "evidenceCount": 1,
351
351
  "evidenceIds": [
352
- "01KXDZ7CX2VQB8W0AK1XY6N7W5"
352
+ "01KXE09KZVKXTT4EEPJTQ7BB7G"
353
353
  ],
354
354
  "recommendedActions": [
355
355
  "Treat retrieved memory as advisory context."
@@ -370,7 +370,7 @@
370
370
  "decisionCorrect": false,
371
371
  "riskScore": 0,
372
372
  "passed": false,
373
- "latencyMs": 0.458,
373
+ "latencyMs": 0.498,
374
374
  "evidenceCount": 0,
375
375
  "evidenceIds": [],
376
376
  "recommendedActions": [],
@@ -397,11 +397,11 @@
397
397
  "decisionCorrect": true,
398
398
  "riskScore": 0.55,
399
399
  "passed": true,
400
- "latencyMs": 4.253,
400
+ "latencyMs": 4.607,
401
401
  "evidenceCount": 2,
402
402
  "evidenceIds": [
403
- "failure:Bash:2026-07-13T14:48:45.755Z",
404
- "01KXDZ7CXVVQDPT9JT83RBRG57"
403
+ "failure:Bash:2026-07-13T15:07:27.131Z",
404
+ "01KXE09M0W99EZNCCVB9T6CYVC"
405
405
  ],
406
406
  "recommendedActions": [
407
407
  "Before re-running Bash, check what changed since the last failure.",
@@ -423,7 +423,7 @@
423
423
  "decisionCorrect": false,
424
424
  "riskScore": 0,
425
425
  "passed": false,
426
- "latencyMs": 0.008,
426
+ "latencyMs": 0.012,
427
427
  "evidenceCount": 0,
428
428
  "evidenceIds": [],
429
429
  "recommendedActions": [],
@@ -443,10 +443,10 @@
443
443
  "decisionCorrect": true,
444
444
  "riskScore": 0.55,
445
445
  "passed": true,
446
- "latencyMs": 0.069,
446
+ "latencyMs": 0.09,
447
447
  "evidenceCount": 1,
448
448
  "evidenceIds": [
449
- "01KXDZ7CYQ5BYW6GC0PFAPF5NP"
449
+ "01KXE09M22ENW4TDBZQCD5D14W"
450
450
  ],
451
451
  "recommendedActions": [
452
452
  "Check the recent failed event before repeating a similar action."
@@ -467,10 +467,10 @@
467
467
  "decisionCorrect": true,
468
468
  "riskScore": 0.35,
469
469
  "passed": true,
470
- "latencyMs": 0.465,
470
+ "latencyMs": 0.519,
471
471
  "evidenceCount": 1,
472
472
  "evidenceIds": [
473
- "01KXDZ7CZ4XMHVMV4MQ65G76KK"
473
+ "01KXE09M2HDYKTN8652470CT9V"
474
474
  ],
475
475
  "recommendedActions": [
476
476
  "Treat retrieved memory as advisory context."
@@ -491,7 +491,7 @@
491
491
  "decisionCorrect": false,
492
492
  "riskScore": 0,
493
493
  "passed": false,
494
- "latencyMs": 0.423,
494
+ "latencyMs": 0.483,
495
495
  "evidenceCount": 0,
496
496
  "evidenceIds": [],
497
497
  "recommendedActions": [],
@@ -518,11 +518,11 @@
518
518
  "decisionCorrect": true,
519
519
  "riskScore": 0.2,
520
520
  "passed": true,
521
- "latencyMs": 11.953,
521
+ "latencyMs": 12.263,
522
522
  "evidenceCount": 2,
523
523
  "evidenceIds": [
524
- "01KXDZ7D00R86YJ387KXYCAGQY",
525
- "01KXDZ7CZYQ6ZA41GTZECZYCQ8"
524
+ "01KXE09M3MDMK9ZED7TX1N9YCJ",
525
+ "01KXE09M3HD3936DV3SY6KN54G"
526
526
  ],
527
527
  "recommendedActions": [
528
528
  "This exact action has succeeded since its last failure; proceed with normal validation.",
@@ -544,7 +544,7 @@
544
544
  "decisionCorrect": true,
545
545
  "riskScore": 0,
546
546
  "passed": true,
547
- "latencyMs": 0.005,
547
+ "latencyMs": 0.007,
548
548
  "evidenceCount": 0,
549
549
  "evidenceIds": [],
550
550
  "recommendedActions": [],
@@ -564,10 +564,10 @@
564
564
  "decisionCorrect": false,
565
565
  "riskScore": 0.55,
566
566
  "passed": false,
567
- "latencyMs": 0.074,
567
+ "latencyMs": 0.089,
568
568
  "evidenceCount": 1,
569
569
  "evidenceIds": [
570
- "01KXDZ7D15SRJYZ8GRVY6YET3F"
570
+ "01KXE09M4Y1SNNTHQHKGPY1BD9"
571
571
  ],
572
572
  "recommendedActions": [
573
573
  "Check the recent failed event before repeating a similar action."
@@ -588,10 +588,10 @@
588
588
  "decisionCorrect": false,
589
589
  "riskScore": 0.35,
590
590
  "passed": false,
591
- "latencyMs": 0.595,
591
+ "latencyMs": 0.622,
592
592
  "evidenceCount": 1,
593
593
  "evidenceIds": [
594
- "01KXDZ7D1J2WAKA39WF2ACKRDD"
594
+ "01KXE09M5EAZWHPSDT7R639VBV"
595
595
  ],
596
596
  "recommendedActions": [
597
597
  "Treat retrieved memory as advisory context."
@@ -612,7 +612,7 @@
612
612
  "decisionCorrect": true,
613
613
  "riskScore": 0,
614
614
  "passed": true,
615
- "latencyMs": 0.41,
615
+ "latencyMs": 0.507,
616
616
  "evidenceCount": 0,
617
617
  "evidenceIds": [],
618
618
  "recommendedActions": [],
@@ -639,7 +639,7 @@
639
639
  "decisionCorrect": true,
640
640
  "riskScore": 0.85,
641
641
  "passed": true,
642
- "latencyMs": 3.697,
642
+ "latencyMs": 4.395,
643
643
  "evidenceCount": 1,
644
644
  "evidenceIds": [
645
645
  "recall:episodic:recall.vector_counts"
@@ -665,7 +665,7 @@
665
665
  "decisionCorrect": false,
666
666
  "riskScore": 0,
667
667
  "passed": false,
668
- "latencyMs": 0.006,
668
+ "latencyMs": 0.008,
669
669
  "evidenceCount": 0,
670
670
  "evidenceIds": [],
671
671
  "recommendedActions": [],
@@ -685,10 +685,10 @@
685
685
  "decisionCorrect": true,
686
686
  "riskScore": 0.85,
687
687
  "passed": true,
688
- "latencyMs": 0.148,
688
+ "latencyMs": 0.164,
689
689
  "evidenceCount": 1,
690
690
  "evidenceIds": [
691
- "01KXDZ7D3BDX746H56BJNSE6Z6"
691
+ "01KXE09M7NN3RTZSMYC3NQ84AA"
692
692
  ],
693
693
  "recommendedActions": [
694
694
  "Review retrieved memory before acting."
@@ -709,7 +709,7 @@
709
709
  "decisionCorrect": false,
710
710
  "riskScore": 0.55,
711
711
  "passed": false,
712
- "latencyMs": 0.292,
712
+ "latencyMs": 0.373,
713
713
  "evidenceCount": 0,
714
714
  "evidenceIds": [],
715
715
  "recommendedActions": [
@@ -737,7 +737,7 @@
737
737
  "decisionCorrect": false,
738
738
  "riskScore": 0,
739
739
  "passed": false,
740
- "latencyMs": 0.388,
740
+ "latencyMs": 0.439,
741
741
  "evidenceCount": 0,
742
742
  "evidenceIds": [],
743
743
  "recommendedActions": [],
@@ -764,11 +764,11 @@
764
764
  "decisionCorrect": true,
765
765
  "riskScore": 0.85,
766
766
  "passed": true,
767
- "latencyMs": 2.763,
767
+ "latencyMs": 2.962,
768
768
  "evidenceCount": 2,
769
769
  "evidenceIds": [
770
770
  "recall:fts:recall.fts_lookup",
771
- "01KXDZ7D4F9H6PC43S0C6FCN6W"
771
+ "01KXE09M92D89KNBQK0BQFN2Y7"
772
772
  ],
773
773
  "recommendedActions": [
774
774
  "Do not proceed until the high-severity memory warning is addressed.",
@@ -791,7 +791,7 @@
791
791
  "decisionCorrect": false,
792
792
  "riskScore": 0,
793
793
  "passed": false,
794
- "latencyMs": 0.006,
794
+ "latencyMs": 0.008,
795
795
  "evidenceCount": 0,
796
796
  "evidenceIds": [],
797
797
  "recommendedActions": [],
@@ -811,10 +811,10 @@
811
811
  "decisionCorrect": false,
812
812
  "riskScore": 0.35,
813
813
  "passed": false,
814
- "latencyMs": 0.107,
814
+ "latencyMs": 0.148,
815
815
  "evidenceCount": 1,
816
816
  "evidenceIds": [
817
- "01KXDZ7D5A65W8JASMQHJDB9ZH"
817
+ "01KXE09MA3ESCT1B6MG4ZMB0Q3"
818
818
  ],
819
819
  "recommendedActions": [
820
820
  "Treat retrieved memory as advisory context."
@@ -835,10 +835,10 @@
835
835
  "decisionCorrect": false,
836
836
  "riskScore": 0.35,
837
837
  "passed": false,
838
- "latencyMs": 0.37,
838
+ "latencyMs": 0.408,
839
839
  "evidenceCount": 1,
840
840
  "evidenceIds": [
841
- "01KXDZ7D5NRMHVW0E774P15SSF"
841
+ "01KXE09MAHS4CCJWTECVYGKJNQ"
842
842
  ],
843
843
  "recommendedActions": [
844
844
  "Treat retrieved memory as advisory context."
@@ -859,7 +859,7 @@
859
859
  "decisionCorrect": false,
860
860
  "riskScore": 0.55,
861
861
  "passed": false,
862
- "latencyMs": 0.135,
862
+ "latencyMs": 0.17,
863
863
  "evidenceCount": 0,
864
864
  "evidenceIds": [],
865
865
  "recommendedActions": [
@@ -894,12 +894,12 @@
894
894
  "decisionCorrect": true,
895
895
  "riskScore": 0.9,
896
896
  "passed": true,
897
- "latencyMs": 4.107,
897
+ "latencyMs": 4.536,
898
898
  "evidenceCount": 3,
899
899
  "evidenceIds": [
900
- "01KXDZ7D6E5JHP9RXZV2XE758W",
901
- "failure:Bash:2026-07-13T14:48:46.030Z",
902
- "01KXDZ7D6FR4DVYEP3RE8N47M1"
900
+ "01KXE09MBGSBCHBAMKCDPKPW1N",
901
+ "failure:Bash:2026-07-13T15:07:27.472Z",
902
+ "01KXE09MBGSBCHBAMKCDPKPW1P"
903
903
  ],
904
904
  "recommendedActions": [
905
905
  "Do not repeat the exact failed action until the prior error is understood or the command is changed.",
@@ -922,7 +922,7 @@
922
922
  "decisionCorrect": false,
923
923
  "riskScore": 0,
924
924
  "passed": false,
925
- "latencyMs": 0.005,
925
+ "latencyMs": 0.007,
926
926
  "evidenceCount": 0,
927
927
  "evidenceIds": [],
928
928
  "recommendedActions": [],
@@ -942,10 +942,10 @@
942
942
  "decisionCorrect": false,
943
943
  "riskScore": 0.55,
944
944
  "passed": false,
945
- "latencyMs": 0.062,
945
+ "latencyMs": 0.075,
946
946
  "evidenceCount": 1,
947
947
  "evidenceIds": [
948
- "01KXDZ7D7B6KSNRGXDM48CCF2B"
948
+ "01KXE09MCKN9T68MQ2SR7S74MX"
949
949
  ],
950
950
  "recommendedActions": [
951
951
  "Check the recent failed event before repeating a similar action."
@@ -966,10 +966,10 @@
966
966
  "decisionCorrect": false,
967
967
  "riskScore": 0.35,
968
968
  "passed": false,
969
- "latencyMs": 0.376,
969
+ "latencyMs": 0.445,
970
970
  "evidenceCount": 1,
971
971
  "evidenceIds": [
972
- "01KXDZ7D7RGKJ0SM9QT3H9WTKG"
972
+ "01KXE09MD21TX971RHKPGH6J0S"
973
973
  ],
974
974
  "recommendedActions": [
975
975
  "Treat retrieved memory as advisory context."
@@ -990,7 +990,7 @@
990
990
  "decisionCorrect": false,
991
991
  "riskScore": 0,
992
992
  "passed": false,
993
- "latencyMs": 0.361,
993
+ "latencyMs": 0.413,
994
994
  "evidenceCount": 0,
995
995
  "evidenceIds": [],
996
996
  "recommendedActions": [],
@@ -1017,11 +1017,11 @@
1017
1017
  "decisionCorrect": true,
1018
1018
  "riskScore": 0.85,
1019
1019
  "passed": true,
1020
- "latencyMs": 3.483,
1020
+ "latencyMs": 4.009,
1021
1021
  "evidenceCount": 2,
1022
1022
  "evidenceIds": [
1023
- "01KXDZ7D8HCCD19F4MX5T8TQW6",
1024
- "01KXDZ7D8GSJYEYEA0PQA2CFEQ"
1023
+ "01KXE09ME0E710DBZCDRM6ZAPK",
1024
+ "01KXE09ME1SDSF3N9ER3MEAYRW"
1025
1025
  ],
1026
1026
  "recommendedActions": [
1027
1027
  "Do not proceed until the high-severity memory warning is addressed.",
@@ -1043,7 +1043,7 @@
1043
1043
  "decisionCorrect": false,
1044
1044
  "riskScore": 0,
1045
1045
  "passed": false,
1046
- "latencyMs": 0.005,
1046
+ "latencyMs": 0.008,
1047
1047
  "evidenceCount": 0,
1048
1048
  "evidenceIds": [],
1049
1049
  "recommendedActions": [],
@@ -1063,11 +1063,11 @@
1063
1063
  "decisionCorrect": true,
1064
1064
  "riskScore": 0.85,
1065
1065
  "passed": true,
1066
- "latencyMs": 0.11,
1066
+ "latencyMs": 0.185,
1067
1067
  "evidenceCount": 2,
1068
1068
  "evidenceIds": [
1069
- "01KXDZ7D9D2T5HC5363DW62Q5Y",
1070
- "01KXDZ7D9CB9AN9NTPT76ER3FS"
1069
+ "01KXE09MF4FRFAGWDMZYWSSYF3",
1070
+ "01KXE09MF382A2BJNVBV44BH5Q"
1071
1071
  ],
1072
1072
  "recommendedActions": [
1073
1073
  "Review retrieved memory before acting."
@@ -1088,11 +1088,11 @@
1088
1088
  "decisionCorrect": true,
1089
1089
  "riskScore": 0.85,
1090
1090
  "passed": true,
1091
- "latencyMs": 0.436,
1091
+ "latencyMs": 0.514,
1092
1092
  "evidenceCount": 2,
1093
1093
  "evidenceIds": [
1094
- "01KXDZ7D9TWCABY6SX2PRFF3GC",
1095
- "01KXDZ7D9RRHJGZX0SGG6KW02G"
1094
+ "01KXE09MFK7PV5JRT20T18GH2Q",
1095
+ "01KXE09MFMTZF8990X4FNN72R6"
1096
1096
  ],
1097
1097
  "recommendedActions": [
1098
1098
  "Review retrieved memory before acting."
@@ -1113,7 +1113,7 @@
1113
1113
  "decisionCorrect": false,
1114
1114
  "riskScore": 0,
1115
1115
  "passed": false,
1116
- "latencyMs": 0.364,
1116
+ "latencyMs": 0.477,
1117
1117
  "evidenceCount": 0,
1118
1118
  "evidenceIds": [],
1119
1119
  "recommendedActions": [],
@@ -1140,22 +1140,22 @@
1140
1140
  "decisionCorrect": true,
1141
1141
  "riskScore": 0.85,
1142
1142
  "passed": true,
1143
- "latencyMs": 21.196,
1143
+ "latencyMs": 21.585,
1144
1144
  "evidenceCount": 13,
1145
1145
  "evidenceIds": [
1146
- "01KXDZ7DVW999JP92H6F6TV4Q7",
1147
- "01KXDZ7DTEMQ67D5HHJD5A4Q4P",
1148
- "01KXDZ7DRHKN6TRS0FD99EXHRR",
1149
- "01KXDZ7DQDBDNYA0CB6026EN7J",
1150
- "01KXDZ7DP9VB7FAQMF83XT32QB",
1151
- "01KXDZ7DMTNF8ZWMG97CX7M3YD",
1152
- "01KXDZ7DM0CM23Y6ETWAB3M3KM",
1153
- "01KXDZ7DJGG0YRG6SYVP00ZC21",
1154
- "01KXDZ7DJCQJV1NCSEF680GCCQ",
1155
- "01KXDZ7DFMRHWTY6X42JC6C1FG",
1156
- "01KXDZ7DEDR8KPS7T6E8QZM1BY",
1157
- "01KXDZ7DDP9V0DGWT3YJADCSM3",
1158
- "01KXDZ7DBFNRKSF2KH1FBXGDJV"
1146
+ "01KXE09N3T9YNYY0NTHA51W62G",
1147
+ "01KXE09N3KVDJJS43TSNP6ENVC",
1148
+ "01KXE09N2CE328SXRXJCH3Z74A",
1149
+ "01KXE09N26TY6RT4C8JST58PXD",
1150
+ "01KXE09N1MNFD0KGFQQR6JNEWS",
1151
+ "01KXE09N1FJESNKB9AYKCVWN0R",
1152
+ "01KXE09N071JZ8HGQ0V4TC4PF7",
1153
+ "01KXE09MZ1ZTJ6T6KVNCBTRJW2",
1154
+ "01KXE09MWBZ49Q7MT4A1DHEMRY",
1155
+ "01KXE09MPH6KST4EQA6MD9MJCA",
1156
+ "01KXE09MN0SN0T0B9Q4GJFFZ79",
1157
+ "01KXE09MKRBKW8GV17QGFFJERT",
1158
+ "01KXE09MJQHKHMH87DPWAPJ852"
1159
1159
  ],
1160
1160
  "recommendedActions": [
1161
1161
  "Do not proceed until the high-severity memory warning is addressed.",
@@ -1198,10 +1198,10 @@
1198
1198
  "decisionCorrect": true,
1199
1199
  "riskScore": 0.85,
1200
1200
  "passed": true,
1201
- "latencyMs": 0.427,
1201
+ "latencyMs": 0.516,
1202
1202
  "evidenceCount": 1,
1203
1203
  "evidenceIds": [
1204
- "01KXDZ7F3CRSBGB5NWMYJGJG6C"
1204
+ "01KXE09PG3YKMX3PNE2KPWDQX4"
1205
1205
  ],
1206
1206
  "recommendedActions": [
1207
1207
  "Review retrieved memory before acting."
@@ -1222,14 +1222,14 @@
1222
1222
  "decisionCorrect": false,
1223
1223
  "riskScore": 0.35,
1224
1224
  "passed": false,
1225
- "latencyMs": 1.589,
1225
+ "latencyMs": 1.502,
1226
1226
  "evidenceCount": 5,
1227
1227
  "evidenceIds": [
1228
- "01KXDZ7FMS8GJ0E0VZDPTNMFWA",
1229
- "01KXDZ7FEEQ4N77YQ6N07WW0GV",
1230
- "01KXDZ7FDQABHDN1HWRRKPKJ7E",
1231
- "01KXDZ7FC5XZMKTACKRSH258TR",
1232
- "01KXDZ7FN452RQXNN7FBJ9EX1K"
1228
+ "01KXE09PX05XK1MNQM6HTSJRDX",
1229
+ "01KXE09PRMM0S934H24B8J9RS3",
1230
+ "01KXE09PS93CENDK7GY7FMJ0F1",
1231
+ "01KXE09PNN1Z71Z9SF29F1PNHS",
1232
+ "01KXE09Q1YJF9QHNDRKJ2HNRFG"
1233
1233
  ],
1234
1234
  "recommendedActions": [
1235
1235
  "Treat retrieved memory as advisory context."
@@ -1250,7 +1250,7 @@
1250
1250
  "decisionCorrect": false,
1251
1251
  "riskScore": 0,
1252
1252
  "passed": false,
1253
- "latencyMs": 0.739,
1253
+ "latencyMs": 0.803,
1254
1254
  "evidenceCount": 0,
1255
1255
  "evidenceIds": [],
1256
1256
  "recommendedActions": [],
@@ -1265,7 +1265,7 @@
1265
1265
  }
1266
1266
  ],
1267
1267
  "artifactRedactionSweep": {
1268
- "checkedAt": "2026-07-13T14:48:49.340Z",
1268
+ "checkedAt": "2026-07-13T15:07:31.207Z",
1269
1269
  "filesChecked": [
1270
1270
  "benchmarks/output/guardbench-manifest.json",
1271
1271
  "benchmarks/output/guardbench-raw.json",