Release/v2.33.1 - #120
Conversation
Score each recall candidate set against one reference time and compare consolidation thresholds against raw cosine similarity. Force semantic duplicate checks, inspect a bounded candidate set, and preserve normal MCP response shapes. Add regression coverage, refresh benchmark documentation, and prepare v2.33.1. Fixes #113
Enable Dependabot coverage for the console, tighten scoring comments, and sync the packaged README navigation.
📝 WalkthroughWalkthroughChangesRetrieval and consolidation corrections
v2.33.1 release metadata and documentation
Console UI dependency automation
Estimated code review effort: 3 (Moderate) | ~25 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 2✅ Passed checks (2 passed)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@CHANGELOG.md`:
- Line 11: Update the changelog entry to call this “semantic duplicate
detection” or “semantic merging,” and clarify that semantic merging is skipped
when no meaning-similarity is available while exact normalized-content
deduplication still handles identical content, including syntax-heavy content.
In `@marm-mcp-server/README.md`:
- Line 182: Update the benchmark section’s release labels to match the actual
benchmark run: use v2.33.1 for both the re-measurement note and the LoCoMo
“Current” row only if the data was rerun against v2.33.1; otherwise mark both
references as historical v2.33.0 results.
In `@README.md`:
- Line 236: Correct the benchmark run-count methodology statement in README.md
at lines 236-236 and
marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md at lines 210-210:
scope the one-run claim to sections 1–4 and explicitly state that LoCoMo results
use two identically configured runs, keeping both documentation copies
consistent.
- Line 180: Update the benchmark version labels in README.md at lines 180-180
and marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md at lines
154-154 to identify the results as a v2.33.0 baseline, removing any implication
that they are current v2.33.1 measurements; keep both documents consistent.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro Plus
Run ID: cd72cf90-b564-4629-aee4-6c270c5b0bea
📒 Files selected for processing (24)
.github/dependabot.ymlCHANGELOG.mdREADME.mddocs/INSTALL-DOCKER.mddocs/INSTALL-LINUX.mddocs/INSTALL-PLATFORMS.mddocs/INSTALL-WINDOWS.mddocs/TECHNICAL-OVERVIEW.mdmarm-mcp-server/Dockerfilemarm-mcp-server/README.mdmarm-mcp-server/docker-compose.ymlmarm-mcp-server/marm_mcp_server/__init__.pymarm-mcp-server/marm_mcp_server/config/settings.pymarm-mcp-server/marm_mcp_server/core/consolidation.pymarm-mcp-server/marm_mcp_server/core/memory.pymarm-mcp-server/marm_mcp_server/core/memory_recall.pymarm-mcp-server/marm_mcp_server/core/memory_scoring.pymarm-mcp-server/marm_mcp_server/core/memory_utils.pymarm-mcp-server/marm_mcp_server/resources/marm-docs/README.mdmarm-mcp-server/marm_mcp_server/server.pymarm-mcp-server/pyproject.tomlmarm-mcp-server/server.jsonmarm-mcp-server/tests/test_consolidation_write_time.pymarm-mcp-server/tests/test_hybrid_search.py
📜 Review details
🧰 Additional context used
📓 Path-based instructions (11)
**/*.{py,json,md,toml,yml,yaml,Dockerfile}
📄 CodeRabbit inference engine (AGENTS.md)
Keep changes surgical: modify only what the task requires, match existing style, and preserve behavior during refactors.
Files:
marm-mcp-server/pyproject.tomlmarm-mcp-server/marm_mcp_server/server.pymarm-mcp-server/docker-compose.ymldocs/TECHNICAL-OVERVIEW.mdmarm-mcp-server/server.jsondocs/INSTALL-LINUX.mddocs/INSTALL-WINDOWS.mdmarm-mcp-server/marm_mcp_server/config/settings.pydocs/INSTALL-PLATFORMS.mdmarm-mcp-server/marm_mcp_server/core/memory_utils.pydocs/INSTALL-DOCKER.mdmarm-mcp-server/marm_mcp_server/core/memory.pyCHANGELOG.mdmarm-mcp-server/marm_mcp_server/__init__.pymarm-mcp-server/tests/test_consolidation_write_time.pymarm-mcp-server/marm_mcp_server/core/consolidation.pymarm-mcp-server/marm_mcp_server/core/memory_scoring.pyREADME.mdmarm-mcp-server/README.mdmarm-mcp-server/marm_mcp_server/resources/marm-docs/README.mdmarm-mcp-server/marm_mcp_server/core/memory_recall.pymarm-mcp-server/tests/test_hybrid_search.py
**/*
📄 CodeRabbit inference engine (AGENTS.md)
**/*: Never commit changes without an explicit user request.
Use semantic versioning: MAJOR for breaking changes, MINOR for new tools, parameters, or features, and PATCH for fixes and documentation updates.
Files:
marm-mcp-server/pyproject.tomlmarm-mcp-server/Dockerfilemarm-mcp-server/marm_mcp_server/server.pymarm-mcp-server/docker-compose.ymldocs/TECHNICAL-OVERVIEW.mdmarm-mcp-server/server.jsondocs/INSTALL-LINUX.mddocs/INSTALL-WINDOWS.mdmarm-mcp-server/marm_mcp_server/config/settings.pydocs/INSTALL-PLATFORMS.mdmarm-mcp-server/marm_mcp_server/core/memory_utils.pydocs/INSTALL-DOCKER.mdmarm-mcp-server/marm_mcp_server/core/memory.pyCHANGELOG.mdmarm-mcp-server/marm_mcp_server/__init__.pymarm-mcp-server/tests/test_consolidation_write_time.pymarm-mcp-server/marm_mcp_server/core/consolidation.pymarm-mcp-server/marm_mcp_server/core/memory_scoring.pyREADME.mdmarm-mcp-server/README.mdmarm-mcp-server/marm_mcp_server/resources/marm-docs/README.mdmarm-mcp-server/marm_mcp_server/core/memory_recall.pymarm-mcp-server/tests/test_hybrid_search.py
marm-mcp-server/{pyproject.toml,server.json,marm_mcp_server/__init__.py,marm_mcp_server/config/settings.py,marm_mcp_server/server.py,Dockerfile,docker-compose.yml}
📄 CodeRabbit inference engine (AGENTS.md)
When bumping the version, update every listed package, manifest, server, Docker, and configuration version occurrence, and audit with
python scripts/find-versions.py.
Files:
marm-mcp-server/pyproject.tomlmarm-mcp-server/Dockerfilemarm-mcp-server/marm_mcp_server/server.pymarm-mcp-server/docker-compose.ymlmarm-mcp-server/server.jsonmarm-mcp-server/marm_mcp_server/config/settings.pymarm-mcp-server/marm_mcp_server/__init__.py
marm-mcp-server/marm_mcp_server/server.py
📄 CodeRabbit inference engine (AGENTS.md)
Keep HTTP MCP tools listed in
MCP_TOOL_OPERATIONS; tools absent from this whitelist must not be exposed over HTTP.
Files:
marm-mcp-server/marm_mcp_server/server.py
marm-mcp-server/marm_mcp_server/**/*.py
📄 CodeRabbit inference engine (AGENTS.md)
marm-mcp-server/marm_mcp_server/**/*.py: All memory writes must use the serialized asynchronous write queue; do not add bypass paths.
A semantic-store failure duringmarm_log_entrymust never fail the corresponding log write.
Keep the memory database and isolated concept-graph database on separate connection pools; never share connections between them.
Graph and concept failures must never break the seven core memory tools.
marm_smart_recallmust keep primary memory ranking authoritative; graph enrichment is bounded, read-only, fail-open, and must be trimmed before primary results when enforcing response limits.
Use one lazy-loadedjinaai/jina-embeddings-v2-small-enencoder with 512 dimensions, serialized behind a lock; writes must succeed if the encoder is unavailable.
Keep orchestration in its current owner file and extract modules only at real boundaries consistent with the existing endpoint split.
Comments must be minimal, explain only non-obvious reasons, and never narrate the next line.
Files:
marm-mcp-server/marm_mcp_server/server.pymarm-mcp-server/marm_mcp_server/config/settings.pymarm-mcp-server/marm_mcp_server/core/memory_utils.pymarm-mcp-server/marm_mcp_server/core/memory.pymarm-mcp-server/marm_mcp_server/__init__.pymarm-mcp-server/marm_mcp_server/core/consolidation.pymarm-mcp-server/marm_mcp_server/core/memory_scoring.pymarm-mcp-server/marm_mcp_server/core/memory_recall.py
**/*.{md,py}
📄 CodeRabbit inference engine (AGENTS.md)
When adding or removing an MCP tool, update every full tool list and tool count in the README, protocol documentation, FAQ, and maintained
marm-docscopies; add tests covering both transports.
Files:
marm-mcp-server/marm_mcp_server/server.pydocs/TECHNICAL-OVERVIEW.mddocs/INSTALL-LINUX.mddocs/INSTALL-WINDOWS.mdmarm-mcp-server/marm_mcp_server/config/settings.pydocs/INSTALL-PLATFORMS.mdmarm-mcp-server/marm_mcp_server/core/memory_utils.pydocs/INSTALL-DOCKER.mdmarm-mcp-server/marm_mcp_server/core/memory.pyCHANGELOG.mdmarm-mcp-server/marm_mcp_server/__init__.pymarm-mcp-server/tests/test_consolidation_write_time.pymarm-mcp-server/marm_mcp_server/core/consolidation.pymarm-mcp-server/marm_mcp_server/core/memory_scoring.pyREADME.mdmarm-mcp-server/README.mdmarm-mcp-server/marm_mcp_server/resources/marm-docs/README.mdmarm-mcp-server/marm_mcp_server/core/memory_recall.pymarm-mcp-server/tests/test_hybrid_search.py
**/*.py
⚙️ CodeRabbit configuration file
**/*.py: Prioritize runtime correctness, async/concurrency safety, SQLite transaction safety, auth/rate-limit behavior, release-breaking packaging issues, and MCP protocol compatibility.
Files:
marm-mcp-server/marm_mcp_server/server.pymarm-mcp-server/marm_mcp_server/config/settings.pymarm-mcp-server/marm_mcp_server/core/memory_utils.pymarm-mcp-server/marm_mcp_server/core/memory.pymarm-mcp-server/marm_mcp_server/__init__.pymarm-mcp-server/tests/test_consolidation_write_time.pymarm-mcp-server/marm_mcp_server/core/consolidation.pymarm-mcp-server/marm_mcp_server/core/memory_scoring.pymarm-mcp-server/marm_mcp_server/core/memory_recall.pymarm-mcp-server/tests/test_hybrid_search.py
**/*.md
⚙️ CodeRabbit configuration file
**/*.md: Only flag documentation issues that are materially wrong, misleading for installation/release behavior, or inconsistent with live MCP behavior. Skip style, phrasing, formatting, and wording preferences.
Files:
docs/TECHNICAL-OVERVIEW.mddocs/INSTALL-LINUX.mddocs/INSTALL-WINDOWS.mddocs/INSTALL-PLATFORMS.mddocs/INSTALL-DOCKER.mdCHANGELOG.mdREADME.mdmarm-mcp-server/README.mdmarm-mcp-server/marm_mcp_server/resources/marm-docs/README.md
marm-mcp-server/{server.py,server_stdio.py,server.json}
📄 CodeRabbit inference engine (AGENTS.md)
When adding or removing an MCP tool, update the endpoint implementation, HTTP route and whitelist, STDIO registration/wrapper, and
server.jsontools array.
Files:
marm-mcp-server/server.json
marm-mcp-server/tests/**/*.py
📄 CodeRabbit inference engine (AGENTS.md)
marm-mcp-server/tests/**/*.py: Run tests withpytestfrommarm-mcp-server/; tests should exercise real FastAPI endpoints and real SQLite, using mocks only when they preserve at least 95% fidelity and provide meaningful speedup.
Every new MARM Console API route must have at least one happy-path FastAPI response-contract test with the MCP adapter stubbed.
Do not write existence-check or coded-to-pass tests; prefer deep tests exercising real paths over broad shallow coverage.
Usepytest.mark.skiponly for genuinely unavailable dependencies, never merely to avoid implementation effort.
Files:
marm-mcp-server/tests/test_consolidation_write_time.pymarm-mcp-server/tests/test_hybrid_search.py
**/tests/**
⚙️ CodeRabbit configuration file
**/tests/**: Focus on tests that are flaky, non-isolated, incorrectly asserting behavior, or missing coverage for a changed high-risk path. Skip minor naming, comments, and layout preferences.
Files:
marm-mcp-server/tests/test_consolidation_write_time.pymarm-mcp-server/tests/test_hybrid_search.py
🔇 Additional comments (24)
.github/dependabot.yml (1)
14-27: LGTM!marm-mcp-server/marm_mcp_server/core/memory_recall.py (1)
5-5: LGTM!Also applies to: 146-168, 292-332, 378-385
marm-mcp-server/marm_mcp_server/core/memory_utils.py (1)
76-93: LGTM!marm-mcp-server/marm_mcp_server/core/memory_scoring.py (1)
14-17: LGTM!marm-mcp-server/tests/test_hybrid_search.py (1)
1366-1409: LGTM!Also applies to: 1412-1528, 1534-1658, 1661-1741, 1745-1800
marm-mcp-server/marm_mcp_server/core/memory.py (1)
412-425: LGTM!marm-mcp-server/marm_mcp_server/core/consolidation.py (1)
12-13: LGTM!Also applies to: 60-89
marm-mcp-server/tests/test_consolidation_write_time.py (1)
132-156: LGTM!Also applies to: 166-176
README.md (1)
8-8: LGTM!Also applies to: 937-937, 1001-1001
docs/INSTALL-DOCKER.md (1)
5-5: LGTM!docs/INSTALL-LINUX.md (1)
5-5: LGTM!Also applies to: 323-323
docs/INSTALL-PLATFORMS.md (1)
1-1: LGTM!docs/INSTALL-WINDOWS.md (1)
5-5: LGTM!Also applies to: 297-297
docs/TECHNICAL-OVERVIEW.md (1)
3-3: LGTM!marm-mcp-server/Dockerfile (1)
76-76: LGTM!marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md (1)
1-1: LGTM!Also applies to: 905-905, 969-969
marm-mcp-server/README.md (2)
238-240: 📐 Maintainability & Code QualityReconcile the benchmark provenance.
Line 238 says the accuracy range comes from two runs, but the benchmark introduction states that all numbers come from one run. Scope the single-run statement to the timed benchmarks or document the multi-run accuracy methodology explicitly.
Source: Path instructions
10-10: LGTM!Also applies to: 939-939, 1003-1003
marm-mcp-server/docker-compose.yml (1)
8-8: LGTM!Also applies to: 21-21
marm-mcp-server/marm_mcp_server/__init__.py (1)
17-20: LGTM!marm-mcp-server/marm_mcp_server/config/settings.py (1)
182-182: LGTM!marm-mcp-server/marm_mcp_server/server.py (1)
8-8: LGTM!marm-mcp-server/pyproject.toml (1)
7-7: LGTM!marm-mcp-server/server.json (1)
6-6: LGTM!Also applies to: 20-25
| ### Fixed: Duplicate Detection Compared the Wrong Score | ||
|
|
||
| - `CONSOLIDATION_THRESHOLD` (default `0.92`) is documented as the meaning-similarity a memory needs before it is merged into an existing one. The check was comparing it against the final ranking score instead, which also folds in keyword relevance and a recency boost. A memory could therefore be merged away when its actual meaning-similarity was below the threshold, because keyword overlap and freshness had pushed its ranking score above it. | ||
| - Duplicate detection now compares the raw meaning-similarity, as documented. It also asks for the meaning-based search lane explicitly: syntax-heavy content such as config keys, file paths and commands is normally routed to exact keyword lookup, which produces no meaning-similarity at all, so duplicate detection could not have worked for it either way. And it looks at a handful of candidates rather than only the top-ranked one, because the closest memory by meaning is not always the one that ranks first. If no meaning-similarity is available, nothing is merged. |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟠 Major | ⚡ Quick win
Qualify “duplicate detection” as semantic duplicate detection.
Exact normalized-content deduplication still handles identical content, including syntax-heavy content. The current wording says duplicate detection cannot work when the semantic lane has no cosine value. State that semantic merging is skipped while exact deduplication remains available.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@CHANGELOG.md` at line 11, Update the changelog entry to call this “semantic
duplicate detection” or “semantic merging,” and clarify that semantic merging is
skipped when no meaning-similarity is available while exact normalized-content
deduplication still handles identical content, including syntax-heavy content.
Source: Path instructions
|
|
||
| These measurements use the fastembed-backed `jinaai/jina-embeddings-v2-small-en` encoder and a throwaway local SQLite database. Every timed path calls the shipped `MARMMemory` code, not a benchmark-local reimplementation. All numbers below come from a single run of [`scripts/benchmarking/performance/bench_hotpath.py`](scripts/benchmarking/performance/bench_hotpath.py) on local hardware; absolute milliseconds vary by machine, so treat the scaling shape as the signal. | ||
|
|
||
| Re-measured for v2.33.0. Latency rose and the hybrid speedup in section 4 fell substantially, both because `FTS_CANDIDATE_LIMIT` was raised from `50` to `200` in v2.32.0 to recover multi-hop accuracy. That was a deliberate trade of speed for correctness, and the older figures below were published before it. |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟠 Major | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf 'README candidate:\n'
fd -a 'README.md' . | sed 's#^\./##'
printf '\nRelevant README excerpts:\n'
if [ -f marm-mcp-server/README.md ]; then
nl -ba marm-mcp-server/README.md | sed -n '160,250p'
fi
printf '\nSearch for v2.33 references:\n'
rg -n 'v2\.33\.0|v2\.33\.1|Current \((.*?)v2\.33|Re-measured for v2\.33' marm-mcp-server/README.md || true
printf '\nDiff stat/name only:\n'
git diff --stat || true
git diff -- marm-mcp-server/README.md | sed -n '1,220p' || trueRepository: Lyellr88/marm-memory
Length of output: 540
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE=marm-mcp-server/README.md
printf 'README excerpts:\n'
sed -n '160,250p' "$FILE" | cat -n -ba
printf '\nSearch for v2.33 references in marm-mcp-server/README.md:\n'
rg -n 'v2\.33\.0|v2\.33\.1|Current \((.*?)v2\.33|Re-measured for v2\.33' "$FILE" || true
printf '\nDiff status and changed README hunks:\n'
git diff -- marm-mcp-server/README.md | sed -n '1,260p'Repository: Lyellr88/marm-memory
Length of output: 241
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE=marm-mcp-server/README.md
printf 'README excerpts:\n'
sed -n '160,250p' "$FILE"
printf '\nSearch for v2.33 references in marm-mcp-server/README.md:\n'
rg -n 'v2\.33\.0|v2\.33\.1|Current \((.*?)v2\.33|Re-measured for v2\.33' "$FILE" || true
printf '\nDiff status and changed README hunks:\n'
git diff -- marm-mcp-server/README.md | sed -n '1,260p'Repository: Lyellr88/marm-memory
Length of output: 10274
Resolve the v2.33 benchmark release label.
marm-mcp-server/README.md is for v2.33.1, but the benchmark section says Re-measured for v2.33.0 and labels the LoCoMo row Current (v2.33.0). If these numbers are from v2.33.0, label them as historical. Update both labels to v2.33.1 only after the benchmark run is from that release.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@marm-mcp-server/README.md` at line 182, Update the benchmark section’s
release labels to match the actual benchmark run: use v2.33.1 for both the
re-measurement note and the LoCoMo “Current” row only if the data was rerun
against v2.33.1; otherwise mark both references as historical v2.33.0 results.
|
|
||
| These measurements use the fastembed-backed `jinaai/jina-embeddings-v2-small-en` encoder and a throwaway local SQLite database. Every timed path calls the shipped `MARMMemory` code, not a benchmark-local reimplementation. All numbers below come from a single run of [`scripts/benchmarking/performance/bench_hotpath.py`](scripts/benchmarking/performance/bench_hotpath.py) on local hardware; absolute milliseconds vary by machine, so treat the scaling shape as the signal. | ||
|
|
||
| Re-measured for v2.33.0. Latency rose and the hybrid speedup in section 4 fell substantially, both because `FTS_CANDIDATE_LIMIT` was raised from `50` to `200` in v2.32.0 to recover multi-hop accuracy. That was a deliberate trade of speed for correctness, and the older figures below were published before it. |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Align benchmark version labels with the released code.
Both documents are labeled v2.33.1 but identify the benchmark results as current v2.33.0 measurements. Rerun the benchmark on v2.33.1, or label the results as a v2.33.0 baseline.
README.md#L180-L180: Update the benchmark version label and remove the current-release ambiguity.marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md#L154-L154: Apply the same version-label correction to the packaged documentation.
As per path instructions: this is a materially misleading documentation claim.
📍 Affects 2 files
README.md#L180-L180(this comment)marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md#L154-L154
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@README.md` at line 180, Update the benchmark version labels in README.md at
lines 180-180 and marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md
at lines 154-154 to identify the results as a v2.33.0 baseline, removing any
implication that they are current v2.33.1 measurements; keep both documents
consistent.
Source: Path instructions
| | Jina v2 Small (v2.29.0) | 53.0% | 43.4% | 47.6% | | ||
| | **Current (v2.33.0)** | **62.9 - 63.5%** | **53.1 - 53.5%** | **57.4 - 57.9%** | | ||
|
|
||
| The v2.33.0 row is a range, not a point, because it is two identically-configured runs. Repeat runs of this harness disagree by up to **0.56pp** (19 of 1,977 questions), so differences smaller than that are not meaningful. The cause is understood: the corpus is written minutes before the benchmark, so `TEMPORAL_WEIGHT` reorders near-ties as the store ages between runs. Quote the range rather than the better run. |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Align the benchmark run-count methodology.
The benchmark introduction claims one run, but the LoCoMo results use two runs. Scope the one-run statement to sections 1–4 and state the two-run method for LoCoMo.
README.md#L236-L236: Correct the run-count description in the main documentation.marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md#L210-L210: Apply the same correction to the packaged documentation.
As per path instructions: this is a materially incorrect methodology statement.
📍 Affects 2 files
README.md#L236-L236(this comment)marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md#L210-L210
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@README.md` at line 236, Correct the benchmark run-count methodology statement
in README.md at lines 236-236 and
marm-mcp-server/marm_mcp_server/resources/marm-docs/README.md at lines 210-210:
scope the one-run claim to sections 1–4 and explicitly state that LoCoMo results
use two identically configured runs, keeping both documentation copies
consistent.
Source: Path instructions
Addresses the three CodeRabbit findings on PR #120, all documentation: - Scope the consolidation entry to semantic merging. Layer 1 exact content-hash deduplication is untouched by the cosine fix and still merges identical content, including syntax-heavy content. The previous wording implied dedup stops entirely when no cosine is available. - Label the benchmark section v2.33.1. Both the LoCoMo accuracy run and the hot-path timings were measured on code carrying the temporal snapshot and consolidation fixes, so v2.33.0 was the wrong label rather than the numbers being stale. - Scope the run-count claim. Sections 1-4 are one bench_hotpath.py run; section 5 is a separate accuracy benchmark reporting two runs. Applied to the root docs and both packaged copies so all three agree. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
1 participant
v2.33.1 - Recall and Consolidation Correctness Fixes
This patch carries forward two independent correctness fixes found during the Phase 4 fusion bake-off. RRF itself was not merged; min-max remains the shipped fusion method.
CONSOLIDATION_THRESHOLDagainst raw cosine similarity not the blended keyword/semantic/temporal ranking score. It explicitly uses the semantic lane, checks multiple candidates, and declines to merge when cosine is unavailable. Fixes Hybrid recall similarity is RRF-scaled, breaking cosine-based consolidation threshold #113.Validation completed: 899 passed, 2 skipped; Ruff, MCP schema validation, version audit, and wheel/sdist build all pass.
Summary by CodeRabbit
Bug Fixes
Documentation
Release