cmd/

Details

Valid go.mod file
Redistributable license
Tagged version
Stable version
Learn more about best practices

Repository

Links

Open Source Insights

README ¶

Commands

One of these is the product. The rest are measurement tools kept in the repository because the numbers in BENCHMARK.md and docs/research are only worth reading if the thing that produced them is readable too.

The product


`contextmaxxer`	the binary a user installs: index, warmup, init, query, mcp, hook, feedback

Everything below builds only when asked and ships in no release archive.

Benchmark runners


`exploreprobe`	SWE-Explore under the paper's protocol (B = 500 budgeted scoring)
`corebench`	CORE-Bench, embedder and seed-fusion stages
`mcpeval`	the evaluation corpus driven through a real MCP server over JSON-RPC, to catch a tuned configuration that never reached the served path
`eval`	the small public regression suite
`goldengate`	golden-output guard

Probes that answer one question each


`chainprobe`	how much of a call chain arrives without a second search, and what the same hops cost as a text search
`regionprobe`	where a missed gold region sits relative to what was returned
`reachprobe`	why a gold file is never retrieved: vocabulary, weighting, or indexer coverage
`rspbreak`	what the response is spending its tokens on
`negprobe`	behaviour on queries with no right answer
`selfsweep`	self-evaluation across configurations
`giteval`	cases derived from repository history
`deepprobe`, `lateexp`	deep reranking and late interaction, both measured and parked
`embprobe`	embedding runtime smoke test
`idxstats`, `diag`	index and environment inspection
`confcal`	confidence calibration
`soak`	long-running stability
`ftdata`	training-pair extraction for the fine-tuning experiments
`askctx`	one-shot query helper

Several of these need corpora that are not in this repository — SWE-Explore is CC-BY-NC-ND and CORE-Bench is CC-BY-NC-SA, so the numbers are published and the data is not. Those runners will build and refuse to run without it.

Directories ¶

?	: This menu
/	: Search site
f or F	: Jump to
y or Y	: Canonical URL

Path	Synopsis
askctx Command askctx runs one find_context against a chosen binary and prints the response an agent host would receive.	Command askctx runs one find_context against a chosen binary and prints the response an agent host would receive.
chainprobe Command chainprobe measures whether the call graph makes a chain followable, without an agent in the loop.	Command chainprobe measures whether the call graph makes a chain followable, without an agent in the loop.
confcal Command confcal calibrates the confidence gate.	Command confcal calibrates the confidence gate.
contextmaxxer
corebench Command corebench evaluates an embedding model on a downloaded subset of CORE-Bench (arXiv:2606.11864, HF: zhangfw123/CORE-Bench) in BEIR format: each repo dir holds corpus.jsonl, queries.jsonl and qrels/test.tsv.	Command corebench evaluates an embedding model on a downloaded subset of CORE-Bench (arXiv:2606.11864, HF: zhangfw123/CORE-Bench) in BEIR format: each repo dir holds corpus.jsonl, queries.jsonl and qrels/test.tsv.
deepprobe Command deepprobe measures what ranking metrics structurally cannot see: whether the answer is inside the excerpt the response actually shows.	Command deepprobe measures what ranking metrics structurally cannot see: whether the answer is inside the excerpt the response actually shows.
diag Diagnostic tool: for selected (project, query_id) pairs, runs retrieval in both vector-only and hybrid modes with top-30 + features, and prints a compact side-by-side comparison highlighting the expected symbol's rank.	Diagnostic tool: for selected (project, query_id) pairs, runs retrieval in both vector-only and hybrid modes with top-30 + features, and prints a compact side-by-side comparison highlighting the expected symbol's rank.
embprobe Command embprobe is a 30-second semantic sanity check for an embedding model in OUR runtime (tokenizer + pooling + prefixes + ORT): it embeds a few fixed code snippets and two NL queries and prints the cosine matrix.	Command embprobe is a 30-second semantic sanity check for an embedding model in OUR runtime (tokenizer + pooling + prefixes + ORT): it embeds a few fixed code snippets and two NL queries and prints the cosine matrix.
eval
exploreprobe Command exploreprobe scores the served retrieval against SWE-Explore, the first external benchmark that grades the WHOLE pipeline rather than its entrance.	Command exploreprobe scores the served retrieval against SWE-Explore, the first external benchmark that grades the WHOLE pipeline rather than its entrance.
ftdata Command ftdata turns CORE-Bench-style BEIR data into embedder fine-tuning triplets: {"query", "pos": [...], "neg": [...]} JSONL, one line per query.	Command ftdata turns CORE-Bench-style BEIR data into embedder fine-tuning triplets: {"query", "pos": [...], "neg": [...]} JSONL, one line per query.
giteval Command giteval scores the served stack against labels nobody on this project authored: each case is a real commit — query = the commit subject, gold = the symbols whose enclosing-function hunk headers appear in that commit's diff, resolved against the index.	Command giteval scores the served stack against labels nobody on this project authored: each case is a real commit — query = the commit subject, gold = the symbols whose enclosing-function hunk headers appear in that commit's diff, resolved against the index.
goldengate Command goldengate is THE regression gate: one command, one report, always the same definition of "our numbers", measured through the shipped binary.	Command goldengate is THE regression gate: one command, one report, always the same definition of "our numbers", measured through the shipped binary.
idxstats Command idxstats prints symbol and edge counts for one or more index DBs.	Command idxstats prints symbol and edge counts for one or more index DBs.
lateexp Command lateexp is an offline experiment: does late-interaction (ColBERT-style MaxSim over per-token vectors) improve paraphrastic SEED recall over the production single-vector mean-pool? It scores the WHOLE corpus by both methods (not just a re-rank of the seed set) because the hypothesis is that MaxSim catches paraphrases the single vector drops at the seed stage — a re-rank of an already-missed seed could never show that.	Command lateexp is an offline experiment: does late-interaction (ColBERT-style MaxSim over per-token vectors) improve paraphrastic SEED recall over the production single-vector mean-pool? It scores the WHOLE corpus by both methods (not just a re-rank of the seed set) because the hypothesis is that MaxSim catches paraphrases the single vector drops at the seed stage — a re-rank of an already-missed seed could never show that.
mcpeval Command mcpeval drives the gen-eval corpus through the REAL served stack — it spawns the shipped binary's `mcp` subcommand per project and talks JSON-RPC over stdio, exactly like an agent host does.	Command mcpeval drives the gen-eval corpus through the REAL served stack — it spawns the shipped binary's `mcp` subcommand per project and talks JSON-RPC over stdio, exactly like an agent host does.
negprobe Command negprobe measures the false-confidence rate: queries about plausible concepts that do NOT exist in the indexed repo.	Command negprobe measures the false-confidence rate: queries about plausible concepts that do NOT exist in the indexed repo.
reachprobe Command reachprobe asks why a gold file is never retrieved.	Command reachprobe asks why a gold file is never retrieved.
regionprobe Command regionprobe asks where the gold sits relative to what we returned.	Command regionprobe asks where the gold sits relative to what we returned.
rspbreak Command rspbreak prices a find_context response by section.	Command rspbreak prices a find_context response by section.
selfsweep Command selfsweep is a label-free ranking radar for ANY indexed repo: every documented symbol must be findable by its own docstring.	Command selfsweep is a label-free ranking radar for ANY indexed repo: every documented symbol must be findable by its own docstring.
soak Command soak asks one long-lived server the same questions over and over and checks that the answers do not drift.	Command soak asks one long-lived server the same questions over and over and checks that the answers do not drift.