Files
FabledCurator/backend/app/services/post_association_service.py
T
bvandeusenandClaude Opus 5 235393c08b
CI / lint (push) Successful in 3s
CI / extension-version (push) Successful in 3s
Build images / sign-extension (push) Successful in 4s
Build images / build-agent (push) Successful in 7s
CI / frontend-build (push) Successful in 25s
CI / backend-lint-and-test (push) Successful in 32s
Build images / build-web (push) Successful in 1m3s
Build images / smoke-web (push) Skipped
Build images / build-ml (push) Successful in 1m53s
Build images / promote (push) Skipped
CI / integration (push) Successful in 2m12s
feat: link the Patreon teaser to the Discord drop it announced (388 step E5)
The point of the milestone rather than its tail. Two of the operator's artists
post a deliberately cropped fragment on Patreon to signal that the real thing
has landed in their Discord; this proposes those pairs.

Confirm-only, following the FC-6.3 series matcher. A wrongly-asserted
association tells the operator two different pieces are one, which is strictly
worse than no link: no link leaves them where they already were, a wrong one
actively misinforms and then propagates into whatever reads it. So the
matcher's job is a SHORT list worth reading, not a long list worth trusting.

**The threshold sits above every single signal weight, and that is the
design.** Proximity is 0.55, declaration 0.45, the cut 0.60 — so neither
signal can carry a pair alone. That makes "time proximity alone is never
sufficient" an arithmetic property rather than an aspiration: on a busy day an
artist posts several times, and a matcher that could pair on proximity alone
would turn every one of those days into false pairs until the review queue got
abandoned. A guard test asserts the relationship against WEIGHTS directly, so
it survives any refactor of the scorer, and says in its own failure message not
to fix it by lowering the assertion.

**Crop-to-source matching is HELD, on the plan's instruction** — real work with
real false-positive risk, worth building only once signals 1 and 2 are shown
insufficient against the operator's actual artists. Worth stating: a naive
whole-image SigLIP similarity is NOT that signal. A cropped teaser and its full
version are precisely the pair a whole-image comparison handles worst, so
adding one as a "bonus" would mostly add noise while looking like progress.

Two premises in the plan corrected in the building:

* **E4 is not actually a prerequisite.** A Patreon Source and a Discord Source
  the operator has added under one Artist already share `Post.artist_id`, and
  the synthetic grouping inherits it. E4 EXTENDS this to creators FC has to
  learn the association for; it is not needed to represent one FC was told.
  Same-artist is then a hard filter, not a scored signal — two different
  creators posting minutes apart is a coincidence, not evidence.
* **`link_extract` cannot supply the declaration signal.** It exists, but
  `SUPPORTED_HOSTS` is file hosts only and `host_for()` returns None for a
  Discord URL, so no ExternalLink row is ever written for one. The signal
  reads the post body directly instead.

And a bug my own test would have caught: `declared_signal` stripped the HTML
before looking for an invite, but `html_to_plain` discards attributes and
these creators put the invite in an anchor's `href` — so the strongest form of
the signal was being thrown away, leaving only whatever the link text said.
The invite now matches the raw body; the bare mention still matches stripped
text, so `\bdiscord\b` is tested against prose rather than against markup.

Dismissed rows are kept, not deleted: the row is what remembers the rejection,
and re-proposing a rejected pair on every scan is the one behaviour that makes
a review queue get ignored. Both FKs CASCADE, so E3's one-DELETE reversal
cannot leave a proposal pointing at a post that no longer exists.

Only ACCEPTED links reach the post payload. A pending proposal is a question
for the review queue, not a claim to render beside the artwork.

UI (rule 27) follows in the next commit.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LNXXULQDjVZmbuNa2G9mD9
2026-09-10 11:47:28 -04:00

309 lines
12 KiB
Python

"""The announcement matcher: which Patreon post announced which Discord drop.
Milestone 388, step E5.
Two of the operator's artists post a deliberately CROPPED fragment on Patreon
to signal that the real thing has landed in their Discord. This service
proposes those pairs, and proposes only — the operator accepts or dismisses,
following the FC-6.3 series matcher (task 737) rather than linking on its own.
## Why confirm-only is not caution for its own sake
A wrongly-asserted association tells the operator that two different pieces are
one. That is strictly worse than no link at all: no link leaves them exactly
where they already were, a wrong one actively misinforms and then propagates
into whatever reads the association. So the matcher's job is to make a SHORT
list worth reading, not a long list worth trusting.
## Signals, and the one deliberately NOT built
1. **Time proximity.** The Patreon post exists in order to announce the drop,
so the two are minutes-to-hours apart. Nearly free, and strong.
2. **The post says so.** These announcements routinely name Discord or carry
an invite link, which is close to a declaration.
3. **Crop-to-source matching is HELD, on the plan's own instruction** — it is
real work with real false-positive risk, and it is only worth building once
1 and 2 are shown to be insufficient against the operator's actual artists.
Nothing here should be read as evidence it is unnecessary; it is deferred,
and the thing that would justify it is an empty review queue on a pair the
operator can see with their own eyes.
Note also that a naive whole-image SigLIP similarity is NOT that signal. A
cropped teaser and its full version are exactly the pair a whole-image
comparison handles worst, so adding one as a "bonus" would mostly add noise
while looking like progress.
## Creator identity comes free, so E4 is not actually a prerequisite
The plan listed E4 (creator identity across the two channels) as a dependency.
It is not one for the pairs that matter today: a Patreon `Source` and a Discord
`Source` the operator has added under the same `Artist` already share
`Post.artist_id`, and the synthetic grouping inherits it (E2). E4 EXTENDS this
to creators whose association FC has to learn rather than being told; it is not
needed to represent an association FC already knows.
## A correction the plan carried, worth restating
`link_extract.py` does exist and does capture off-platform links — but only
file hosts (`SUPPORTED_HOSTS` is mega/gdrive/mediafire/dropbox/pixeldrain).
`host_for()` returns None for a Discord URL, so no `ExternalLink` row is ever
written for one. The declaration signal therefore reads the post body itself
rather than the extracted-links table the plan assumed it could use.
"""
from __future__ import annotations
import logging
import re
from datetime import UTC, datetime, timedelta
from sqlalchemy import func, or_, select
from sqlalchemy.ext.asyncio import AsyncSession
from ..models import ImportSettings, Post, PostAssociation
from ..utils.text import html_to_plain
from .discord_grouping import DROP_GROUPER
log = logging.getLogger(__name__)
# Additive weights, summing to 1.0. Kept as constants rather than settings —
# the sensitivity knob that matters is the threshold, and per-signal weights
# are an over-tune (same call as series_match_service.WEIGHTS).
#
# THE RELATIONSHIP TO THE THRESHOLD IS THE DESIGN. No single weight may reach
# the default threshold, which is what makes "time proximity alone is never
# enough" arithmetic rather than aspirational: on a busy day an artist posts
# several times, and a matcher that could pair on proximity alone would turn
# every busy day into false pairs. A guard test pins this.
WEIGHTS = {"proximity": 0.55, "declared": 0.45}
# A Discord INVITE in the body is close to a declaration; the bare word is
# weaker but still meaningful, because these posts are short and on-topic.
_INVITE = re.compile(r"discord\.(?:gg|com/invite)/", re.I)
_MENTION = re.compile(r"\bdiscord\b", re.I)
DECLARED_INVITE = 1.0
DECLARED_MENTION = 0.6
MAX_CANDIDATES = 25
def proximity_signal(gap: timedelta, window: timedelta) -> float:
"""1.0 when the two posts are simultaneous, decaying linearly to 0 at the
window's edge. Linear rather than a step, so a pair an hour outside a
hand-tuned window degrades instead of vanishing."""
if window <= timedelta(0):
return 0.0
seconds = abs(gap.total_seconds())
if seconds >= window.total_seconds():
return 0.0
return round(1.0 - (seconds / window.total_seconds()), 4)
def declared_signal(description: str | None) -> float:
"""Does the announcement say, in its own body, that this is about Discord?
The invite is matched against the RAW body and the bare mention against the
stripped text, which is not fussiness — post bodies are HTML, and these
creators put the invite in an anchor's `href`. `html_to_plain` discards
attributes, so stripping first would have thrown away the strongest form of
the signal and left only whatever the link text happened to say.
The mention still reads stripped text, so `\bdiscord\b` is matched against
prose rather than against markup and URLs, where it would fire on any link
that merely passes through a discord domain.
"""
if not description:
return 0.0
if _INVITE.search(description):
return DECLARED_INVITE
if _MENTION.search(html_to_plain(description) or ""):
return DECLARED_MENTION
return 0.0
def weighted_score(signals: dict) -> float:
return round(sum(WEIGHTS[k] * signals.get(k, 0.0) for k in WEIGHTS), 4)
def _post_time(post: Post) -> datetime:
return post.post_date or post.downloaded_at
class PostAssociationService:
def __init__(self, session: AsyncSession):
self.session = session
async def _decided(self, announcement_id: int) -> set[int]:
"""Payload posts already proposed for this announcement, in ANY status.
Dismissed pairs are included deliberately: re-proposing a pair the
operator has already rejected on every subsequent scan is the single
behaviour that makes a review queue get ignored.
"""
rows = (await self.session.execute(
select(PostAssociation.payload_post_id)
.where(PostAssociation.announcement_post_id == announcement_id)
)).scalars().all()
return set(rows)
async def _candidate_groups(
self, announcement: Post, *, window: timedelta,
) -> list[Post]:
"""Synthetic Discord groupings by the SAME artist, inside the window.
Same-artist is the identity signal and it is free (see the module
docstring on E4). It is also a hard filter rather than a scored one:
two different creators posting minutes apart is a coincidence, not
evidence, and letting it score at all would mean a busy hour across the
library could out-vote everything else.
"""
at = _post_time(announcement)
sort_key = func.coalesce(Post.post_date, Post.downloaded_at)
return (await self.session.execute(
select(Post)
.where(
Post.artist_id == announcement.artist_id,
Post.synthesized_by == DROP_GROUPER,
Post.id != announcement.id,
sort_key >= at - window,
sort_key <= at + window,
)
.order_by(sort_key)
.limit(MAX_CANDIDATES)
)).scalars().all()
async def match_post(
self, announcement_id: int, *, threshold: float, window_hours: float,
) -> int:
"""Score one announcement against nearby groupings. Returns proposals made."""
announcement = await self.session.get(Post, announcement_id)
if announcement is None or announcement.synthesized_by is not None:
# A synthetic post cannot announce anything — FC wrote it.
return 0
window = timedelta(hours=window_hours)
declared = declared_signal(announcement.description)
already = await self._decided(announcement_id)
made = 0
for group in await self._candidate_groups(announcement, window=window):
if group.id in already:
continue
signals = {
"proximity": proximity_signal(
_post_time(group) - _post_time(announcement), window,
),
"declared": declared,
}
score = weighted_score(signals)
if score < threshold:
continue
self.session.add(PostAssociation(
announcement_post_id=announcement.id,
payload_post_id=group.id,
score=score,
signals=signals,
status="pending",
))
made += 1
return made
async def list_pending(self) -> list[dict]:
rows = (await self.session.execute(
select(PostAssociation)
.where(PostAssociation.status == "pending")
.order_by(PostAssociation.score.desc(), PostAssociation.id.desc())
)).scalars().all()
return [
{
"id": a.id,
"announcement_post_id": a.announcement_post_id,
"payload_post_id": a.payload_post_id,
"score": a.score,
"signals": a.signals,
}
for a in rows
]
async def accept(self, association_id: int) -> dict | None:
a = await self.session.get(PostAssociation, association_id)
if a is None:
return None
a.status = "linked"
return {"id": a.id, "status": a.status}
async def dismiss(self, association_id: int) -> dict | None:
a = await self.session.get(PostAssociation, association_id)
if a is None:
return None
# Kept, not deleted — the row is what remembers the rejection.
a.status = "dismissed"
return {"id": a.id, "status": a.status}
async def linked_for(self, post_ids: list[int]) -> dict[int, list[dict]]:
"""Accepted links touching these posts, keyed by post id, BOTH ways.
A post is either end of the relationship, and each end wants the other
one: the teaser wants "the full set is over here", the grouping wants
"this is what announced me". One query, both directions.
"""
if not post_ids:
return {}
rows = (await self.session.execute(
select(PostAssociation).where(
PostAssociation.status == "linked",
or_(
PostAssociation.announcement_post_id.in_(post_ids),
PostAssociation.payload_post_id.in_(post_ids),
),
)
)).scalars().all()
out: dict[int, list[dict]] = {}
for a in rows:
if a.announcement_post_id in post_ids:
out.setdefault(a.announcement_post_id, []).append(
{"role": "announces", "post_id": a.payload_post_id, "id": a.id}
)
if a.payload_post_id in post_ids:
out.setdefault(a.payload_post_id, []).append(
{"role": "announced_by", "post_id": a.announcement_post_id, "id": a.id}
)
return out
async def rescan(session: AsyncSession, *, now: datetime | None = None) -> dict:
"""Score every recent non-synthetic post against nearby groupings."""
settings = await ImportSettings.load(session)
if not settings.discord_link_enabled:
return {"enabled": False, "scanned": 0, "proposed": 0}
now = now or datetime.now(UTC)
window_hours = float(settings.discord_link_window_hours)
# Only look at announcements that could still have a partner in range —
# a full-library rescan is the manual button's job, not the sweep's.
horizon = now - timedelta(hours=window_hours * 2)
sort_key = func.coalesce(Post.post_date, Post.downloaded_at)
ids = (await session.execute(
select(Post.id).where(
Post.synthesized_by.is_(None),
Post.absorbed_by_post_id.is_(None),
sort_key >= horizon,
)
)).scalars().all()
svc = PostAssociationService(session)
proposed = 0
for pid in ids:
proposed += await svc.match_post(
pid,
threshold=float(settings.discord_link_threshold),
window_hours=window_hours,
)
log.info(
"discord announcement matcher: scanned %d post(s), proposed %d pair(s)",
len(ids), proposed,
)
return {"enabled": True, "scanned": len(ids), "proposed": proposed}