feat: platform_membership — the learned roster of what the account pays for (milestone 387 step C1)
CI / extension-version (push) Successful in 3s
CI / lint (push) Successful in 3s
Build images / sign-extension (push) Successful in 4s
Build images / build-agent (push) Successful in 8s
CI / frontend-build (push) Successful in 23s
CI / backend-lint-and-test (push) Successful in 39s
Build images / build-web (push) Successful in 1m14s
Build images / smoke-web (push) Skipped
Build images / build-ml (push) Successful in 2m18s
Build images / promote (push) Skipped
CI / integration (push) Successful in 2m21s

FC knows which creators it was TOLD to follow and nothing about which
ones the operator is subscribed to. Those two sets drift both ways and
neither drift is currently visible: a subscription FC doesn't track is
content the operator believes they're archiving and aren't, and a
source walked after the subscription lapsed is requests spent on a wall
reported as a creator gone quiet.

Sibling of service_seen (milestone 365) and the same insight — an
absence is only observable against a record of presence. touch_membership
reuses the recorded touch_service shape (snippet 3447): upsert rather
than read-modify-write, first_seen_at deliberately outside the update
set because it's the one field that makes a later DISAPPEARANCE
readable as a lapse rather than as a creator we never knew.

Nothing populates it yet, and that's the intended intermediate state.
The sweep (C3) needs a client seam (C2) that needs Patreon's real
response characterised from a captured sample (C0), which needs the
operator's browser session. The table's SHAPE doesn't wait on that,
because it's deliberately free-form exactly where C0's findings would
otherwise dictate a column.

status is an unconstrained String holding the PLATFORM's own word, not
a normalised FC value. Rule 36 considered and declined, same reasoning
service_seen.kind records: the vocabulary isn't ours to invent, and
picking a lowest-common-denominator enum before any platform has been
characterised would bake a guess into the schema. The service owns the
whitelist and the mapping; the column owns the evidence.

MEMBERSHIP_STATUS ships EMPTY, guarded by a test that fails if anyone
adds an entry — every one must come from a characterised response, not
from API docs. That's rule 130 at the one place it's easiest to break,
and the failure message says so.

has_paid_access returns None, never False, for a word it hasn't been
taught. The difference is load-bearing: False means the operator lost
access, which C4 turns into an offer to disable the source, so
asserting it from an unrecognised word would tell them to cancel a
subscription they're still paying for.

Retention decided here rather than deferred (rule 89): a membership
that stops appearing is aged out on time, never deleted on absence —
deleting would destroy the signal at the moment it became interesting.

Rule 90 check, done on the right thing this time: the per-test TRUNCATE
teardown derives its table list from Base.metadata.sorted_tables, so
the new table is picked up automatically; test_models asserts a subset,
so it doesn't break. 0091 follows 0090 on the collapsed baseline.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LNXXULQDjVZmbuNa2G9mD9
This commit is contained in:
2026-09-10 10:58:25 -04:00
co-authored by Claude Opus 5
parent 6b19012bb6
commit 4fe792b61c
5 changed files with 504 additions and 0 deletions
@@ -0,0 +1,81 @@
"""platform_membership — the learned roster of what the account actually pays for.
Milestone 387, phase C. FC knows which creators it was told to follow and
nothing about which ones the operator is subscribed to; this table is the
memory that makes the drift in both directions observable. See the model
docstring for why the roster is learned rather than looked up live, and why
`status` holds the platform's own word rather than a normalised FC value.
## Nothing populates this yet, on purpose
The sweep that fills it (C3) depends on a client seam (C2) that depends on
characterising Patreon's real membership response from a captured sample (C0),
which needs the operator's authenticated browser session. The table's SHAPE
does not wait on that: it is deliberately free-form where C0's findings would
otherwise dictate a column — `status` is an unconstrained String and `details`
keeps the raw payload — so no capture can invalidate what is created here.
An empty table is the correct intermediate state. It is not dead code: C5 reads
it to explain a tier-limited source, and C4 reads it to reconcile.
Revision ID: 0091
Revises: 0090
Create Date: 2026-09-10
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
revision: str = "0091"
down_revision: Union[str, None] = "0090"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
op.create_table(
"platform_membership",
sa.Column("id", sa.Integer(), nullable=False),
sa.Column("platform", sa.String(length=64), nullable=False),
# Text, not a bounded String: an opaque upstream identifier we do not
# mint, and guessing a ceiling for one is how a walk dies on a silent
# truncation.
sa.Column("external_campaign_id", sa.Text(), nullable=False),
sa.Column("display_name", sa.Text(), nullable=True),
sa.Column("url", sa.Text(), nullable=True),
# No CHECK, deliberately (rule 36 considered and declined): the
# vocabulary is each platform's own and is not ours to fix before C0
# has characterised even one of them. The service owns the whitelist.
sa.Column("status", sa.String(length=32), nullable=True),
sa.Column("tier_names", sa.JSON(), nullable=True),
sa.Column("amount_cents", sa.Integer(), nullable=True),
sa.Column("currency", sa.String(length=8), nullable=True),
sa.Column(
"first_seen_at", sa.DateTime(timezone=True),
server_default=sa.text("now()"), nullable=False,
),
sa.Column(
"last_seen_at", sa.DateTime(timezone=True),
server_default=sa.text("now()"), nullable=False,
),
sa.Column("details", sa.JSON(), nullable=False),
sa.PrimaryKeyConstraint("id", name=op.f("pk_platform_membership")),
# The upsert's conflict target. Named explicitly because
# touch_membership references it by name in ON CONFLICT — an
# autogenerated name would make that call break on a rename nobody
# connected to it.
sa.UniqueConstraint(
"platform", "external_campaign_id",
name="uq_platform_membership_platform_campaign",
),
)
# No secondary indexes. This table holds one row per subscription — tens,
# not millions — so every query against it is a short scan and an index
# would be write cost buying nothing (#3301 removed seven of that shape).
# The unique constraint above already backs the only lookup that matters.
def downgrade() -> None:
op.drop_table("platform_membership")
+2
View File
@@ -26,6 +26,7 @@ from .patreon_failed_media import PatreonFailedMedia
from .patreon_seen_media import PatreonSeenMedia from .patreon_seen_media import PatreonSeenMedia
from .pixiv_failed_media import PixivFailedMedia from .pixiv_failed_media import PixivFailedMedia
from .pixiv_seen_media import PixivSeenMedia from .pixiv_seen_media import PixivSeenMedia
from .platform_membership import PlatformMembership
from .post import Post from .post import Post
from .post_attachment import PostAttachment, attachment_download_url from .post_attachment import PostAttachment, attachment_download_url
from .presentation_review import PresentationReview from .presentation_review import PresentationReview
@@ -64,6 +65,7 @@ __all__ = [
"SeriesChapter", "SeriesChapter",
"SeriesPage", "SeriesPage",
"SeriesSuggestion", "SeriesSuggestion",
"PlatformMembership",
"ServiceSeen", "ServiceSeen",
"ImageRecord", "ImageRecord",
"ImageProvenance", "ImageProvenance",
+130
View File
@@ -0,0 +1,130 @@
"""platform_membership — the learned roster of what the account actually pays for.
Milestone 387, phase C. FabledCurator knows which creators it has been TOLD to
follow (`source`), and nothing about which ones the operator is actually
subscribed to. Those two sets drift in both directions and the app cannot
currently see either drift:
* A subscription the operator pays for that FC does not track is content they
believe they are archiving and are not.
* A source FC keeps walking after the subscription lapsed is requests spent on
a wall, reported as a creator who has gone quiet.
This table is the memory that makes both visible — every membership the account
has been observed to hold, and when it was last seen.
## Why a learned roster rather than a live lookup
Same reasoning as `service_seen` (milestone 365), and the same shape: an
absence is only observable against a record of presence. A membership that
stops appearing in a sweep is the signal — "you were subscribed to this, now
you aren't" — and there is nowhere to read that from a live call, because a
live call returns what IS, never what stopped being.
It also means the reconciliation surface keeps working when Patreon is
unreachable, degraded to a stale roster with a visible age rather than an empty
page (rule 164).
## Roster truth, NOT per-post truth
The single most important thing about this table: `tier_names` says which tiers
the account holds. It does **not** say which posts those tiers unlock. A
creator can gate a post behind an access rule that maps onto no tier name at
all.
`current_user_can_view` — read per post by `patreon_client.post_is_gated` — is
the authoritative signal, and phase A already turned it into a durable
per-source state. This roster EXPLAINS that state ("you are no longer a patron"
vs "your tier doesn't cover these posts"). It must never be used to decide
whether to fetch something. Getting that backwards would make FC silently stop
fetching content the operator is paying for, which is the worst failure
available in this milestone.
## status is a plain String, and deliberately the platform's own word
Not a Postgres ENUM, not CHECK-gated — matching `service_seen.kind`,
`gpu_job.status` and `source.error_type`. Two reasons, and the first is the
real one:
1. **The vocabulary is not ours to invent.** Patreon says `active_patron` /
`former_patron` / `declined_patron`; SubscribeStar and FANBOX will say
something else. Storing each platform's own word verbatim and mapping to
FC's meaning at the READ site keeps this table a record of what was
observed rather than a lossy translation of it. A lowest-common-denominator
enum picked before any platform has been characterised (step C0) would be a
guess baked into the schema.
2. A constraint swap per new value (rule 36) would be cost with no invariant
behind it, exactly as `service_seen.kind` records.
The service layer owns the whitelist and the mapping; the column owns the
evidence.
## Retention: aged out, never deleted on disappearance
A membership that stops appearing in a sweep is NOT removed. Its disappearance
is the fact the reconciliation surface reads, and deleting the row would
destroy the signal at the moment it became interesting. `last_seen_at` is what
makes "gone" decidable, and a retention policy ages rows out on time rather
than on absence.
"""
from datetime import datetime
from sqlalchemy import JSON, DateTime, Integer, String, Text, UniqueConstraint, func
from sqlalchemy.orm import Mapped, mapped_column
from .base import Base
class PlatformMembership(Base):
__tablename__ = "platform_membership"
__table_args__ = (
# The natural key the sweep's upsert conflicts on. Named explicitly
# because `touch_membership` references it by name in ON CONFLICT.
UniqueConstraint(
"platform", "external_campaign_id",
name="uq_platform_membership_platform_campaign",
),
)
id: Mapped[int] = mapped_column(Integer, primary_key=True)
platform: Mapped[str] = mapped_column(String(64), nullable=False)
# The platform's own id for the thing subscribed to — a Patreon campaign
# id, whatever SubscribeStar and FANBOX call theirs. Text rather than a
# bounded String: these are opaque upstream identifiers and guessing a
# ceiling for a value we do not mint is how a walk dies on a truncation.
external_campaign_id: Mapped[str] = mapped_column(Text, nullable=False)
# For the reconciliation UI, and for matching against Source.url — the
# vanity/URL is what the two sides actually have in common.
display_name: Mapped[str | None] = mapped_column(Text, nullable=True)
url: Mapped[str | None] = mapped_column(Text, nullable=True)
# The platform's own word. See the module docstring — this is evidence,
# not a normalised FC status.
status: Mapped[str | None] = mapped_column(String(32), nullable=True)
# Nullable throughout: a free follow has no tier and no money attached, and
# a platform may not expose an amount at all. Absent must stay
# distinguishable from zero — "free" and "we don't know" are different
# answers to "what is this costing".
tier_names: Mapped[list | None] = mapped_column(JSON, nullable=True)
amount_cents: Mapped[int | None] = mapped_column(Integer, nullable=True)
currency: Mapped[str | None] = mapped_column(String(8), nullable=True)
# NEVER updated after insert. The one field that answers "has this ever
# been true", which is what makes a disappearance readable rather than
# indistinguishable from never having existed. `touch_membership`
# deliberately excludes it from the ON CONFLICT update set.
first_seen_at: Mapped[datetime] = mapped_column(
DateTime(timezone=True), nullable=False, server_default=func.now(),
)
last_seen_at: Mapped[datetime] = mapped_column(
DateTime(timezone=True), nullable=False, server_default=func.now(),
)
# The raw membership as the platform returned it, so a later question can
# be answered without re-fetching — and so a field we did not think to
# model is not lost. Displayed and never queried, like service_seen.details.
details: Mapped[dict] = mapped_column(JSON, nullable=False, default=dict)
+116
View File
@@ -0,0 +1,116 @@
"""The learned membership roster: what the account actually subscribes to.
Milestone 387, phase C. Sibling of `service_roster` (milestone 365) and built
on the same insight — an absence is only observable against a record of
presence. There, a stopped worker; here, a subscription that lapsed.
## Nothing calls this yet
`touch_membership` is written before its caller because the caller (the sweep,
C3) needs a client seam (C2) that needs Patreon's real response characterised
from a captured sample (C0), and that capture needs the operator's browser
session. The write side does not depend on any of it: an upsert keyed on
(platform, external_campaign_id) is the same regardless of what the payload
turns out to look like, and `details` carries whatever C0 finds.
## Why the whitelist lives here and not in the column
`platform_membership.status` is an unconstrained String holding the PLATFORM's
own word — `active_patron`, not some normalised FC value. The mapping from
those words to FC's meaning is a read-site concern and belongs in code that can
be corrected without a migration, because the vocabulary comes from whatever
each platform says and will be discovered per platform rather than designed up
front. `MEMBERSHIP_STATUS` below is a place for that knowledge to accumulate as
platforms are characterised; it is deliberately empty of guesses today.
"""
from __future__ import annotations
import logging
from sqlalchemy import func
from sqlalchemy.dialects.postgresql import insert as pg_insert
from sqlalchemy.ext.asyncio import AsyncSession
from ..models import PlatformMembership
log = logging.getLogger(__name__)
# Platform word -> whether the account currently has paid access.
#
# EMPTY ON PURPOSE. Every entry here must come from a characterised response
# (step C0), not from what the API docs or a plausible guess suggest — that is
# the whole point of project rule 130, and inventing `active_patron` before
# seeing it in a real payload is exactly the failure it names. Populate per
# platform as each is characterised.
#
# Unknown words are NOT an error: an unrecognised status means the roster
# records evidence it cannot yet interpret, which is a better state than
# dropping the row or asserting a meaning for it.
MEMBERSHIP_STATUS: dict[str, dict[str, bool]] = {}
def has_paid_access(platform: str, status: str | None) -> bool | None:
"""Does this status mean the account currently pays for access?
Returns None for a status this code has not been taught, which callers must
treat as "unknown" rather than as False. The difference matters: False says
the operator has lost access, and asserting that from an unrecognised word
would tell them to cancel a source they are still paying for.
"""
if status is None:
return None
entry = MEMBERSHIP_STATUS.get(platform, {})
return entry.get(status)
async def touch_membership(
session: AsyncSession,
*,
platform: str,
external_campaign_id: str,
display_name: str | None = None,
url: str | None = None,
status: str | None = None,
tier_names: list | None = None,
amount_cents: int | None = None,
currency: str | None = None,
details: dict | None = None,
) -> None:
"""Record that this membership was observed just now.
Upsert rather than read-modify-write, for the same reason as
`service_roster.touch_service`: a sweep may overlap its own previous run,
and the last writer is simply the most recent sighting.
`first_seen_at` is deliberately NOT in the update set. It is the one field
that answers "has this ever been true", which is what makes a membership's
later DISAPPEARANCE readable as a lapse rather than indistinguishable from
a creator FC never knew about. Every other column is last-writer-wins,
including status — a membership that goes from active to former must move.
"""
stmt = pg_insert(PlatformMembership).values(
platform=platform,
external_campaign_id=external_campaign_id,
display_name=display_name,
url=url,
status=status,
tier_names=tier_names,
amount_cents=amount_cents,
currency=currency,
details=details or {},
)
stmt = stmt.on_conflict_do_update(
constraint="uq_platform_membership_platform_campaign",
set_={
"display_name": stmt.excluded.display_name,
"url": stmt.excluded.url,
"status": stmt.excluded.status,
"tier_names": stmt.excluded.tier_names,
"amount_cents": stmt.excluded.amount_cents,
"currency": stmt.excluded.currency,
"details": stmt.excluded.details,
"last_seen_at": func.now(),
},
)
await session.execute(stmt)
+175
View File
@@ -0,0 +1,175 @@
"""Milestone 387 C1: the learned membership roster.
The load-bearing property is that `first_seen_at` survives every re-observation
— it is what makes a membership's later DISAPPEARANCE readable as a lapse
rather than indistinguishable from a creator FC never knew about. Everything
else is last-writer-wins, including status, because a membership that goes from
active to former has to move.
"""
import pytest
from sqlalchemy import select
from backend.app.models import PlatformMembership
from backend.app.services.membership_roster import (
MEMBERSHIP_STATUS,
has_paid_access,
touch_membership,
)
pytestmark = pytest.mark.integration
async def _row(db, platform="patreon", campaign="c1"):
return (await db.execute(
select(PlatformMembership).where(
PlatformMembership.platform == platform,
PlatformMembership.external_campaign_id == campaign,
)
)).scalar_one()
@pytest.mark.asyncio
async def test_touch_inserts_a_new_membership(db):
await touch_membership(
db, platform="patreon", external_campaign_id="c-new",
display_name="Maewix Studios", url="https://patreon.com/maewix",
status="some_platform_word", tier_names=["Sketches"],
amount_cents=500, currency="USD", details={"raw": 1},
)
await db.commit()
row = await _row(db, campaign="c-new")
assert row.display_name == "Maewix Studios"
assert row.status == "some_platform_word"
assert row.tier_names == ["Sketches"]
assert row.amount_cents == 500
assert row.details == {"raw": 1}
assert row.first_seen_at is not None
@pytest.mark.asyncio
async def test_re_observation_preserves_first_seen_and_advances_last_seen(db):
"""The whole point of the table.
Committed between touches deliberately: `func.now()` is the TRANSACTION
timestamp in Postgres, so two touches in one transaction would share a
last_seen_at and this test would pass without proving anything. (That
sharing is correct for a sweep — every row it touches is one observation —
but it makes an in-transaction assertion vacuous.)
"""
await touch_membership(
db, platform="patreon", external_campaign_id="c-again", status="active_ish",
)
await db.commit()
original = await _row(db, campaign="c-again")
first_seen, first_last_seen = original.first_seen_at, original.last_seen_at
db.expunge_all()
await touch_membership(
db, platform="patreon", external_campaign_id="c-again", status="former_ish",
)
await db.commit()
row = await _row(db, campaign="c-again")
assert row.first_seen_at == first_seen, "first_seen_at must never move"
assert row.last_seen_at >= first_last_seen
# Status is last-writer-wins: a lapse has to be able to overwrite.
assert row.status == "former_ish"
@pytest.mark.asyncio
async def test_re_observation_overwrites_the_mutable_fields(db):
"""A creator who renames, retiers or changes price must not leave the
roster asserting the old value — every column except first_seen_at moves."""
await touch_membership(
db, platform="patreon", external_campaign_id="c-mut",
display_name="Old Name", amount_cents=500, tier_names=["Cheap"],
details={"v": 1},
)
await db.commit()
db.expunge_all()
await touch_membership(
db, platform="patreon", external_campaign_id="c-mut",
display_name="New Name", amount_cents=1500, tier_names=["Pricey"],
details={"v": 2},
)
await db.commit()
row = await _row(db, campaign="c-mut")
assert row.display_name == "New Name"
assert row.amount_cents == 1500
assert row.tier_names == ["Pricey"]
assert row.details == {"v": 2}
@pytest.mark.asyncio
async def test_the_same_campaign_id_on_two_platforms_is_two_rows(db):
"""The key is (platform, external_campaign_id). Nothing stops two platforms
minting the same opaque id, and collapsing them would merge one creator's
membership into another's."""
await touch_membership(db, platform="patreon", external_campaign_id="shared-id")
await touch_membership(db, platform="subscribestar", external_campaign_id="shared-id")
await db.commit()
rows = (await db.execute(
select(PlatformMembership).where(
PlatformMembership.external_campaign_id == "shared-id"
)
)).scalars().all()
assert {r.platform for r in rows} == {"patreon", "subscribestar"}
@pytest.mark.asyncio
async def test_a_free_follow_keeps_absent_distinct_from_zero(db):
""""Free" and "we don't know what this costs" are different answers, and
the reconciliation UI has to be able to tell them apart."""
await touch_membership(
db, platform="patreon", external_campaign_id="c-free", status="free_ish",
)
await db.commit()
row = await _row(db, campaign="c-free")
assert row.amount_cents is None
assert row.tier_names is None
# --- has_paid_access: unknown must never read as "lost access" -------------
def test_unknown_status_is_unknown_not_false():
"""The dangerous case. False means "the operator lost access", which C4
turns into an offer to disable the source. Asserting that from a word this
code simply has not been taught would tell them to cancel a subscription
they are still paying for."""
assert has_paid_access("patreon", "a_word_nobody_characterised_yet") is None
def test_absent_status_is_unknown():
assert has_paid_access("patreon", None) is None
def test_unknown_platform_is_unknown():
assert has_paid_access("a-platform-with-no-mapping", "active") is None
def test_the_status_map_starts_empty_and_that_is_deliberate():
"""Guards project rule 130 at the one place it is easiest to break.
Every entry must come from a characterised response (step C0), never from
API docs or a plausible-looking guess. If this assertion fails, either C0
happened — in which case update this test along with the map, citing the
capture — or somebody guessed, which is the thing the rule exists to stop.
"""
assert MEMBERSHIP_STATUS == {}
def test_the_map_is_consulted_once_it_has_entries(monkeypatch):
"""The map is empty today, so exercise the lookup with a stand-in — proving
the plumbing works without pretending to know a real platform's word."""
monkeypatch.setitem(
MEMBERSHIP_STATUS, "testplat", {"paying": True, "lapsed": False},
)
assert has_paid_access("testplat", "paying") is True
assert has_paid_access("testplat", "lapsed") is False
assert has_paid_access("testplat", "something_else") is None