fix: ask 검색에서 제외 소스를 숨긴다

docs/reference/ignore-eject.en.md

@@ -7,9 +7,10 @@
 `/factlog sync` re-extracts **every** source on each run. To keep specific
 sources out of that — a draft, a work-in-progress, an external doc — add them to
 the per-KB **sync-ignore list** (`policy/sync-ignore.md`). Ignored sources are
-skipped by `/factlog sync`, `factlog ingest --scan`, and coverage gap reporting,
-**even when modified**. Their already-merged facts are kept untouched (use
-`factlog eject` to actually remove a fact).
+skipped by `/factlog sync`, `factlog ingest --scan`, coverage gap reporting, and
+the wiki-exploration evidence in `/factlog ask`, **even when modified**. Their
+already-merged facts are kept untouched (use `factlog eject` to actually remove
+a fact).
 
 ```bash
 factlog ignore drafts/*.md sources/wip-notes.md   # add pattern(s)

docs/reference/ignore-eject.md

@@ -7,8 +7,9 @@
 `/factlog sync` 는 매 실행마다 **모든** 소스를 다시 추출합니다. 특정 소스를
 그 대상에서 빼려면 — 초안, 작업 중인 문서, 외부 문서 등 — KB별 **sync-ignore
 목록**(`policy/sync-ignore.md`)에 추가하십시오. 무시된 소스는 **수정되더라도**
-`/factlog sync`, `factlog ingest --scan`, 커버리지 누락 보고에서 건너뜁니다. 이미
-머지된 사실은 그대로 유지됩니다(사실을 실제로 제거하려면 `factlog eject` 사용).
+`/factlog sync`, `factlog ingest --scan`, 커버리지 누락 보고와 `/factlog ask`의
+wiki 탐색 근거에서 건너뜁니다. 이미 머지된 사실은 그대로 유지됩니다(사실을 실제로
+제거하려면 `factlog eject` 사용).
 
 ```bash
 factlog ignore drafts/*.md sources/wip-notes.md   # add pattern(s)

factlog/cli.py

@@ -472,9 +472,10 @@ def cmd_doctor(_args: argparse.Namespace) -> int:
 # Sync-ignore list
 #
 # Source files matching these glob patterns are SKIPPED by `/factlog sync`
-# (re-extraction), `factlog ingest --scan`, and coverage gap reporting — even
-# when modified. Their already-merged facts are KEPT (use `factlog eject` to
-# remove those). Manage with `factlog ignore [--remove] <pattern>`.
+# (re-extraction), `factlog ingest --scan`, coverage gap reporting, and
+# `/factlog ask` wiki exploration — even when modified. Their already-merged
+# facts are KEPT (use `factlog eject` to remove those). Manage with
+# `factlog ignore [--remove] <pattern>`.
 #
 # One pattern per line; '#' comments and '-' bullets allowed; quote a pattern
 # with spaces (or one starting with '#') in `backticks`. A pattern matches a
@@ -1373,12 +1374,13 @@ def _matches_extended(r: dict) -> bool:
 
 
 def cmd_ignore(args: argparse.Namespace) -> int:
-    """Manage policy/sync-ignore.md — glob patterns of sources excluded from sync.
+    """Manage policy/sync-ignore.md — sources excluded from sync and wiki evidence.
 
     No patterns: list current entries and the on-disk sources each matches.
     With pattern(s): add them, or remove them with --remove. Excluding a source
-    only stops its re-extraction (ingest --scan / sync / coverage); its already-
-    merged facts are untouched (use `factlog eject` to remove those).
+    stops its re-extraction (ingest --scan / sync / coverage) and keeps it out
+    of `/factlog ask` wiki evidence; its already-merged facts are untouched
+    (use `factlog eject` to remove those).
     """
     import re
     import unicodedata
@@ -2760,7 +2762,7 @@ def build_parser() -> argparse.ArgumentParser:
 
     ignore = sub.add_parser(
         "ignore",
-        help="manage policy/sync-ignore.md: glob patterns of sources excluded from sync",
+        help="manage policy/sync-ignore.md: glob patterns excluded from sync and wiki evidence",
     )
     ignore.add_argument("patterns", nargs="*", help="glob/path pattern(s) to add (omit to list)")
     ignore.add_argument("--remove", action="store_true", help="remove the given pattern(s) instead of adding")

skills/factlog/SKILL.md

@@ -385,8 +385,9 @@ Sources matching `policy/sync-ignore.md` are skipped. Then extract from **both**
 source whose path matches one of its glob patterns — by full ref (`sources/...`
 or `runs/sources/...`) or by the path within the source root (so `drafts/*.md`
 matches `sources/drafts/x.md`). These sources are excluded from re-extraction on
-purpose; their already-merged facts are left as-is. (Manage the list with
-`factlog ignore`.)
+purpose and are also excluded from `/factlog ask` wiki-exploration evidence;
+their already-merged facts are left as-is. (Manage the list with `factlog
+ignore`.)
 
 For each *non-ignored* file under `sources/<name>` **and** `runs/sources/<name>`
 in the KB root:

tests/test_ask_router.sh

@@ -193,6 +193,21 @@ if router search "WidgetX ToolA" | "$PYTHON" -c "import json,sys; d=json.load(sy
 # pages/ candidate content must never appear in search citations
 if router search "Datadog may_use" | grep -qE 'pages/|may_use|confidence=0\.40'; then bad "pages/ candidate content leaked into search results"; else ok "pages/ excluded from search (no candidate leak)"; fi
 
+# Sync-ignored primary sources are deliberately excluded from wiki evidence;
+# non-ignored sources and supplementary decisions retain their normal behavior.
+mkdir -p "$KB/sources/drafts" "$KB/runs/sources/drafts"
+printf 'syncignoretoken appears only in an ignored original.\n' > "$KB/sources/drafts/ignored.md"
+printf 'syncignoretoken appears only in an ignored conversion.\n' > "$KB/runs/sources/drafts/ignored.md"
+printf 'syncignoretoken appears in the retained source.\n' > "$KB/sources/retained.md"
+printf -- '- drafts/**\n' >> "$KB/policy/sync-ignore.md"
+ignored_search="$(router search "syncignoretoken")"
+if printf '%s' "$ignored_search" | grep -qE 'sources/drafts/ignored\.md|runs/sources/drafts/ignored\.md'; then bad "sync-ignored sources leaked into search results"; else ok "sync-ignored sources excluded from search results"; fi
+if printf '%s' "$ignored_search" | grep -qF 'sources/retained.md'; then ok "non-ignored source remains in search results"; else bad "non-ignored source missing from search results"; fi
+if router wiki "syncignoretoken" | grep -qE 'sources/drafts/ignored\.md|runs/sources/drafts/ignored\.md'; then bad "sync-ignored sources leaked into rendered wiki evidence"; else ok "sync-ignored sources excluded from rendered wiki evidence"; fi
+rm -rf "$KB/sources/drafts" "$KB/runs/sources/drafts" "$KB/sources/retained.md"
+# Keep later search cases independent from this filtering fixture.
+sed -i.bak '/^- drafts\/\*\*$/d' "$KB/policy/sync-ignore.md" && rm -f "$KB/policy/sync-ignore.md.bak"
+
 wiki_out="$(router wiki "what uses FastAPI" --reason "unknown entity")"
 if printf '%s' "$wiki_out" | grep -qF "UNVERIFIED — wiki exploration"; then ok "wiki answer carries UNVERIFIED marker"; else bad "wiki answer missing UNVERIFIED marker"; fi
 if printf '%s' "$wiki_out" | grep -qF "sources/acme.md:"; then ok "wiki answer cites a source path:line"; else bad "wiki answer missing citation"; fi

tools/ask_router.py

@@ -84,6 +84,8 @@
     policy_predicates,
     relation_aliases,
     run_wirelog,
+    is_sync_ignored,
+    sync_ignore_patterns,
 )
 from factlog import literal_types  # noqa: E402
 
@@ -653,6 +655,7 @@ def search(question: str, root: Path, *, limit: int = 10) -> list[dict[str, obje
     if not patterns:
         return []
     scored: list[tuple[tuple[int, int], dict[str, object]]] = []
+    ignored_patterns = sync_ignore_patterns(root)
     for rel, label in _wiki_corpus():
         base = root / rel
         if not base.is_dir():
@@ -662,6 +665,12 @@ def search(question: str, root: Path, *, limit: int = 10) -> list[dict[str, obje
             # Stay within the corpus root: never follow a symlink out of the KB.
             if not path.resolve().is_relative_to(base_resolved):
                 continue
+            ref = path.relative_to(root).as_posix()
+            # Sync-ignore means this primary source is not evidence for wiki
+            # exploration either. Supplementary decisions remain searchable:
+            # they are explicitly labeled and are not source files.
+            if rel in WIKI_SOURCE_DIRS and is_sync_ignored(ref, ignored_patterns):
+                continue
             try:
                 text = path.read_text(encoding="utf-8")
             except (OSError, UnicodeDecodeError):
@@ -681,7 +690,7 @@ def search(question: str, root: Path, *, limit: int = 10) -> list[dict[str, obje
                 last_end = end - 1
                 excerpt = "\n".join(_sanitize(line_text) for line_text in lines[start:end])
                 result = {
-                    "file": path.relative_to(root).as_posix(),
+                    "file": ref,
                     "line": i + 1,
                     "excerpt": excerpt,
                     "dir": label,