feat: facts audit/extract/crm shebang wrappers (D14) (#19)

Python stays the implementation. Commands are bin/facts/{audit,extract,crm}.go
like postgres/query.go.

AGENTS.md

@@ -77,12 +77,14 @@ bin/brain/index.go --rebuild                                            # rebuil
 ## Tools
 
 ```bash
-bin/facts/audit ["self"|"facts"|"info"|"stale"]   # 2-source + staleness gate
+bin/facts/audit.go ["self"|"facts"|"info"|"stale"]  # 2-source + staleness gate
-bin/facts/crm [--dry-run]                         # proof person↔company/company↔project (ooCRM × corpus SoT)
+bin/facts/crm.go [--dry-run]                       # proof person↔company/company↔project (ooCRM × corpus SoT)
 bin/kb/search "query" [--repo X]                  # deprecated wrapper → bin/brain/search.go
 bin/brain/search.go "query" [--root facts|info]   # deduction search → YAML
 bin/brain/search.go "query" --no-web              # local graph only
-bin/brain/get.go <id> [--body]
+bin/brain/get.go <id> [--body] [--json]          # Go read; Python bin/kb/get CI fallback
+bin/brain/stats.go [--json]
+bin/brain/eval.go [--json]                       # recall@5; questions in internal/brain/rank
 bin/markdown/import.go [dir]                      # mistune leaves → YAML
 bin/git/import.go [REPO] [--json] [--limit N]     # go-git history → commit leafs
 bin/web/search.go "query" [--json]                # SearXNG; throttled ≠ absence

PLAN.md

@@ -29,7 +29,7 @@ detective method: **a fact needs ≥2 independent sources or it is
 | D3 | web search | Go client `bin/web/search.go` (`internal/websearch`). SearXNG URL is config (`BRAIN_SEARCH_URL`). Optional Compose profile `searxng` (sanitized settings). Do not run a second copy on a host that already has one. Empty/`throttled` ≠ “nothing exists”. |
 | D4 | embeddings | **model2vec** `minishlab/potion-multilingual-128M` instead of embeddinggemma. |
 | D5 | parser | **mistune** for MD → leaf extraction (duckdb-md documented as future optional SQL/export layer, not v1). |
-| D6 | graph engine | **LadybugDB**. Go is the service (`bin/brain/search.go`, `bin/brain/serve.go` in-process, `internal/brain`); Python remains for index/write until the Go write path is safe. |
+| D6 | graph engine | **LadybugDB**. Go is the service (`bin/brain/search.go`, `bin/brain/serve.go` in-process, `internal/brain`). Read path (`get.go` / `stats.go` / `eval.go`) is Go + cgo. Python `bin/kb/{get,stats,eval}` is the CI fallback (GitHub runners have no ladybug cgo). Index/write stays Python until the Go write path is safe. |
 | D7 | db access | `db-yaml`/`psql-yq`-style, read-only, YAML out. OnlyOffice Postgres via SSH tunnel (`127.0.0.1:5433`). |
 | D8 | evidence | detective method: ≥2 independent sources or `(not confirmed)`. Auto-pair docker ps × compose × ssh-config × docs. |
 | D9 | facts/goal model | Who / What / How / Where / When + evidence + confidence on every edge. |
@@ -52,11 +52,11 @@ detective method: **a fact needs ≥2 independent sources or it is
   docs/                     published docs (this conversation → docs/ as md)
   skills/                   in-project skills (web-search, db-yaml, brain, diataxis-docs)
   bin/
-    facts/extract           auto-pair 2 sources → lexicon yaml + graph
+    facts/extract.go audit.go crm.go  # D14 shebang; Python implementation
-    facts/audit             ["self"|"facts"|"info"|"stale"] 2-source + staleness gate
     kb/index                Python write path (called by bin/brain/index.go)
     brain/index.go          rebuild FTS + HNSW (incl. --with-mail)
-    brain/get.go stats.go eval.go watch.go
+    brain/get.go stats.go eval.go  # Go read (cgo); Python bin/kb/* CI fallback
+    brain/watch.go
     brain/search.go         deduction: facts → info → web-search
     brain/serve.go          HTTP API in-process (internal/httpapi + internal/brain)
     mail/import.go          JSON → markdown (no brain write)
@@ -132,8 +132,10 @@ Common props on every node/edge: `root`, `confidence`, `evidence[]`, `how`,
 1. go vet + go test ./... (root module; packages without ladybug cgo)
 2. `go test ./internal/brain/rank` (cgo-free ranking + flag parser)
 3. python -m unittest discover -s bin/tools (includes published-docs SoT)
-4. bin/facts/audit self  (lexicon internal consistency)
+4. `bin/facts/audit self` (lexicon internal consistency; `bin/facts/audit.go` is the D14 wrapper)
-5. bin/brain/eval.go       (recall@5 ≥ 0.95, gates index regressions)
+5. `bin/kb/eval` (recall@5 ≥ 0.95). Local SoT is `bin/brain/eval.go`; CI uses
+   the Python twin until the runner has ladybug cgo. Questions live in
+   `internal/brain/rank`.
 6. md-docs build/lint if docs tooling arrives.
 
 Feedback loop: every commit → PR → CI → green/gate → merge. Same discipline as

README.md

@@ -28,8 +28,8 @@ graph TB
     end
 
     subgraph dph["2dph tools"]
-        EX["bin/facts/extract<br/>2-source pairing"]
+        EX["bin/facts/extract.go<br/>2-source pairing"]
-        AU["bin/facts/audit<br/>confidence + staleness"]
+        AU["bin/facts/audit.go<br/>confidence + staleness"]
         IDX["bin/brain/index.go<br/>chunk + embed"]
         MD["bin/markdown/import.go<br/>mistune leaves"]
         SR["bin/brain/search.go<br/>deduction"]
@@ -126,7 +126,8 @@ bin/brain/search.go "invoice from last week"                            # same s
 
 - **LadybugDB** — single `var/kb.lbug`, Cypher property graph, HNSW + BM25
   in one engine, embedded (no server), ACID, read-only-safe for concurrent
-  readers. **Never `DROP INDEX` FTS/VECTOR** on Ladybug 0.19: DROP leaves
+  readers. Read tools (`get` / `stats` / `eval`) are Go + cgo; Python
+  `bin/kb/{get,stats,eval}` is the CI fallback. **Never `DROP INDEX` FTS/VECTOR** on Ladybug 0.19: DROP leaves
   ghost catalog tables (`_0_Leaf_vec_UPPER`) so recreate fails while
   `SHOW_INDEXES` omits HNSW. Fresh indexes = delete `var/kb.lbug` +
   `bin/brain/index.go --rebuild`. Use `ensure_indexes()` after upserts.
@@ -147,7 +148,7 @@ machines. Tests gate every commit. HTTP: `bin/brain/serve.go` calls
 ```bash
 uv venv .venv                                  # Python 3.12, uv-managed
 uv pip install -r requirements.lock.txt        # pinned toolchain
-bin/facts/audit self                           # lexicon consistency gate
+bin/facts/audit.go self                        # lexicon consistency gate
 go test ./... && python -m unittest discover -s bin/tools -t .
 ```
 

bin/brain/eval.go

@@ -1,20 +1,22 @@
-//usr/bin/env go run -tags=brain_eval "$0" "$@"; exit
+//usr/bin/env go run -tags=system_ladybug,brain_eval "$0" "$@"; exit
-//go:build brain_eval
+//go:build cgo && system_ladybug && brain_eval
 //
 // bin/brain/eval.go - recall@5 gate.
 //
 //	./bin/brain/eval.go
 //	./bin/brain/eval.go --json
 //
+// Needs CGO + libladybug. Python bin/kb/eval is the CI fallback (no cgo).
+// Control questions live in internal/brain/rank (cgo-free).
 // NOTE: never run `gofmt -w` on this file — it breaks the shebang.
 package main
 
 import (
 	"os"
 
-	"github.com/eSlider/2dph/internal/cmdbin"
+	"github.com/eSlider/2dph/internal/brain"
 )
 
 func main() {
-	os.Exit(cmdbin.ExecFile("bin/kb/eval", os.Args[1:]))
+	os.Exit(brain.MainEval(os.Args[1:]))
 }

bin/brain/get.go

@@ -1,20 +1,22 @@
-//usr/bin/env go run -tags=brain_get "$0" "$@"; exit
+//usr/bin/env go run -tags=system_ladybug,brain_get "$0" "$@"; exit
-//go:build brain_get
+//go:build cgo && system_ladybug && brain_get
 //
 // bin/brain/get.go - read one leaf by id.
 //
 //	./bin/brain/get.go <id>
 //	./bin/brain/get.go <id> --body
+//	./bin/brain/get.go <id> --json
 //
+// Needs CGO + libladybug. Python bin/kb/get is the CI fallback (no cgo).
 // NOTE: never run `gofmt -w` on this file — it breaks the shebang.
 package main
 
 import (
 	"os"
 
-	"github.com/eSlider/2dph/internal/cmdbin"
+	"github.com/eSlider/2dph/internal/brain"
 )
 
 func main() {
-	os.Exit(cmdbin.ExecFile("bin/kb/get", os.Args[1:]))
+	os.Exit(brain.MainGet(os.Args[1:]))
 }

bin/brain/stats.go

@@ -1,20 +1,21 @@
-//usr/bin/env go run -tags=brain_stats "$0" "$@"; exit
+//usr/bin/env go run -tags=system_ladybug,brain_stats "$0" "$@"; exit
-//go:build brain_stats
+//go:build cgo && system_ladybug && brain_stats
 //
 // bin/brain/stats.go - index health.
 //
 //	./bin/brain/stats.go
 //	./bin/brain/stats.go --json
 //
+// Needs CGO + libladybug. Python bin/kb/stats is the CI fallback (no cgo).
 // NOTE: never run `gofmt -w` on this file — it breaks the shebang.
 package main
 
 import (
 	"os"
 
-	"github.com/eSlider/2dph/internal/cmdbin"
+	"github.com/eSlider/2dph/internal/brain"
 )
 
 func main() {
-	os.Exit(cmdbin.ExecFile("bin/kb/stats", os.Args[1:]))
+	os.Exit(brain.MainStats(os.Args[1:]))
 }

bin/facts/audit.go

@@ -0,0 +1,21 @@
+//usr/bin/env go run -tags=facts_audit "$0" "$@"; exit
+//go:build facts_audit
+//
+// bin/facts/audit.go - 2-source + lexicon checks.
+//
+//	./bin/facts/audit.go self
+//	./bin/facts/audit.go db
+//
+// Python bin/facts/audit is the implementation (CI runs it directly).
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/facts/audit", os.Args[1:]))
+}

bin/facts/crm.go

@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=facts_crm "$0" "$@"; exit
+//go:build facts_crm
+//
+// bin/facts/crm.go - prove person↔company / company↔project (ooCRM × corpus).
+//
+//	./bin/facts/crm.go [--dry-run] [--mismatches]
+//
+// Python bin/facts/crm is the implementation. Graph write stays Python.
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/facts/crm", os.Args[1:]))
+}

bin/facts/extract.go

@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=facts_extract "$0" "$@"; exit
+//go:build facts_extract
+//
+// bin/facts/extract.go - acquire confirmed facts (2-source each).
+//
+//	./bin/facts/extract.go [--json] [--dry-run]
+//
+// Python bin/facts/extract is the implementation. Graph write stays Python.
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/facts/extract", os.Args[1:]))
+}

bin/tools/test_bin_layout.py

@@ -77,6 +77,44 @@ class BinLayoutTest(unittest.TestCase):
         for method in ("index.go", "get.go", "stats.go", "eval.go", "watch.go"):
             self._assert_shebang(f"bin/brain/{method}")
 
+    def test_brain_get_stats_eval_are_not_python_exec(self) -> None:
+        for method in ("get.go", "stats.go", "eval.go"):
+            text = (ROOT / "bin" / "brain" / method).read_text()
+            self.assertNotIn(
+                "ExecFile",
+                text,
+                f"bin/brain/{method} must call internal/brain, not ExecFile Python",
+            )
+            self.assertNotIn(
+                "cmdbin",
+                text,
+                f"bin/brain/{method} must not import internal/cmdbin",
+            )
+            self.assertIn(
+                "system_ladybug",
+                text.splitlines()[0],
+                f"bin/brain/{method} shebang must pass -tags=system_ladybug",
+            )
+            self.assertIn(
+                "github.com/eSlider/2dph/internal/brain",
+                text,
+            )
+
+    def test_eval_control_questions_live_in_rank(self) -> None:
+        rank = (ROOT / "internal" / "brain" / "rank" / "evalq.go").read_text()
+        py = (ROOT / "bin" / "kb" / "eval").read_text()
+        for frag in ("BM25", "DevOps", "LadybugDB"):
+            self.assertIn(frag, rank)
+            self.assertIn(frag, py)
+        self.assertIn("0.95", rank)
+
+    def test_facts_methods_are_shebangs(self) -> None:
+        for method in ("audit.go", "extract.go", "crm.go"):
+            self._assert_shebang(f"bin/facts/{method}")
+            text = (ROOT / "bin" / "facts" / method).read_text()
+            self.assertIn("cmdbin.ExecFile", text)
+            self.assertIn(f"bin/facts/{method.removesuffix('.go')}", text)
+
     def test_mail_import_is_shebang_not_brain_write(self) -> None:
         self._assert_shebang("bin/mail/import.go")
         index_mail = (ROOT / "bin" / "mail" / "index_mail").read_text()

bin/tools/test_published_docs.py

@@ -59,6 +59,14 @@ class PublishedDocsTest(unittest.TestCase):
         self.assertNotIn("password", settings.lower())
         self.assertIn("json", settings)
 
+    def test_readme_read_path_is_go(self) -> None:
+        plan = (ROOT / "PLAN.md").read_text()
+        self.assertIn("get.go", plan)
+        self.assertIn("CI fallback", plan)
+        design = (ROOT / "docs" / "design.md").read_text()
+        self.assertIn("internal/brain/rank", design)
+        self.assertIn("They do not exec Python", design)
+
     def test_readme_search_escalates_web(self) -> None:
         text = (ROOT / "README.md").read_text()
         self.assertIn("--no-web", text)

docs/crm-associations-proof.md

@@ -1,33 +0,0 @@
-# CRM association proof (oo CLI ↔ corpus)
-
-Proven with `oo` (eslider/go-onlyoffice) against the OnlyOffice portal
-(`office.produktor.io`). Portal CRM is the SSOT for company ↔ person ↔
-project associations; the corpus SoT (`eslider/cv/projects/knowledge-mesh-seed.yaml`)
-is the second, independent source. Facts that can be backed by both are
-written to the brain under `root=facts` by `bin/facts/crm`.
-
-## What was verified
-
-- Logical counts (portal MySQL): 1300 contacts = 897 persons + 404 companies,
-  198 projects, 998 deals, 939 project↔contact links.
-- Every client company linked to a project has ≥1 person underneath.
-- Every person `company_id` resolves to an existing company.
-- Corpus org list (9) maps 1:1 onto CRM companies:
-  ProProdukt SL / produktor.io, Dyvenia, Immowelt AG, WhereGroup,
-  Keynote SIGOS, D2S/SYSTEMS, GRID, Pack und Cup, Markets Platform.
-- 78 person↔company association facts written to the brain
-  (`how=crm-crosscheck`, `type=association`). Recall@5 in `bin/kb/eval` = 1.0.
-
-## Mistakes found
-
-| # | Mistake | Fix |
-|---|---------|-----|
-| 1 | Duplicate legal entity `GoldenRatio.Exchange` (contact 759) vs `Golden Ratio Exchange` (763); 3 deals (211, 287, 559) were linked to 759 | `oo contacts merge 759 763` — 763 kept, 759 removed, deal links re-pointed to 763 |
-| 2 | `env/`-wide: OnlyOffice creds file used wrong UX (user `eslider`, password with `$2` suffix) making `oo` auth fail | `.env` fixed to `eslider@gmail.com` + clean password; `.env` stays gitignored |
-
-## Gates after fix
-
-- `uv run python -m unittest discover -s bin/tools -t .` → 26 tests OK
-- `bin/facts/audit self` + `bin/facts/audit db` → ok
-- `bin/kb/eval` → recall@5 = 1.0
-- `go test ./...` (bin/server + bin/watch) → ok

docs/design.md

@@ -62,3 +62,11 @@ corpus HEAD.
 
 Confirmed = A×B or B×C agreement. Single source = hypothesis + `(not confirmed)`.
 Conflicting pairings (≥2 yes vs ≥2 no) = hypothesis (OQ1 → v2 resolution).
+
+## Read path
+
+`bin/brain/get.go`, `stats.go`, and `eval.go` call `internal/brain` with cgo
+(`system_ladybug`). They do not exec Python. Control questions for recall@5
+live in `internal/brain/rank` so CI can test the table without libladybug.
+Python `bin/kb/{get,stats,eval}` remain for GitHub Actions until the runner
+has ladybug cgo. Index/write is still `bin/kb/index`.

internal/brain/modelid.go

@@ -0,0 +1,3 @@
+package brain
+
+const ModelID = "minishlab/potion-multilingual-128M"

internal/brain/rank/evalq.go

@@ -0,0 +1,16 @@
+package rank
+
+// Eval control questions (recall@5). Kept here so CI can test the gate
+// table without ladybug cgo. The runner lives in internal/brain (cgo).
+const EvalRecallThreshold = 0.95
+
+type EvalQuestion struct {
+	Query    string
+	Fragment string
+}
+
+var EvalQuestions = []EvalQuestion{
+	{"hybrid search fts and vector", "BM25"},
+	{"eslider devops engineer", "DevOps"},
+	{"ladybugdb graph engine storage", "LadybugDB"},
+}

internal/brain/rank/evalq_test.go

@@ -0,0 +1,17 @@
+package rank
+
+import "testing"
+
+func TestEvalQuestionsAreThreeAndThreshold(t *testing.T) {
+	if EvalRecallThreshold != 0.95 {
+		t.Fatalf("threshold = %v", EvalRecallThreshold)
+	}
+	if len(EvalQuestions) != 3 {
+		t.Fatalf("questions = %d, want 3", len(EvalQuestions))
+	}
+	for _, q := range EvalQuestions {
+		if q.Query == "" || q.Fragment == "" {
+			t.Fatalf("empty control: %+v", q)
+		}
+	}
+}

internal/brain/read.go

@@ -0,0 +1,281 @@
+//go:build cgo && system_ladybug
+
+package brain
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"sort"
+	"strings"
+	"unicode/utf8"
+
+	"github.com/eSlider/2dph/internal/brain/rank"
+)
+
+func MainGet(args []string) int {
+	id, body, jsonOut := "", false, false
+	for _, a := range args {
+		switch {
+		case a == "--body":
+			body = true
+		case a == "--json":
+			jsonOut = true
+		case a == "-h" || a == "--help":
+			fmt.Fprintln(os.Stderr, `usage: bin/brain/get.go <id> [--body] [--json]`)
+			return 0
+		case strings.HasPrefix(a, "-"):
+			fmt.Fprintf(os.Stderr, "brain/get: unknown flag %s\n", a)
+			return 2
+		default:
+			id = a
+		}
+	}
+	if id == "" {
+		fmt.Fprintln(os.Stderr, "brain/get: id required")
+		return 2
+	}
+	if err := openBrain(); err != nil {
+		fmt.Fprintf(os.Stderr, "open brain: %v\n", err)
+		return 1
+	}
+	defer closeBrain()
+	meta, text, err := lookupLeaf(id)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "brain/get: %v\n", err)
+		return 1
+	}
+	out := Dict{
+		{"id", meta["id"]},
+		{"root", meta["root"]},
+		{"confidence", meta["confidence"]},
+		{"source", meta["source"]},
+		{"type", meta["type"]},
+	}
+	if body {
+		out = append(out, KV{"text", text})
+	} else {
+		out = append(out, KV{"snippet", clip(text, 280)})
+	}
+	if jsonOut {
+		m := map[string]any{}
+		for _, kv := range out {
+			m[kv.K] = kv.V
+		}
+		enc := json.NewEncoder(os.Stdout)
+		enc.SetIndent("", "  ")
+		enc.SetEscapeHTML(false)
+		return b2i(enc.Encode(m))
+	}
+	fmt.Print(toYAML(out, 0))
+	return 0
+}
+
+func MainStats(args []string) int {
+	jsonOut := false
+	for _, a := range args {
+		switch a {
+		case "--json":
+			jsonOut = true
+		case "-h", "--help":
+			fmt.Fprintln(os.Stderr, `usage: bin/brain/stats.go [--json]`)
+			return 0
+		default:
+			if strings.HasPrefix(a, "-") {
+				fmt.Fprintf(os.Stderr, "brain/stats: unknown flag %s\n", a)
+				return 2
+			}
+		}
+	}
+	if err := openBrain(); err != nil {
+		fmt.Fprintf(os.Stderr, "open brain: %v\n", err)
+		return 1
+	}
+	defer closeBrain()
+	s, err := leafStats()
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "brain/stats: %v\n", err)
+		return 1
+	}
+	if jsonOut {
+		enc := json.NewEncoder(os.Stdout)
+		enc.SetIndent("", "  ")
+		enc.SetEscapeHTML(false)
+		return b2i(enc.Encode(s))
+	}
+	by := s["by_root"].(map[string]int)
+	keys := make([]string, 0, len(by))
+	for k := range by {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	byRoot := make(Dict, 0, len(keys))
+	for _, k := range keys {
+		byRoot = append(byRoot, KV{k, by[k]})
+	}
+	out := Dict{
+		{"total", s["total"]},
+		{"by_root", byRoot},
+		{"db", s["db"]},
+		{"model", s["model"]},
+	}
+	fmt.Print(toYAML(out, 0))
+	return 0
+}
+
+func MainEval(args []string) int {
+	jsonOut := false
+	for _, a := range args {
+		switch a {
+		case "--json":
+			jsonOut = true
+		case "-h", "--help":
+			fmt.Fprintln(os.Stderr, `usage: bin/brain/eval.go [--json]`)
+			return 0
+		default:
+			if strings.HasPrefix(a, "-") {
+				fmt.Fprintf(os.Stderr, "brain/eval: unknown flag %s\n", a)
+				return 2
+			}
+		}
+	}
+	if err := openBrain(); err != nil {
+		fmt.Fprintf(os.Stderr, "open brain: %v\n", err)
+		return 1
+	}
+	defer closeBrain()
+	recalled := 0
+	details := make([]any, 0, len(rank.EvalQuestions))
+	jsDetails := make([]map[string]any, 0, len(rank.EvalQuestions))
+	for _, q := range rank.EvalQuestions {
+		hits, err := queryFTS(q.Query, 5)
+		ok := false
+		if err == nil {
+			frag := strings.ToLower(q.Fragment)
+			for _, h := range hits {
+				if strings.Contains(strings.ToLower(h.Text), frag) {
+					ok = true
+					break
+				}
+			}
+		}
+		if ok {
+			recalled++
+		}
+		details = append(details, Dict{
+			{"q", q.Query},
+			{"fragment", q.Fragment},
+			{"in_top5", ok},
+		})
+		jsDetails = append(jsDetails, map[string]any{
+			"q": q.Query, "fragment": q.Fragment, "in_top5": ok,
+		})
+	}
+	n := len(rank.EvalQuestions)
+	recall := 0.0
+	if n > 0 {
+		recall = float64(recalled) / float64(n)
+	}
+	passed := recall >= rank.EvalRecallThreshold
+	if jsonOut {
+		enc := json.NewEncoder(os.Stdout)
+		enc.SetIndent("", "  ")
+		enc.SetEscapeHTML(false)
+		_ = enc.Encode(map[string]any{
+			"recall@5": round3(recall),
+			"passed":   passed,
+			"gate":     n,
+			"details":  jsDetails,
+		})
+	} else {
+		out := Dict{
+			{"recall@5", round3(recall)},
+			{"passed", passed},
+			{"gate", n},
+			{"details", details},
+		}
+		fmt.Print(toYAML(out, 0))
+	}
+	if !passed {
+		return 2
+	}
+	return 0
+}
+
+func lookupLeaf(id string) (map[string]string, string, error) {
+	if conn == nil {
+		return nil, "", fmt.Errorf("brain not open")
+	}
+	stmt, err := conn.Prepare(
+		"MATCH (l:Leaf {id:$id}) RETURN l.id, l.text, l.root, l.confidence, l.source, l.type",
+	)
+	if err != nil {
+		return nil, "", err
+	}
+	defer stmt.Close()
+	res, err := conn.Execute(stmt, map[string]any{"id": id})
+	if err != nil {
+		return nil, "", err
+	}
+	if !res.HasNext() {
+		return nil, "", fmt.Errorf("no leaf %s", id)
+	}
+	row, err := res.Next()
+	if err != nil {
+		return nil, "", err
+	}
+	vals, err := row.GetAsSlice()
+	if err != nil || len(vals) < 6 {
+		return nil, "", fmt.Errorf("leaf row")
+	}
+	meta := map[string]string{
+		"id":         fmt.Sprint(vals[0]),
+		"root":       fmt.Sprint(vals[2]),
+		"confidence": fmt.Sprint(vals[3]),
+		"source":     fmt.Sprint(vals[4]),
+		"type":       fmt.Sprint(vals[5]),
+	}
+	return meta, fmt.Sprint(vals[1]), nil
+}
+
+func leafStats() (map[string]any, error) {
+	if conn == nil {
+		return nil, fmt.Errorf("brain not open")
+	}
+	res, err := conn.Query("MATCH (l:Leaf) RETURN l.root, count(*)")
+	if err != nil {
+		return nil, err
+	}
+	byRoot := map[string]int{}
+	total := 0
+	for res.HasNext() {
+		row, err := res.Next()
+		if err != nil {
+			return nil, err
+		}
+		vals, err := row.GetAsSlice()
+		if err != nil || len(vals) < 2 {
+			continue
+		}
+		n := int(asInt(vals[1]))
+		byRoot[fmt.Sprint(vals[0])] = n
+		total += n
+	}
+	return map[string]any{
+		"total":   total,
+		"by_root": byRoot,
+		"db":      dbPath(),
+		"model":   ModelID,
+	}, nil
+}
+
+func clip(s string, n int) string {
+	if utf8.RuneCountInString(s) <= n {
+		return s
+	}
+	return string([]rune(s)[:n])
+}
+
+func round3(f float64) float64 {
+	return float64(int(f*1000+0.5)) / 1000
+}

skills/brain/SKILL.md

@@ -28,7 +28,7 @@ bin/brain/search.go "onlyoffice postgres" --root facts # restrict to confirmed
 bin/brain/search.go "where is cs-lexicon" --json | yq '.[].ref'
 bin/brain/get.go <id> --body                           # full chunk only when needed
 bin/brain/stats.go                                     # index health
-bin/brain/eval.go                                      # recall@5 >= 0.95 gate
+bin/brain/eval.go                                      # recall@5 >= 0.95 gate (Go; Python bin/kb/eval is CI fallback)
 ```
 
 `bin/kb/search` is a deprecated wrapper. `--hop` errors (File/FROM_FILE edges