- Compound entity chaining now stops one level deep. Previously "TDS
motorsports team history" would inflate the expected entity set with
"team" and "history", causing false-positive entity-preservation
penalties during GRPO. Now only {tds, motorsports} are detected.
- Add INTERIOR_FILLER_WORDS penalty (-3/line): lex lines containing
"overview" or "basics" absent from the original query are penalised.
Targets template-generator noise, e.g. "ancient overview rome timeline".
- Raise is_diverse threshold 2→3: requires 3 unique words between lex
lines before they count as diverse. Reduces reward for near-duplicate
pairs like "auth setup" / "auth configuration".
- Broaden quoted-phrase bonus: was gated on named entities existing;
now any multi-word query earns +3 for using quotes in lex lines.
Better incentivises BM25-aware syntax like "memory leak" python.
Fixes scoring noise identified while working on issue #247.
node-llama-cpp throws a hard error when any document + query + template
overhead exceeds the ranking context size. Truncate oversized documents
using the rerank model's tokenizer before passing them to rankAll().
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace ad-hoc JSON parsing with a strict Pydantic model
(TrainingExample with typed OutputPair). All data loading goes
through load_examples() which fails loudly on invalid data.
- Convert v3_structured.jsonl from "searches" to "output" format
- Rewrite all consumer scripts (prepare, validate, score, analyze)
to load through the Pydantic schema
- Prepared train/val files are ephemeral build artifacts
- Restore LFM2 and GEPA experiments under experiments/
- Add pydantic>=2.0 to dependencies
Training data:
- Expand lex phrases/negation examples from 12 to 74 with intent field
- Add 50 personal entity examples (meetings, emails, projects with names)
Reward function:
- Detect entities at position 0 (fixes "Bob asked about deploy")
- Per-entity coverage penalty: -20 per entity absent from all lex+vec
- Phrase quoting bonus: +3 when lex uses quotes for multi-word terms
- Expanded stopwords to reduce false positive entity detection
Eval queries: add 21 test queries for personal entities, quoted phrases,
and negation/disambiguation scenarios.
Remove one-off data generator/fix scripts, superseded data files (v2, v3
replaced by v3_structured), LFM2 experiment, GEPA directory, duplicate
job scripts, and historical docs. Clean up Justfile.
These are restored under experiments/ in a later commit.
Replaces the inner test script with an outer driver that runs individual
podman/docker commands against a pre-built image. Tests sqlite-vec
loading and store unit tests under both node and bun runtimes.
Supports --build (image only), --shell (interactive), and -- CMD
(arbitrary command) for debugging install issues in isolation.
The qmd bin was a custom bash script that discovered node via hardcoded
fallback paths (mise, asdf, nvm, homebrew). This was nonstandard and
caused ABI mismatches when installed via bun (native modules compiled
for bun but executed with node).
Now uses the standard npm bin convention: dist/qmd.js with a node
shebang, added by the build script. The isMain guard resolves symlinks
so it works when npm/bun create symlinked bin entries.
Also converts all dynamic require() calls in tests to ESM imports, and
adds container-based smoke tests (test/smoke-install.sh) that verify
install + run under both node and bun via mise in a Debian container.
The 'query document' is now a first-class concept in QMD: a structured
document with typed sub-queries that combine for best recall.
## Query types
- lex: BM25 keyword search with phrase and negation syntax
- vec: Semantic vector search (natural language questions)
- hyde: Hypothetical document (write the expected answer)
- expand: Auto-expand via local LLM (max 1, default for plain queries)
## Lex syntax
Full BM25 operator support:
"exact phrase" verbatim match, no prefix
-term exclude documents containing term
-"exact phrase" exclude documents containing phrase
Examples:
"C++ performance" optimization -sports -athlete
"connection pool" timeout -redis
"machine learning" -sports -athlete
## MCP tool description rewritten
The 'query' tool description now fully teaches AI agents the query
document format, lex syntax, and strategy for combining types.
Includes worked examples including intent-aware lex (C++ performance,
not sports) which is critical for disambiguation in dense corpora.
## Unit tests
11 new lex parser tests covering:
- plain terms, quoted phrases, negation, combined
- intent-aware disambiguation (performance -sports -athlete)
- only-negation returns null (FTS5 constraint)
- empty/whitespace handling
## Training data
12 new intent-aware examples for next model training round:
- Real technical topics with lex phrase+negation combinations
- Covers: C++ perf, Python memory, DB connections, rate limiting,
SQL optimization, ML overfitting, Docker, JWT, async/await,
git conflicts, Kubernetes, React state
- Each shows how context/intent shapes lex query construction
(e.g. performance with C++ context → -sports -athlete exclusions)
New collection subcommands:
- show <name> Show collection details
- update-cmd <name> [cmd] Set pre-update command (runs before indexing)
- include <name> Include in default queries (default)
- exclude <name> Exclude from default queries
Collections with includeByDefault=false are skipped unless
explicitly named with -c flag.
CLI improvements:
- 'qmd collection' shows help instead of error
- 'qmd collection list' shows [excluded] tag
- Better command descriptions and examples
Lex queries now support:
- "exact phrase" - quoted exact matching (no prefix)
- -term or -"phrase" - exclude from results
- term1 OR term2 - match either term
Semantic queries (vec/hyde) validate and reject these operators
with helpful error messages.
Examples:
performance -sports → matches "performance" excluding "sports"
"machine learning" → exact phrase match
auth OR authentication → matches either term
- structured_search now accepts collections[] for OR filtering
- Updated skill docs with detailed query writing guidance
- lex: 2-5 keywords, include synonyms, exact names
- vec: full natural language questions with context
- hyde: 50-100 word hypothetical answer passages
BREAKING CHANGE: MCP tools search, vector_search, deep_search removed.
Use structured_search with lex/vec/hyde queries instead.
- Remove search, vector_search, deep_search MCP tool registrations
- Update MCP instructions to focus on structured_search
- Update skill docs to reflect simplified API
- Rename test describes to reflect they test store functions
- CLI commands (qmd search, vsearch, query) unchanged for backwards compat
Lines prefixed with lex:, vec:, or hyde: route directly to
structured search, skipping automatic query expansion.
Examples:
qmd query 'lex: CAP theorem'
qmd query $'lex: keywords\nvec: natural language question'
qmd query $'lex: terms\nvec: question\nhyde: hypothetical answer...'
Plain queries (single line, no prefix) still use automatic expansion.
Multiple plain lines without prefixes error with helpful message.
This lets CLI users leverage the same structured search as MCP,
useful when piping from scripts or when you know exactly what
query variations you want.
- New MCP tool: structured_search - lets capable LLMs provide their own
lex/vec/hyde query variations instead of using local expansion model
- New REST endpoint: POST /search - same functionality without MCP protocol
- Updated skill docs to prioritize structured_search for LLM callers
- Added installation instructions for Claude Code, Desktop, and OpenClaw
Pipeline: lex→FTS, vec/hyde→batch embed, RRF fusion (first query 2x weight),
chunk + rerank, position-aware blending, dedup.
This is the recommended endpoint for capable LLMs - they generate better
query variations than the small local model, especially for domain-specific
or nuanced queries.
collections-config.test.ts set currentIndexName to "myindex" in its
last test but only restored env vars in afterEach — not the module
variable. Under bun test (single process), this leaked into mcp.test.ts,
causing it to look for myindex.yml instead of index.yml.
Fix: reset setConfigIndexName("index") in afterEach, and add defensive
reset in mcp.test.ts beforeAll.
node-gyp needs python3 to compile better-sqlite3, and on macOS it also
needs libtool from cctools to create static libraries. Without these,
`nix build` fails in the sandbox.
