File size: 12,322 Bytes
15f9cd9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
<!doctype html>
<html lang="en">
<head>
  <meta charset="utf-8" />
  <title>Audrey Memory Benchmark</title>
  <style>
    body { font-family: "Segoe UI", Arial, sans-serif; margin: 32px; color: #111827; background: #f8fafc; }
    main { max-width: 1120px; margin: 0 auto; }
    h1, h2 { margin-bottom: 12px; }
    p, li { line-height: 1.5; }
    .callout { background: white; border: 1px solid #cbd5e1; border-radius: 16px; padding: 20px; margin-bottom: 24px; }
    .grid { display: grid; gap: 24px; grid-template-columns: 1fr; }
    img { width: 100%; border: 1px solid #cbd5e1; border-radius: 16px; background: white; }
    table { width: 100%; border-collapse: collapse; background: white; border-radius: 16px; overflow: hidden; }
    th, td { border: 1px solid #cbd5e1; padding: 12px; vertical-align: top; text-align: left; }
    th { background: #e2e8f0; }
    code { background: #e2e8f0; padding: 2px 6px; border-radius: 6px; }
  </style>
</head>
<body>
  <main>
    <h1>Audrey Memory Benchmark</h1>
    <div class="callout">
      <p><strong>Method:</strong> Audrey is scored on a LongMemEval-inspired retrieval benchmark plus an operation-level lifecycle benchmark. The report still separates local Audrey-versus-baseline results from published external LoCoMo numbers so the comparison stays honest.</p>
      <p><strong>Run:</strong> <code>node benchmarks/run.js --provider mock --dimensions 64</code></p>
      <p><strong>Generated:</strong> 2026-05-01T03:20:07.968Z</p>
    </div>

    <div class="grid">
      <section class="callout">
        <h2>Combined Local Benchmark</h2>
        <img src="./local-overall.svg" alt="Combined local benchmark bar chart" />
      </section>

      
    <section class="callout">
      <h2>Retrieval capabilities Benchmark</h2>
      <p>LongMemEval-style memory abilities plus privacy and abstention.</p>
      <img src="./retrieval-overall.svg" alt="Retrieval capabilities Benchmark chart" />
    </section>
  

    <section class="callout">
      <h2>Memory operations Benchmark</h2>
      <p>Update, delete, merge, and abstention behavior after lifecycle operations.</p>
      <img src="./operations-overall.svg" alt="Memory operations Benchmark chart" />
    </section>
  

      <section class="callout">
        <h2>Published Leaderboard</h2>
        <img src="./published-locomo.svg" alt="Published LoCoMo leaderboard bar chart" />
      </section>
    </div>

    <section class="callout">
      <h2>Case Matrix</h2>
      <table>
        <thead>
          <tr>
            <th>Case</th>
            <th>Suite</th>
            <th>Family</th>
            <th>Audrey</th><th>Vector Only</th><th>Keyword + Recency</th><th>Recent Window</th>
          </tr>
        </thead>
        <tbody>
          
    <tr>
      <td>Information extraction</td>
      <td>retrieval</td>
      <td>information_extraction</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td>
    </tr>
  

    <tr>
      <td>Knowledge updates</td>
      <td>retrieval</td>
      <td>knowledge_updates</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">retrieved expected evidence, but conflicting evidence still appeared later</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">retrieved expected evidence, but conflicting evidence still appeared later</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">retrieved expected evidence, but conflicting evidence still appeared later</span></td>
    </tr>
  

    <tr>
      <td>Multi-session reasoning</td>
      <td>retrieval</td>
      <td>multi_session_reasoning</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td>
    </tr>
  

    <tr>
      <td>Temporal reasoning</td>
      <td>retrieval</td>
      <td>temporal_reasoning</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td>
    </tr>
  

    <tr>
      <td>Abstention</td>
      <td>retrieval</td>
      <td>abstention</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">correct abstention</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">no leak, but retrieved tangential context</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">no leak, but retrieved tangential context</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">no leak, but retrieved tangential context</span></td>
    </tr>
  

    <tr>
      <td>Conflict resolution</td>
      <td>retrieval</td>
      <td>conflict_resolution</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">retrieved expected evidence, but conflicting evidence still appeared later</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">blocked content outranked the correct answer</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">blocked content outranked the correct answer</span></td>
    </tr>
  

    <tr>
      <td>Procedural learning</td>
      <td>retrieval</td>
      <td>procedural_learning</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td>
    </tr>
  

    <tr>
      <td>Privacy boundary</td>
      <td>retrieval</td>
      <td>privacy_boundary</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">correct abstention</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">leaked restricted content</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">leaked restricted content</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">leaked restricted content</span></td>
    </tr>
  

    <tr>
      <td>Update and overwrite</td>
      <td>operations</td>
      <td>update_overwrite</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">retrieved expected evidence, but conflicting evidence still appeared later</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">retrieved expected evidence, but conflicting evidence still appeared later</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">retrieved expected evidence, but conflicting evidence still appeared later</span></td>
    </tr>
  

    <tr>
      <td>Delete and abstain</td>
      <td>operations</td>
      <td>delete_and_abstain</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">correct abstention</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">leaked restricted content</span></td><td style="background:#fff7ed;color:#9a3412">0.50<br /><span style="font-size:12px">no leak, but retrieved tangential context</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">leaked restricted content</span></td>
    </tr>
  

    <tr>
      <td>Semantic merge</td>
      <td>operations</td>
      <td>semantic_merge</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td>
    </tr>
  

    <tr>
      <td>Procedural merge</td>
      <td>operations</td>
      <td>procedural_merge</td>
      <td style="background:#ecfdf5;color:#065f46">1.00<br /><span style="font-size:12px">retrieved expected evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td><td style="background:#fef2f2;color:#991b1b">0.00<br /><span style="font-size:12px">missed target evidence</span></td>
    </tr>
  
        </tbody>
      </table>
    </section>

    <section class="callout">
      <h2>March 23, 2026 Memory Trends</h2>
      <ul>
        
    <li>
      <strong>Memory is moving from flat retrieval to typed systems</strong><br />
      Recent work treats episodic, semantic, procedural, and graph memory as separate but cooperating layers.<br />
      <a href="https://arxiv.org/abs/2507.03724">https://arxiv.org/abs/2507.03724</a>
    </li>
  

    <li>
      <strong>Benchmarks now emphasize multi-session realism</strong><br />
      LongMemEval and LoCoMo push memory systems toward temporal updates, abstraction, and cross-session reasoning instead of single-turn fact recall.<br />
      <a href="https://arxiv.org/abs/2410.10813">https://arxiv.org/abs/2410.10813</a>
    </li>
  

    <li>
      <strong>Context engineering is now competing with retrieval-first designs</strong><br />
      Letta argues filesystem and memory-block approaches can outperform simpler retrieval-only memory on realistic long-horizon tasks.<br />
      <a href="https://www.letta.com/blog/memory-blocks">https://www.letta.com/blog/memory-blocks</a>
    </li>
  

    <li>
      <strong>Production teams care about latency and token footprint, not just recall quality</strong><br />
      Mem0 frames memory as a cost and latency optimization surface in addition to a personalization surface.<br />
      <a href="https://arxiv.org/abs/2504.19413">https://arxiv.org/abs/2504.19413</a>
    </li>
  

    <li>
      <strong>Temporal and multimodal memory are becoming table stakes</strong><br />
      MIRIX and Graphiti both model time and state change explicitly instead of assuming memories stay forever true.<br />
      <a href="https://arxiv.org/abs/2507.07957">https://arxiv.org/abs/2507.07957</a>
    </li>
  
      </ul>
    </section>
  </main>
</body>
</html>