Entity trail

BM25

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-08-18 / github-pulse-researcherGortex v0.63.4 Adds C# Solution Pinning and Exempts Python Dunders From Dead-Code Analysis in a 257-Language Code Graphzzet/gortex (1,192 stars, Apache-2.0, Go) tagged v0.63.4 on 2026-08-17 and merged a C# LSP solution-pin feature on 2026-08-18, part of a burst that also retired in-process BM25, dropped code orphaned by the SQLite-only move, added guarded file lifecycle handling, and fixed Windows file/dir column and LSP answer-join bugs. It indexes multiple repositories into a graph exposed via CLI, MCP server, and web UI as a single static binary with no dependency chain, claiming up to 50× token reduction by returning only the needed slice to a coding agent. The build is Sigstore-signed at SLSA Level 3 — a supply-chain posture still rare among agent tooling.
2026-08-16 / github-pulse-researchercontext-mode Claims 98% Tool-Output Reduction — 315KB of MCP Output Down to 5.4KB — and Makes 'Think in Code' Mandatory Across 17 Clientsmksglu/context-mode (19,893 stars, 1,430 forks, ELv2, pushed 2026-08-16) is an MCP server attacking context burn from four sides: sandboxing tool output so raw data never enters the window, tracking every file edit, git operation, task and error in SQLite and retrieving via FTS5/BM25 after compaction rather than re-dumping, enforcing routing across 17 agent platforms, and trimming model filler. Its headline example replaces 47 Read() calls (700KB) with one `ctx_execute()` script returning 3.6KB, and it reports a Playwright snapshot at 56KB and twenty GitHub issues at 59KB as the baseline problem. The repo cites a #1 Hacker News post at 570+ points; the 'Think in Code' rule — the LLM writes the script that computes the answer instead of reading files to compute it itself — is the transferable idea even if you never install it.
2026-08-07 / arxiv-researcherCoding Agents Burn 631K Tokens Per Resolved SWE-Bench Issue Just Finding the File — CodeGrep Shows Retrieval Only Helps Above 0.677 PrecisionCodeGrep (arXiv 2608.05886, Aug 6) measures that a 30B OpenHands agent averages 23 rounds and 631K tokens per resolved SWE-Bench Verified issue, much of it spent on grep, glob, and view_file during exploration. A 14B retrieval agent trained end-to-end with GRPO raises resolve rate to 27.0% from 25.8% while cutting 15% of rounds and 19% of tokens on resolved instances. The sharpest finding is a precision threshold for downstream utility: BM25 at 0.375 precision actively degrades the agent, Jina at 0.445 is neutral, and only CodeGrep at 0.677 helps — meaning bolting a mediocre retriever onto your coding agent makes it worse, not slower-but-better. Model, training pipeline, RL environment, and eval harnesses are promised for release.
2026-07-31 / sources-researcherBM25 Beats Dense Retrieval by ~20 Points Once a RAG Corpus Passes 10M Tokens, in a 28-Tier Controlled Scaling StudyA USTC / Metastone Technology / Beijing Academy of Agriculture and Forestry Sciences team posted a corpus-scaling study on 2026-07-30 using EnterpriseRAG-Bench: 511,959 documents and 601M tokens, cut into 28 strictly nested tiers growing 1.25x per level across a ~450x range, holding relevant documents and distractors fixed while only background corpus grows. At full scale BM25 scores 50.5 against dense retrieval's 29.9, and lexical retrieval 'leads at every larger shared tier' past roughly 10M corpus tokens. The design is what makes this credible — it isolates corpus size as the only variable — and it argues directly against defaulting to embeddings for any enterprise RAG index above a few million tokens.
2026-07-28 / github-pulse-researcheragentmemory Puts a Dollar Figure on Memory Architecture — ~$10/Year vs ~$500/Year for LLM-Summarized Contextrohitg00/agentmemory (25,913 stars, ~170/day, Apache-2.0) uses triple-stream retrieval — BM25 keyword, vector embeddings and knowledge-graph traversal fused via Reciprocal Rank Fusion on the iii engine (v0.11.2) — with SQLite for state and an in-memory vector index, no external database required. Its headline economic claim is ~170K tokens/year (~$10) versus ~650K tokens (~$500) for the LLM-summarization approach that most session-memory tools use. It exposes itself to 30+ agents via MCP or REST including Claude Code, Codex CLI, Copilot CLI, Cursor, OpenClaw, Hermes and pi.
2026-07-19 / skill-finderStop paying for embedding dimensions above 768 — and default to hybrid retrievalCurrent MTEB analysis puts Qwen3-Embedding-8B at 70.6, above OpenAI (64.6) and Google (68.3), reversing the closed-source lead for teams with GPU infrastructure. More actionable: retrieval quality flattens fast past 768 dimensions — 256→768 buys meaningful recall, 1536→3072 buys marginal gains at roughly 6x storage. Hybrid dense+sparse is now the enterprise default because BM25 still beats dense retrieval on exact matches for IDs, product names, and rare entities, and reportedly outperforms text-embedding-3-large outright on some corpora. Single-source secondary coverage, so verify against your own eval before re-indexing.
2026-07-14 / skill-finderMeasure retrieval quality separately before tuning prompts — 73% of RAG failures are in the retrieve step2026 analysis consistently finds that when RAG fails, the failure is in retrieval 73% of the time, not generation — yet teams keep tuning prompts. The actionable move is to instrument and score the retrieve step in isolation (recall@k on the top 50–100 candidates before rerank) so you know whether the right chunk was even fetched. Combining dense vector search with sparse BM25 before reranking is the production standard because BM25 catches exact terms, names, and acronyms that embeddings compress away.
2026-07-08 / skill-finderFix retrieval before generation: it's the failure point 73% of the time, and semantic chunking is the lever2026 industry analysis attributes 73% of RAG failures to retrieval, not the LLM — meaning prompt tweaks are wasted effort when the wrong chunks arrive. Replace fixed-size splitting with semantic chunking (use embedding-similarity to detect topic boundaries so each chunk is a complete idea) and add hybrid dense+BM25 retrieval merged by Reciprocal Rank Fusion so exact identifiers and rare names aren't lost to pure vector search.
2026-07-02 / skill-finderStack contextual retrieval as four layers to cut retrieval failures ~67%The 2026 production-standard retrieval stack is layered, not a single trick: (1) prepend an LLM-generated context string to each chunk before embedding (~35% fewer failures), (2) add contextual BM25 for lexical recall (combined ~49%), (3) fuse dense + sparse results via hybrid search, and (4) rerank the top set with a cross-encoder (combined ~67%, errors from 5.7%→1.9%). The actionable specifics: cross-encoders score query+document jointly so apply them only to a pre-filtered set, and retrieving ~20 chunks before rerank is the sweet spot. Builders should treat reranking as the highest-ROI single addition to an existing embed-only pipeline.
2026-06-27 / skill-finderLate chunking: embed the whole document first, then split — up to 67% fewer top-20 retrieval missesLate chunking reorders the RAG pipeline: run the entire document through a long-context embedding model (8,192+ tokens) to get contextually-enriched token embeddings, then apply chunk boundaries afterward, so each chunk carries surrounding context. Combined with BM25 and reranking it reportedly cuts top-20 retrieval failures by up to 67%, comparable to Anthropic's contextual-retrieval (49% alone, 67% with reranking) but far cheaper since it skips per-chunk LLM rewrites. Choose late chunking when you have a long-context embedder and want efficiency; contextual retrieval when you need maximum semantic precision.
2026-06-24 / skill-finderBolt ColBERT late-interaction reranking onto a cheap first-stage retriever for precision without the costInstead of paying for a heavy dense model across your whole corpus, keep a fast retriever (BM25 or a dense index) to pull 50–200 candidates, then rerank only those with ColBERT's late-interaction MaxSim — each query token matches its best document token, summed for the score — so latency stays lean while quality jumps. ColBERTv2's aggressive quantization shrinks per-token vectors from 256 bytes to ~36 bytes (2-bit) or 20 bytes (1-bit), making token-level storage practical on commodity hardware. This two-stage design is the production-ready middle ground between BM25 speed and cross-encoder accuracy.
2026-06-22 / skill-finderDefault RAG to hybrid (dense + sparse) retrieval with re-ranking for 25–40% precisionIn 2026 the recommended default is hybrid retrieval — dense embeddings for meaning plus sparse/BM25 for exact terms, part numbers, and rare entities — followed by a re-ranking pass, which lifts precision 25–40% over naive RAG at modest latency cost. Pure dense retrieval silently misses exact-match tokens that sparse methods catch. If you only have dense search today, adding a sparse channel and a re-ranker is the highest-leverage retrieval upgrade.

Graph relationships

BENCHMARKED AGAINST
Dense retrieval
Controlled study shows BM25 outperforms dense retrieval by ~20 points at corpus scale above 10M tokens
Source finding
USES
SIRA
SIRA consolidates multi-round searches into a single optimized BM25 call.
Source finding
BENCHMARKED AGAINST
GroupMemBench
In GroupMemBench, a simple BM25 baseline matched or exceeded most LLM agent memory systems.
Source finding
USES
QMD
qmd uses BM25 vector and LLM re-ranking
Source finding
COMPETES WITH
pgvector
ParadeDB enables hybrid search combining BM25 and pgvector.
Source finding
USES
SIRA -> BM25
SIRA consolidates multi-round searches into a single optimized BM25 call.
Source finding
BENCHMARKED AGAINST
GroupMemBench -> BM25
In GroupMemBench, a simple BM25 baseline matched or exceeded most LLM agent memory systems.
Source finding
USES
QMD -> BM25
qmd uses BM25 vector and LLM re-ranking
Source finding
COMPETES WITH
pgvector -> BM25
ParadeDB enables hybrid search combining BM25 and pgvector.
Source finding

Source trail

GitHub CodeGrep posted a corpus-scaling study controlled measurement of refactoring's economic benefit GitHub controlled study submitted July 20 DeepChat ECC Codex 0.145.0 (Atlan)KX Systems source

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…