fix(search): show the passage that matched, not the opening of the body (#4243)
CI & Build / Plugin hooks (push) Successful in 10s
CI & Build / Python lint (push) Successful in 3s
CI & Build / TypeScript typecheck (push) Successful in 53s
CI & Build / Python tests (push) Failing after 1m5s
CI & Build / Build & push image (push) Skipped
CI & Build / integration (push) Successful in 46s
CI & Build / Plugin hooks (push) Successful in 10s
CI & Build / Python lint (push) Successful in 3s
CI & Build / TypeScript typecheck (push) Successful in 53s
CI & Build / Python tests (push) Failing after 1m5s
CI & Build / Build & push image (push) Skipped
CI & Build / integration (push) Successful in 46s
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
This commit is contained in:
@@ -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),
|
||||
|
||||
@@ -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:
|
||||
|
||||
@@ -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:
|
||||
|
||||
@@ -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
|
||||
Reference in New Issue
Block a user