Skip to content

Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors

Rust

docs: publish the 3.1 release benchmark

AkitaOnRailsefe6b96171bf
diff --git a/docs/benchmarks/README.md b/docs/benchmarks/README.md
index a0f12236..609874f7 100644
--- a/docs/benchmarks/README.md
+++ b/docs/benchmarks/README.md
@@ -13,6 +13,8 @@ cargo run --release -p ai-memory-eval -- retrieval --fetch
| date | dataset | mode | overall hit@5 | file |
|---|---|---|---|---|
+| **2026-10-16** | LongMemEval-S (v1) | **zero-llm FTS + observation stream (3.1 release)** | **0.951** | [longmemeval-s-3.1.0-release.md](longmemeval-s-3.1.0-release.md) |
+| **2026-10-16** | LongMemEval-S (v1) | **local embeddings + observation stream (3.1 release)** | **0.940** | [longmemeval-s-3.1.0-release.md](longmemeval-s-3.1.0-release.md) |
| **2026-10-15** | LongMemEval-S (v1) | **zero-llm FTS + observation stream (3.0.1 release)** | **0.951** | [longmemeval-s-3.0.1-release.md](longmemeval-s-3.0.1-release.md) |
| **2026-10-15** | LongMemEval-S (v1) | **local embeddings + observation stream (3.0.1 release)** | **0.943** | [longmemeval-s-3.0.1-release.md](longmemeval-s-3.0.1-release.md) |
| **2026-10-13** | LongMemEval-S (v1) | **zero-llm FTS + observation stream (3.0.0 release; on by default)** | **0.951** | [longmemeval-s-3.0.0-release.md](longmemeval-s-3.0.0-release.md) |
@@ -23,6 +25,13 @@ cargo run --release -p ai-memory-eval -- retrieval --fetch
| 2026-09-01 | LongMemEval-S (v1) | zero-llm, stopword-filtered FTS (prior) | 0.668 | [longmemeval-s-2026-09-01-fts.md](longmemeval-s-2026-09-01-fts.md) |
| 2026-09-01 | LongMemEval-S (v1) | zero-llm (pre-2.0 FTS) | 0.617 | [longmemeval-s-2026-09-01.md](longmemeval-s-2026-09-01.md) |
+The 2026-10-16 minor-release run found no material retrieval regression
+against 3.0.1. FTS is identical at every overall cutoff. Local embeddings
+moved from 0.943 to 0.940 `hit@5`, inside the pre-registered +/-0.005 noise
+band; its latency and context size stayed stable. See the
+[3.1 release report](longmemeval-s-3.1.0-release.md) for every slice, ingest
+performance, commands, and the explicit post-measurement commit boundary.
+
The 2026-10-15 patch-release re-run found no retrieval regression against
3.0.0. FTS accuracy is identical (`hit@5` 0.951), while local embeddings
moved 0.945 -> 0.943, inside the +/-0.005 overall noise band. Query p50/p95
diff --git a/docs/benchmarks/longmemeval-s-3.1.0-release.md b/docs/benchmarks/longmemeval-s-3.1.0-release.md
new file mode 100644
index 00000000..76bd36ba
--- /dev/null
+++ b/docs/benchmarks/longmemeval-s-3.1.0-release.md
@@ -0,0 +1,81 @@
+# LongMemEval-S release benchmark: 3.1.0
+
+The 3.1 release benchmark runs all 470 answerable LongMemEval-S questions
+through the shipped hook capture pipeline. It measures both supported default
+retrieval paths: zero-LLM FTS and the in-process local embedder, with the
+observation stream enabled as it is in production.
+
+## Provenance
+
+- code: both modes used `a288a1093550b096075087665319054c3ef598ba`.
+ Changes added after measurement are IBM Bob lifecycle integration,
+ documentation, benchmark publication, and an immutable source pin for the
+ same `zip` 8.6.0 code already in the lockfile. None changes retrieval,
+ existing-agent capture, model weights, or ranking. The benchmark was not
+ repeated for those post-measurement changes.
+- dataset: `longmemeval_s`, sha256
+ `08d8dad4be43ee2049a22ff5674eb86725d0ce5ff434cde2627e5e8e7e117894`.
+- hardware: AMD Ryzen 9 7950X3D (32 threads). No build or test workload ran
+ concurrently with either measurement.
+- commands, after building `ai-memory` and `ai-memory-eval` in release mode:
+
+ ```bash
+ ai-memory-eval retrieval \
+ --server-bin target/release/ai-memory \
+ --concurrency 6 --http-timeout-secs 900 \
+ --server-env AI_MEMORY_RETRIEVAL__OBSERVATION_STREAM=true
+
+ ai-memory-eval retrieval \
+ --server-bin target/release/ai-memory \
+ --concurrency 6 --http-timeout-secs 900 \
+ --embeddings local \
+ --server-env AI_MEMORY_RETRIEVAL__OBSERVATION_STREAM=true
+ ```
+
+- run dirs: `evals/runs/2026-10-16T01-10-11-488161722Z-retrieval` (FTS)
+ and `evals/runs/2026-10-16T02-15-24-986694723Z-retrieval` (local).
+
+## Results
+
+Overall accuracy, latency, and returned context remain within the
+pre-registered stability bounds against 3.0.1:
+
+| mode | hit@1 | hit@5 | hit@10 | p50 | p95 | mean context tokens |
+|---|---:|---:|---:|---:|---:|---:|
+| FTS 3.0.1 | 0.857 | 0.951 | 0.972 | 12 ms | 101 ms | 528 |
+| **FTS 3.1** | **0.857** | **0.951** | **0.972** | **12 ms** | **101 ms** | **528** |
+| local 3.0.1 | 0.768 | 0.943 | 0.977 | 98 ms | 228 ms | 528 |
+| **local 3.1** | **0.755** | **0.940** | **0.972** | **99 ms** | **225 ms** | **528** |
+
+FTS is identical at every overall cutoff. Local `hit@5` changed by -0.003,
+inside the pre-registered +/-0.005 run-to-run noise band. Local `hit@1` moved
+-0.013 and `hit@10` -0.005, without a systematic slice regression. Latency
+and returned context are unchanged in practical terms.
+
+Per-slice `hit@5`:
+
+| slice | n | FTS 3.0.1 | FTS 3.1 | local 3.0.1 | local 3.1 |
+|---|---:|---:|---:|---:|---:|
+| knowledge-update | 72 | 1.000 | 1.000 | 1.000 | 1.000 |
+| multi-session | 121 | 0.967 | 0.967 | 0.975 | 0.975 |
+| temporal-reasoning | 127 | 0.921 | 0.921 | 0.913 | 0.898 |
+| single-session-assistant | 56 | 1.000 | 1.000 | 0.893 | 0.893 |
+| single-session-preference | 30 | 0.700 | 0.700 | 0.867 | 0.900 |
+| single-session-user | 64 | 1.000 | 1.000 | 0.953 | 0.953 |
+
+All FTS slices are exact. Four of six local slices are exact;
+temporal-reasoning changed by -0.015, within its pre-registered slice band,
+and preference gained one question (+0.033) in the 30-question slice. The
+pre-registration explicitly called out one-question steps on that smallest
+slice.
+
+## Ingest
+
+| mode | items | requests | throughput | wall clock | fallback answers |
+|---|---:|---:|---:|---:|---:|
+| FTS | 278,946 | 1,403 | 12 items/s per question stream | 3,910.0 s | 0/470 |
+| local | 278,946 | 1,403 | 9 items/s per question stream | 4,975.3 s | 0/470 |
+
+No document exceeded the 16 KiB capture bound. The raw-observation fallback
+was not needed, so the result measures indexed retrieval rather than fallback
+behavior.