fix(search): show the passage that matched, not the opening of the body (#4243)

Raised by the operator: are we limiting what comes back by character count,
and how do we verify the pertinent part is the part displayed?

We were not. mcp/tools/search.py sent (note.body or "")[:240] — a head cut,
with no marker that anything had been removed, so a 240-character preview of
a 4000-character record was indistinguishable from a complete short one.

The opening is the wrong span. The match is semantic and per chunk, and
semantic_search_notes collapses to best-chunk-per-note — its own comment at
the collapse says "the first appearance of a note is its best chunk". So the
system identified the passage that earned the hit and then discarded it:
select(Note, distance) kept no chunk column. A record could rank first on its
sixth paragraph, be previewed by its first, and be judged irrelevant on a
span the search had already scored lower. That biases against long records,
and it is self-concealing — the caller who does not open it never learns the
preview was misleading.

  - embeddings: chunk_index/chunk_text ride along in the select, and the
    collapse records the winner in report["best_chunk"]. Carried in `report`,
    NOT by widening the return tuple: ten callers unpack (score, note) at
    ~18 sites and nothing would catch the misses (lesson #4207). `report` is
    the side-channel this function already uses for best_available_score.
  - search(): excerpt / excerpt_is / body_length, and read_full when there is
    more. A caller that cannot tell a matched passage from a document opening
    cannot judge whether to look deeper, which is the only decision the field
    supports.

elide() moves to services/text.py so both callers share one copy, and it
keeps BOTH ends with a stated gap — it is the fallback for when nothing
identifies a better span than "all of it", not the goal.

Also fixes a guard that produced a false failure on the previous commit:
test_pull_telemetry checked `"project_id: int = 0" in body.split("\n")[0]`,
which sees only the first line, so wrapping get_task's signature over four
lines made it report a function that does take the project as one that does
not. Parsed with ast now, and proven to still reject an absent or
wrongly-typed parameter rather than being appeased by reflowing the code.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01821k5B3Ysecp9fNYs92Kuy

src/scribe/mcp/tools/search.py

@@ -11,6 +11,7 @@ import time
 
 from scribe.mcp._context import current_user_id
 from scribe.services.access import owner_names_for
+from scribe.services.text import elide
 from scribe.services.embeddings import (
     DEFAULT_SIMILARITY_THRESHOLD, semantic_search_milestones, semantic_search_notes,
     semantic_search_rules,
@@ -102,6 +103,55 @@ async def _search_milestones(uid: int, q: str, limit: int, project_id: int) -> d
     }
 
 
+# A matched chunk is at most _CHUNK_CHAR_BUDGET (1400) characters, and it is
+# the evidence the ranking was built on — so it is worth more room than the 240
+# characters of document opening this used to send. Elision inside a chunk is
+# far less lossy than a head cut of a whole record: the region is already the
+# right one.
+_EXCERPT_CHARS = 1000
+
+
+def result_excerpt(note, chunk: dict | None) -> dict:
+    """The part of a record a caller judges "should I open this?" on.
+
+    This used to be `(note.body or "")[:240]` — the document's opening, with
+    no marker that anything had been cut, so a 240-character preview of a
+    4000-character record was indistinguishable from a complete short one.
+
+    The opening is the wrong span. The match was semantic and per-chunk, and
+    `semantic_search_notes` collapses to best-chunk-per-note — so the system
+    already knows which passage earned the hit and used to discard it. A
+    record could rank first on its sixth paragraph and be previewed by its
+    first, which the search had already judged less relevant, and the caller
+    would decide from that and never know (#4243).
+
+    So: show the matched passage when there is one, the body when it fits, and
+    in either case SAY which of the two this is. A caller that cannot tell an
+    excerpt from a whole record cannot tell whether looking deeper is worth it,
+    which is the only decision this field supports.
+    """
+    body = note.body or ""
+    matched = (chunk or {}).get("text") or ""
+    out: dict = {"body_length": len(body)}
+
+    if matched.strip():
+        text, cut = elide(matched.strip(), _EXCERPT_CHARS)
+        out["excerpt"] = text
+        out["excerpt_is"] = "matched_passage"
+        if (chunk or {}).get("index") is not None:
+            out["chunk_index"] = int(chunk["index"])
+    else:
+        # No stored chunk — an un-embedded record, or a caller that passed no
+        # report. Fall back to the opening, and name it as the opening rather
+        # than letting it pass for the relevant part.
+        text, cut = elide(body, _EXCERPT_CHARS)
+        out["excerpt"] = text
+        out["excerpt_is"] = "body_opening"
+    if cut or (out["excerpt_is"] == "matched_passage" and len(matched) < len(body)):
+        out["read_full"] = "get_note / get_task by id for the whole record."
+    return out
+
+
 async def search(
     q: str,
     content_type: str = "all",
@@ -150,9 +200,21 @@ async def search(
             list_system_records gives the same slice unranked.
 
     Returns:
-        {"results": [{"id", "title", "body", "is_task", "tags", "similarity"}],
+        {"results": [{"id", "title", "excerpt", "excerpt_is", "body_length",
+                      "is_task", "tags", "similarity"}],
          "total": int}
 
+        `excerpt` is a SPAN of the record, not the record. `excerpt_is` says
+        which span: "matched_passage" is the chunk that actually earned the
+        hit — the evidence the ranking was built on, and the right thing to
+        judge relevance from. "body_opening" is a fallback for a record with
+        no stored chunk, and is only the beginning of the text, which may say
+        nothing about why it matched. `body_length` is the whole record's
+        size, so a long record previewed by a short span is visible as one;
+        `read_full` appears when there is more, and opening the id by
+        get_note / get_task is how you get it. Judge from the passage, not
+        from the fact that a preview looked thin.
+
     A result marked `shared: true` with an `owner` belongs to another user —
     that person's suggestion, not the operator's own record or settled practice.
     Weigh it on its merits and say whose it is when you use it.
@@ -195,12 +257,13 @@ async def search(
     owners = await owner_names_for(
         {int(note.user_id) for _s, note in raw if note.user_id != uid}
     )
+    chunks = report.get("best_chunk") or {}
     return {
         "results": [
             {
                 "id": note.id,
                 "title": note.title,
-                "body": (note.body or "")[:240],
+                **result_excerpt(note, chunks.get(int(note.id))),
                 "is_task": bool(note.is_task),
                 "tags": list(note.tags or []),
                 "similarity": float(score),

src/scribe/mcp/tools/tasks.py

@@ -45,6 +45,7 @@ from scribe.services import task_logs as task_logs_svc
 from scribe.services import trash as trash_svc
 from scribe.services.note_usage import record_pulled
 from scribe.services.record_refs import refuse_guessed_ids
+from scribe.services.text import elide
 
 
 # A work log entry is prose, often long — the discipline asks for what was
@@ -64,31 +65,6 @@ _WORK_LOG_ADVICE = (
 )
 
 
-def elide(text: str, budget: int) -> tuple[str, bool]:
-    """Cut to `budget` characters from the MIDDLE, keeping both ends.
-
-    A head-only cut — `text[:800]` — decides what a reader sees by character
-    position, which is uncorrelated with what matters. Prose does not put its
-    conclusion first: a log entry that opens with what was tried and closes
-    with "so this shipped in 04775c3" loses exactly the sentence that answers
-    the question, and the reader cannot tell, because a truncation marker says
-    that something was removed and never whether it mattered.
-
-    So keep the opening (what this entry is about) AND the closing (where it
-    landed), and say in between how much went. Two thirds to the head because
-    that is where the subject is established; the tail needs less to carry a
-    conclusion. `budget` of 0 means no cut.
-    """
-    if budget <= 0 or len(text) <= budget:
-        return text, False
-    head_len = max(1, budget * 2 // 3)
-    tail_len = max(1, budget - head_len)
-    omitted = len(text) - head_len - tail_len
-    head = text[:head_len].rstrip()
-    tail = text[-tail_len:].lstrip()
-    return f"{head}\n\n[… {omitted} characters omitted …]\n\n{tail}", True
-
-
 def work_log_payload(
     logs: list, total: int, chars: int, latest_chars: int = _WORK_LOG_LATEST_CHARS
 ) -> dict:

src/scribe/services/embeddings.py

@@ -686,7 +686,16 @@ async def semantic_search_notes(
             # to the note's owner, so filtering it would pin every scope to "own"
             # and leave shared records unreachable by meaning.
             stmt = (
-                select(Note, distance.label("distance"))
+                # chunk_index/chunk_text ride along so the collapse below can
+                # say WHICH passage matched. Without them the caller is left
+                # previewing the head of the body — a span this query has
+                # already determined is not why the record ranked (#4243).
+                select(
+                    Note,
+                    distance.label("distance"),
+                    NoteEmbedding.chunk_index,
+                    NoteEmbedding.chunk_text,
+                )
                 .select_from(NoteEmbedding)
                 .join(Note, NoteEmbedding.note_id == Note.id)
                 .where(
@@ -774,11 +783,23 @@ async def semantic_search_notes(
     # Recover similarity (1 - distance); order stays highest-first.
     scored: list[tuple[float, Note]] = []
     seen: set[int] = set()
-    for note, dist in rows:
+    # The winning row IS the best chunk, by the ordering above — so this is the
+    # one place that knows which passage earned the hit. Kept beside the score
+    # rather than returned with it: the return type is list[tuple[float, Note]]
+    # and ten callers unpack it at ~18 sites, so widening the tuple would be an
+    # interface change to every one of them with nothing to catch the misses
+    # (lesson #4207). `report` is the side-channel this function already uses
+    # for best_available_score.
+    best_chunk: dict[int, dict] = {}
+    for note, dist, chunk_index, chunk_text in rows:
         if int(note.id) in seen:
             continue
         seen.add(int(note.id))
         scored.append((1.0 - float(dist), note))
+        best_chunk[int(note.id)] = {
+            "index": int(chunk_index),
+            "text": chunk_text or "",
+        }
     # The best score anything reached, bar or no bar. Recorded BEFORE the
     # filter because a call that returns nothing is exactly when it matters.
     if report is not None:
@@ -797,8 +818,18 @@ async def semantic_search_notes(
         report["best_available_id"] = int(best[1].id) if best else None
     scored = [pair for pair in scored if pair[0] >= threshold]
     if not demote_superseded:
-        return scored[:limit]
-    return await _apply_supersession_penalty(scored, limit)
+        final = scored[:limit]
+    else:
+        final = await _apply_supersession_penalty(scored, limit)
+    # Only for what actually came back, so a caller can key straight off the
+    # results without carrying chunks for records it never saw.
+    if report is not None:
+        report["best_chunk"] = {
+            int(n.id): best_chunk[int(n.id)]
+            for _s, n in final
+            if int(n.id) in best_chunk
+        }
+    return final
 
 
 async def backfill_note_embeddings() -> None:

src/scribe/services/text.py

@@ -0,0 +1,38 @@
+"""Text shortening for surfaces that cannot show a record whole.
+
+One function, in one place, because both doors shorten and two copies would
+drift — and because the reasoning below is the part that matters and should
+not have to be re-derived at each call site.
+"""
+
+
+def elide(text: str, budget: int) -> tuple[str, bool]:
+    """Cut to `budget` characters from the MIDDLE, keeping both ends.
+
+    A head-only cut — `text[:800]` — decides what a reader sees by character
+    position, which is uncorrelated with what matters. Prose does not put its
+    conclusion first: a passage that opens with what was attempted and closes
+    with "so this shipped in 04775c3" loses exactly the sentence that answers
+    the question. Worse, the reader cannot tell: a truncation marker says that
+    something was removed, never whether it mattered, so the decision "should
+    I look deeper?" gets made on evidence selected by length.
+
+    So keep the opening (what this is about) AND the closing (where it landed),
+    and state in between how much went. Two thirds to the head because that is
+    where the subject is established; a conclusion needs less room to carry.
+
+    Returns (text, was_cut). A `budget` of 0 or less means no cut.
+
+    This is the fallback, not the goal. Where the system knows WHICH span of a
+    record is the relevant one — a semantic search knows exactly that, and
+    stores it (#4243) — show that span and say so. Reach for this only when
+    nothing identifies a better part than "all of it".
+    """
+    if budget <= 0 or len(text) <= budget:
+        return text, False
+    head_len = max(1, budget * 2 // 3)
+    tail_len = max(1, budget - head_len)
+    omitted = len(text) - head_len - tail_len
+    head = text[:head_len].rstrip()
+    tail = text[-tail_len:].lstrip()
+    return f"{head}\n\n[… {omitted} characters omitted …]\n\n{tail}", True