Every commit scored for change-risk against this repo's own history, so 'elevated' means elevated here rather than on some global curve.
Needs review
201 commits sit in this repo's top risk tercile, which is 34% of the 597scored. The cut is drawn against this codebase's own history rather than a global curve, so a quiet repo still fills its top band, and here it starts at 7.6 out of 10. What pushes a commit up is size and spread together: a large change confined to one area scores below a smaller one scattered across a dozen files.
Commit categories over time, read off the subject line. Fixes carry the accent because that is the series this chart exists to show.
Consistently other-driven across its history.
Ranked by change-risk, highest first. Priority is a tercile of this repo's own distribution, so a quiet repo still fills its top band.
| # | Commit | Author | When | Lines | Risk | Top driver |
|---|---|---|---|---|---|---|
| 1 | d2864cfefeat(benchmarks): add LongMemEval benchmark + BEAM 10M results | buildingjoshbetter | 3mo ago | +91.0K -16 | 100%Elevated | more lines added than baseline |
| 2 | b4a9981ddocs: update BEAM-1M to 3-run mean 76.6% (up from 75.0% single run) | buildingjoshbetter | 3mo ago | +25.3K -14 | 100%Elevated | more lines added than baseline |
| 3 | 2b834d55fix(benchmarks): update all docs and results to v0.6.0 3-run validated scores | buildingjoshbetter | 3mo ago | +210.3K -50.2K | 100%Elevated | more lines added than baseline |
| 4 | 3adc7f7eNeuromem v0.2.0 — Local-first memory for AI agents | buildingjoshbetterclaudeassisted | 4mo ago | +310.4K -0 | 100%Elevated | more lines added than baseline |
| 5 | 178962acdocs: overhaul README with new banner, terminal demo, comparison table, and 19 translations | buildingjoshbetter | 1mo ago | +6.5K -238 | 99%Elevated | more lines added than baseline |
| 6 | d6de715cfix(benchmarks): add Base run1 (91.8%) completing 3-run set, add BEAM 10M script | buildingjoshbetter | 3mo ago | +25.1K -2 | 99%Elevated | more lines added than baseline |
| 7 | e33a0e81feat(instrumentation): opt-in SQLite telemetry overlay | buildingjoshbetter | 1mo ago | +1.1K -0 | 99%Elevated | more lines added than baseline |
| 8 | 0883bee5feat(gate): encoding-specific salience scorer (closes #108) | buildingjoshbetter | 3mo ago | +1.1K -26 | 99%Elevated | more lines added than baseline |
| 9 | cdd84ffdfeat: universal hook framework for multi-CLI support (#186) | buildingjoshbetter | 2mo ago | +900 -0 | 98%Elevated | more lines added than baseline |
| 10 | 7f172ee2fix(observability): replace silent-swallow anti-patterns in mcp_server.py (closes #8, #9, #28, #29) | buildingjoshbetter | 3mo ago | +716 -92 | 98%Elevated | more lines added than baseline |
| 11 | 21a121e5MCP server: parallel search, tiered rerankers, background preloading | buildingjoshbetter | 3mo ago | +843 -38 | 98%Elevated | more lines added than baseline |
| 12 | 65fb8b76fix: dedup × gate coherence — shared markers, category plumbing, lock + txn hygiene (#649) (#661) | NOTCAVEMAN | 1mo ago | +685 -146 | 98%Elevated | more lines added than baseline |
| 13 | d86b2b3aperf: compute detect_contradictions and build_structured_facts outside transaction (#591) (#609) | NOTCAVEMAN | 1mo ago | +636 -132 | 98%Elevated | more lines added than baseline |
| 14 | c2312cb4fix(adapters): rewrite Hermes and OpenClaw adapters from actual source code | buildingjoshbetter | 2mo ago | +898 -175 | 98%Elevated | more lines added than baseline |
| 15 | 98c4eaeddocs: comprehensive API documentation (#212) | buildingjoshbetter | 2mo ago | +703 -1 | 98%Elevated | more lines added than baseline |
| 16 | 70de5293feat(l0): char-n-gram style vectors replace keyword extraction (MEMORIST-L0) | buildingjoshbetter | 3mo ago | +848 -53 | 98%Elevated | more lines added than baseline |
| 17 | 52284a67fix: model server lifecycle — exclusive bind lock, throttler TOCTOU, server-side deadline, protocol handshake (#646) (#662) | NOTCAVEMAN | 1mo ago | +639 -30 | 96%Elevated | more lines added than baseline |
| 18 | 0e5256cefix: backlog drain + maintenance hardening — corrupt-marker quarantine, claim races, watermark, inline fallback (#644) (#660) | NOTCAVEMAN | 1mo ago | +652 -30 | 96%Elevated | more lines added than baseline |
| 19 | 0d082905fix: declare cosine distance on vec0 tables + rebuild migration (#631) (#658) | NOTCAVEMAN | 1mo ago | +528 -16 | 96%Elevated | more lines added than baseline |
| 20 | a06bc8cefeat: add memory intensity preference (search_intensity, store_intensity) (#396) (#628) | NOTCAVEMAN | 1mo ago | +713 -4 | 96%Elevated | more lines added than baseline |
| 21 | 99265a58feat: Antigravity MCP adapter, Groq provider, DeepSearch model selection (#625) | NOTCAVEMAN | 1mo ago | +703 -7 | 96%Elevated | more lines added than baseline |
| 22 | 674f77d4fix: widen contradiction detection + supersede stale facts (#580) (#610) | NOTCAVEMAN | 1mo ago | +619 -5 | 96%Elevated | more lines added than baseline |
| 23 | 45370ebctest: issue #577 regression suite | buildingjoshbetter | 1mo ago | +579 -0 | 96%Elevated | more lines added than baseline |
| 24 | 2b8f180dtest: issue #589 regression suite | buildingjoshbetter | 1mo ago | +510 -0 | 96%Elevated | more lines added than baseline |
| 25 | eb1fe3e7refactor: extract 3 modules from engine.py god class (#137) (#553) | NOTCAVEMAN | 1mo ago | +540 -537 | 96%Elevated | more lines added than baseline |
| 26 | f9df0ee7phase 2: create tier_switch module with DynamicThrottler and VectorCacheRegistry | buildingjoshbetter | 2mo ago | +521 -0 | 96%Elevated | more lines added than baseline |
| 27 | 879fbe64feat: dynamic spawn cap with correct per-tier metrics and 8-bug fix | buildingjoshbetter | 2mo ago | +518 -61 | 96%Elevated | more lines added than baseline |
| 28 | c387cf05fix: flock-based spawn cap, drain race, and 6 hunter-v5 issues | buildingjoshbetter | 2mo ago | +642 -189 | 96%Elevated | more lines added than baseline |
| 29 | 656a1b1cfeat: Cursor integration — MCP server config + lifecycle hooks (#234) | buildingjoshbetter | 2mo ago | +763 -19 | 96%Elevated | more lines added than baseline |
| 30 | 96c0281ffeat: Codex CLI integration — MCP config + lifecycle hooks (#182) | buildingjoshbetter | 2mo ago | +751 -19 | 96%Elevated | more lines added than baseline |
| 31 | d5ef3ba8docs: multi-CLI setup guide and compatibility matrix (#187) | buildingjoshbetter | 2mo ago | +403 -0 | 96%Elevated | more lines added than baseline |
| 32 | 686357f7license: switch from Apache-2.0 to AGPL-3.0 | buildingjoshbetter | 3mo ago | +664 -193 | 96%Elevated | more lines added than baseline |
| 33 | 28c48f1dfeat(bench): paper-aligned Edge/Base/Pro bench scripts (Qwen3 256d + gte-reranker-modernbert) | buildingjoshbetter | 3mo ago | +479 -39 | 96%Elevated | more lines added than baseline |
| 34 | f62773a1fix: DB corruption UX — backup rotation/restore, quick_check, actionable WAL/permission errors (#650) (#672) | NOTCAVEMAN | 1mo ago | +538 -90 | 94%Elevated | more lines added than baseline |
| 35 | 09ec22c9fix: installer + adapter honesty — hook-entry dedup, alwaysLoad target + atomic writes, capability-parameterized prompts, antigravity verify (#651) | NOTCAVEMAN | 1mo ago | +520 -34 | 94%Elevated | more lines added than baseline |
| 36 | 8c6f9ef0fix: intensity validation fail-closed + recall flow coherence (#636) (#666) | NOTCAVEMAN | 1mo ago | +495 -48 | 94%Elevated | more lines added than baseline |
| 37 | 46335ecdfix: vector rebuild durability — in-progress marker, no destructive pre-wipe, metadata atomicity, sep-vector batch commits (#647) (#663) | NOTCAVEMAN | 1mo ago | +504 -15 | 94%Elevated | more lines added than baseline |
| 38 | d98e4807test: regression tests for Windows TCP socket fallback (#620) | NOTCAVEMAN | 1mo ago | +433 -0 | 94%Elevated | more lines added than baseline |
| 39 | 9b1753a9feat(custom-tier): centralize tier configs and add custom presets | buildingjoshbetter | 1mo ago | +511 -51 | 94%Elevated | more lines added than baseline |
| 40 | 9cf67a46fix(consolidation,mcp): shrink summary write-lock window + self-only orphan watcher | buildingjoshbetter | 2mo ago | +345 -36 | 94%Elevated | more lines added than baseline |
| 41 | e2cd8a7cfeat: shared model server for cross-process memory deduplication (#333, #335) | buildingjoshbetter | 2mo ago | +512 -6 | 94%Elevated | more lines added than baseline |
| 42 | f3b3ce20feat: Gemini CLI integration — MCP server config + lifecycle hooks (#233) | buildingjoshbetter | 2mo ago | +647 -19 | 94%Elevated | more lines added than baseline |
| 43 | 6abe4bd2feat(retrieval): wire L5 surprise signal into post-rerank boost (opt-in, α=0 default) | buildingjoshbetter | 3mo ago | +433 -3 | 94%Elevated | more lines added than baseline |
| 44 | 53b761baRebrand Neuromem to TrueMemory across the entire repo | buildingjoshbetter | 3mo ago | +703 -703 | 94%Elevated | more lines added than baseline |
| 45 | b9f6bcf0fix: recall cache contract — key completeness, invalidate on mutation, no negative caching (#645) (#678) | NOTCAVEMAN | 1mo ago | +521 -46 | 91%Elevated | more lines added than baseline |
| 46 | 9f1b3fe5fix: vec compatibility surfacing — dim check on open, propagate migration error, gate clustering/HyDE, tier-switch guards (#648) (#676) | NOTCAVEMAN | 1mo ago | +401 -40 | 91%Elevated | more lines added than baseline |
| 47 | bc9e5350fix: per-exchange store coordination — route through pipeline dedup; tighten _STORABLE_RE (#635) (#674) | NOTCAVEMAN | 1mo ago | +400 -24 | 91%Elevated | more lines added than baseline |
| 48 | 9d89b6fcfix: atomic + cross-process-locked config writes (#641) (#670) | NOTCAVEMAN | 1mo ago | +420 -53 | 91%Elevated | more lines added than baseline |
| 49 | b998cde2fix: heuristic extractor — role-scope, IGNORECASE, smart-quotes, 3 categories, validation (#611) | NOTCAVEMAN | 1mo ago | +369 -42 | 91%Elevated | more lines added than baseline |
| 50 | 21371693ChatGPT MCP Addition | Naresh Karthigeyan | 1mo ago | +379 -27 | 91%Elevated | more lines added than baseline |
Two views of the same model: where the cuts fall, and what commit shape lands you above them.
Every scored commit, binned on the raw 0 to 10 score rather than the percentile. Percentile ranks are uniform by construction, so that axis has no shape to draw. The dashed lines are the tercile cuts behind each row's priority pill.
The 200 most recent commits, on their own recency sample rather than the feed above: that defaults to risk-sorted, so reusing it would plot only the top tercile and call it the spread. Big and scattered is what the model penalises. Click a dot to open it.
Repowise tracks change history across 278 files in buildingjoshbetter/truememory. In the last 90 days 278 files were touched, 1,363 times in total, most often mcp_server.py. Every commit is scored for change risk from its size, spread and the history of the files it touches.
