fix(auth): match identity on a canonical email, so one Gmail mailbox is one account (E9.5 S3)

Gmail delivers x.y@, xy@ and xy+tag@ to one inbox; the invite gate only lowercased.
On 2026-09-03 that turned an approved user's Google sign-in into a silent SECOND
access request -- OAuth returns the account's stored primary, which need not be the
spelling they typed -- and left prod with two user rows for one mailbox (ids 13, 14).
The +tag case fails every time: a Google account cannot contain +, so OAuth can only
ever answer with the base address.

Mail still goes to the address AS TYPED; only identity matching uses the canonical
form. utils.canonical_email is the single definition, and a model validator derives
users.email_canonical / invites.email_canonical on every write, so no call site can
forget it. Scope is narrow on purpose: dots are significant on most providers and the
Gmail rule does not hold for Workspace custom domains.

Swept every sibling of the same principle, not just the reported one: is_allowed,
upsert_pending_invite, the Google-callback adoption, _register_account (so no new
duplicate pair can be created), _activate_user_by_email, admin _accounts_for (an
invite backing a live user under another spelling used to look deletable) and
add_invite, plus purge_user, which now erases every canonically-equal invite instead
of leaving one behind that still grants access.

Migration 0079 backfills both columns and indexes them NON-uniquely: prod already
carries canonical duplicates (approved invite pairs since 2026-06-16), and merging
accounts is a judgement call about people's data, not something a migration does on
its own. A test pins the backfill SQL against the Python helper.

backend/app/auth.py

@@ -12,6 +12,7 @@ from app.config import settings
 from app.db import get_db
 from app.models import Invite, User
 from app.security import decrypt_optional
+from app.utils import canonical_email
 
 PASSWORD_MIN_LEN = 10
 
@@ -65,13 +66,29 @@ def has_write_scope(user: User) -> bool:
 
 def is_allowed(db: Session, email: str) -> bool:
     """May this email sign in? An approved Invite is the source of truth; the env
-    ALLOWED_EMAILS/ADMIN_EMAILS remain a bootstrap fallback (e.g. a fresh DB)."""
+    ALLOWED_EMAILS/ADMIN_EMAILS remain a bootstrap fallback (e.g. a fresh DB).
+
+    Matched on the CANONICAL form (utils.canonical_email), not the literal string. Gmail delivers
+    every dotted / `+tagged` spelling to one mailbox, so approving one spelling always meant approving
+    that mailbox — matching literally only produced the 2026-09-03 lockout, where an approved user's
+    Google sign-in became a silent second access request because OAuth returned the undotted primary.
+
+    One canonical form can match SEVERAL invite rows (prod carries approved pairs for one mailbox,
+    the oldest from 2026-06-16). An approved row among them grants access: the API cannot flip an
+    approved invite to denied (`_decide` only accepts a pending one), so approved-beside-denied means
+    the admin approved that mailbox under a different spelling, not that they revoked it. Revoking is
+    deleting the invite row or suspending the account — both still work, since a deleted row is no
+    longer here to match and suspension is enforced at sign-in.
+    """
     if not email:
         return False
+    canon = canonical_email(email)
+    statuses = (
+        db.execute(select(Invite.status).where(Invite.email_canonical == canon)).scalars().all()
+    )
+    if statuses:
+        return "approved" in statuses
     email = email.lower()
-    inv = db.execute(select(Invite).where(Invite.email == email)).scalar_one_or_none()
-    if inv is not None:
-        return inv.status == "approved"
     return email in settings.allowed_email_set or email in settings.admin_email_set
 
 
@@ -130,7 +147,10 @@ def purge_user(db: Session, user: User, background: BackgroundTasks) -> None:
     )
     # Raw delete so Postgres applies ON DELETE CASCADE / SET NULL on every dependent table.
     db.execute(delete(User).where(User.id == user_id))
-    db.execute(delete(Invite).where(Invite.email == email))
+    # Every canonically-equal invite goes, not just the literal spelling: is_allowed now matches on
+    # the canonical form, so leaving a sibling row behind would keep granting access to a mailbox
+    # whose account was just erased — and prod really does carry approved pairs for one mailbox.
+    db.execute(delete(Invite).where(Invite.email_canonical == canonical_email(email)))
     db.commit()
     if google_token:
         background.add_task(revoke_google_token, google_token)

backend/app/models.py

@@ -17,9 +17,10 @@ from sqlalchemy import (
     func,
 )
 from sqlalchemy.dialects.postgresql import JSONB, TSVECTOR
-from sqlalchemy.orm import Mapped, mapped_column, relationship
+from sqlalchemy.orm import Mapped, mapped_column, relationship, validates
 
 from app.db import Base
+from app.utils import canonical_email
 
 
 class TimestampMixin:
@@ -29,6 +30,21 @@ class TimestampMixin:
     created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
 
 
+class CanonicalEmailMixin:
+    """Keeps `email_canonical` derived from `email` on every write.
+
+    The invariant — "identity is matched on the canonical form, mail is sent to the typed one" —
+    lives HERE, in code, instead of in a comment every write site has to remember. That distinction
+    is what the 2026-09-03 lockout was made of: one gate lowercased, nothing else did, and an
+    approved user's Google sign-in silently became a second access request.
+    """
+
+    @validates("email")
+    def _derive_email_canonical(self, _key: str, value: str | None) -> str | None:
+        self.email_canonical = canonical_email(value)
+        return value
+
+
 class UpdatedAtMixin:
     """Adds an `updated_at` that server-defaults on insert and bumps on every UPDATE."""
 
@@ -37,7 +53,7 @@ class UpdatedAtMixin:
     )
 
 
-class User(Base, TimestampMixin):
+class User(Base, TimestampMixin, CanonicalEmailMixin):
     __tablename__ = "users"
 
     id: Mapped[int] = mapped_column(primary_key=True)
@@ -45,6 +61,11 @@ class User(Base, TimestampMixin):
     # multiple NULLs under a unique index). Set on Google sign-in / in-app linking.
     google_sub: Mapped[str | None] = mapped_column(String(64), unique=True, index=True)
     email: Mapped[str] = mapped_column(String(320), unique=True, index=True)
+    # The same mailbox, in the form used to MATCH identities (utils.canonical_email) — Gmail dots and
+    # `+tags` folded away. Kept by the validator below so no write site can forget it. Deliberately
+    # NOT unique: prod already carries two rows for one mailbox (a password registration and the
+    # Google sign-in that failed to find it), and a migration must not merge accounts on its own.
+    email_canonical: Mapped[str] = mapped_column(String(320), index=True, server_default="")
     # argon2id hash for email+password login; NULL for Google-only / demo accounts.
     password_hash: Mapped[str | None] = mapped_column(String(255))
     # When SEC-10 dropped a preset password on Google adoption (see the single caller of
@@ -121,7 +142,7 @@ class AuthToken(Base, TimestampMixin):
     used_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
 
 
-class Invite(Base):
+class Invite(Base, CanonicalEmailMixin):
     """Access-request / whitelist row. Source of truth for who may sign in; the env
     ALLOWED_EMAILS/ADMIN_EMAILS act only as a bootstrap fallback (see auth.is_allowed)."""
 
@@ -129,6 +150,9 @@ class Invite(Base):
 
     id: Mapped[int] = mapped_column(primary_key=True)
     email: Mapped[str] = mapped_column(String(320), unique=True, index=True)
+    # See User.email_canonical. Not unique here either: prod has approved invites for both spellings
+    # of the same mailbox, dating back to 2026-06-16.
+    email_canonical: Mapped[str] = mapped_column(String(320), index=True, server_default="")
     # pending | approved | denied
     status: Mapped[str] = mapped_column(String(16), default="pending", server_default="pending")
     note: Mapped[str | None] = mapped_column(Text)

backend/app/routes/admin.py

@@ -27,20 +27,27 @@ from app.models import (
     Invite,
     User,
 )
-from app.utils import iso, valid_email
+from app.utils import canonical_email, iso, valid_email
 
 router = APIRouter(prefix="/api/admin", tags=["admin"])
 
 
 def _accounts_for(db: Session, emails: Iterable[str]) -> set[str]:
-    """Which of these addresses already have an account. Batched, so listing the invites costs one
-    extra query. An invite whose address has an account must not be deletable: `is_allowed` reads
-    this row, so removing it would silently revoke a live user's access."""
-    wanted = {e.lower() for e in emails if e}
+    """Which of these addresses already have an account, as a set of CANONICAL forms. Batched, so
+    listing the invites costs one extra query. An invite whose address has an account must not be
+    deletable: `is_allowed` reads this row, so removing it would silently revoke a live user's access.
+
+    Canonical on both sides, because `is_allowed` matches that way: an invite for `x.y@gmail.com`
+    whose account registered as `xy@gmail.com` would otherwise look account-less, offer the admin a
+    delete button, and take away a live user's access — precisely the accident this guard exists to
+    prevent."""
+    wanted = {canonical_email(e) for e in emails if e}
     if not wanted:
         return set()
-    rows = db.execute(select(User.email).where(User.email.in_(wanted))).scalars()
-    return {e.lower() for e in rows if e}
+    rows = db.execute(
+        select(User.email_canonical).where(User.email_canonical.in_(wanted))
+    ).scalars()
+    return {e for e in rows if e}
 
 
 def _env_allowed(email: str) -> bool:
@@ -82,7 +89,7 @@ def list_invites(
     order = {"pending": 0, "approved": 1, "denied": 2}
     rows.sort(key=lambda i: order.get(i.status, 9))
     accounts = _accounts_for(db, (i.email for i in rows))
-    return [_serialize(i, has_account=(i.email or "").lower() in accounts) for i in rows]
+    return [_serialize(i, has_account=canonical_email(i.email) in accounts) for i in rows]
 
 
 def _activate_user_by_email(db: Session, email: str) -> None:
@@ -97,7 +104,18 @@ def _activate_user_by_email(db: Session, email: str) -> None:
     DEFAULT experience, because approve-before-verify is the normal ordering. The revoke stays on
     the Google-adoption path (auth.revoke_preset_password_if_unverified), where the real owner does
     attest ownership and the preset password must not survive."""
-    user = db.execute(select(User).where(User.email == email.lower())).scalar_one_or_none()
+    # Canonical, with the literal spelling preferred when both exist: approving
+    # `npeter83.temu@gmail.com` must activate the account that registered as `npeter83temu@gmail.com`,
+    # since Gmail delivers both to one mailbox and the admin decided about the mailbox.
+    email = email.lower()
+    candidates = (
+        db.execute(
+            select(User).where(User.email_canonical == canonical_email(email)).order_by(User.id)
+        )
+        .scalars()
+        .all()
+    )
+    user = next((u for u in candidates if u.email == email), None) or next(iter(candidates), None)
     if user is not None and not user.is_active:
         user.is_active = True
         db.commit()
@@ -169,7 +187,12 @@ def add_invite(
     email = body.email.strip().lower()
     if not valid_email(email):
         raise HTTPException(status_code=400, detail="Enter a valid email address.")
-    inv = db.execute(select(Invite).where(Invite.email == email)).scalar_one_or_none()
+    # Reuse a canonically-equal row instead of adding a second one for the same mailbox. Prod carries
+    # exactly that pair for one person from 2026-06-16 (hartmann.peter1986@ / hartmannpeter1986@),
+    # which is how an admin ends up approving the same human twice and still seeing them bounce.
+    canon = canonical_email(email)
+    rows = db.execute(select(Invite).where(Invite.email_canonical == canon)).scalars().all()
+    inv = next((r for r in rows if r.email == email), None) or next(iter(rows), None)
     if inv is None:
         inv = Invite(email=email)
         db.add(inv)

backend/app/routes/auth.py

@@ -37,7 +37,7 @@ from app.db import SessionLocal, get_db
 from app.models import AuthToken, DemoWhitelist, Invite, OAuthToken, User
 from app.ratelimit import RateLimiter
 from app.security import encrypt, hash_password, hash_token, verify_password
-from app.utils import valid_email
+from app.utils import canonical_email, valid_email
 
 VERIFY_TTL = timedelta(hours=24)
 RESET_TTL = timedelta(hours=1)
@@ -149,11 +149,21 @@ def _client_ip(request: Request) -> str:
 
 def upsert_pending_invite(db: Session, email: str) -> Invite | None:
     """Idempotently record an access request. Returns the Invite if a *new* pending row
-    was created (so the caller can notify admins), else None for already-known emails."""
+    was created (so the caller can notify admins), else None for already-known emails.
+
+    "Already known" is decided on the CANONICAL form: `npeter83.temu@` and `npeter83temu@` are one
+    Gmail mailbox, and treating them as two is what put a second, silent access request in the admin
+    queue on 2026-09-03 for an already-approved user. An approved sibling wins — re-requesting must
+    not reopen a decision that has already been made for that mailbox. The row is still STORED under
+    the address as typed, because that is where mail has to go.
+    """
     email = (email or "").lower()
     if not valid_email(email):
         return None
-    inv = db.execute(select(Invite).where(Invite.email == email)).scalar_one_or_none()
+    canon = canonical_email(email)
+    rows = db.execute(select(Invite).where(Invite.email_canonical == canon)).scalars().all()
+    # Prefer a decided row over a pending one, so a denied duplicate can't mask an approval.
+    inv = next((r for r in rows if r.status == "approved"), None) or next(iter(rows), None)
     if inv is not None:
         # Let a previously-denied person ask again; don't disturb approved/pending rows.
         if inv.status == "denied":
@@ -292,11 +302,21 @@ async def callback(
 
     user = db.query(User).filter(User.google_sub == sub).one_or_none()
     if user is None:
-        # No account carries this Google identity yet. If one already exists for this email —
+        # No account carries this Google identity yet. If one already exists for this MAILBOX —
         # e.g. an email+password registration — adopt the Google identity onto it instead of
-        # inserting a duplicate (which would collide on the unique email and 500). Google has
-        # verified the address, so this safely unifies password + Google sign-in on one account.
-        existing = db.query(User).filter(User.email == email).one_or_none()
+        # inserting a duplicate. Google has verified the address, so this safely unifies password +
+        # Google sign-in on one account.
+        #
+        # Matched canonically, and that is not cosmetic: on 2026-09-03 a registration as
+        # `npeter83.temu@gmail.com` met an OAuth claim of `npeter83temu@gmail.com`, the literal
+        # lookup found nothing, and prod ended up with TWO active user rows for one Gmail mailbox
+        # (ids 13 and 14). Exact spelling still wins when both exist, so an account that already
+        # matches literally is never passed over for a canonical sibling.
+        canon = canonical_email(email)
+        candidates = db.query(User).filter(User.email_canonical == canon).order_by(User.id).all()
+        existing = next((u for u in candidates if u.email == email), None) or next(
+            iter(candidates), None
+        )
         if existing is not None:
             if existing.google_sub is not None and existing.google_sub != sub:
                 # A different Google identity is already bound to this email — refuse rather than
@@ -615,7 +635,13 @@ def _register_account(email: str, password: str) -> None:
     notice) for a genuinely new email; no-op for an already-registered one. Runs after the response
     with its own DB session, so the request's timing never reveals whether the email already exists."""
     with SessionLocal() as db:
-        if db.execute(select(User).where(User.email == email)).scalar_one_or_none() is not None:
+        # Canonical, so a second spelling of one Gmail mailbox cannot open a SECOND account. Prod
+        # already carries such a pair (user ids 13 and 14, both for npeter83temu@gmail.com); this is
+        # the end that stops new ones appearing, while the Google callback copes with the existing.
+        existing = db.execute(
+            select(User).where(User.email_canonical == canonical_email(email))
+        ).first()
+        if existing is not None:
             return  # already registered — nothing to do (the uniform "ok" was already returned)
         # Without working SMTP we can't deliver a verification link, so email ownership can't gate
         # sign-in. Admin approval stays the real gate (is_active=False); mark the account verified so

backend/app/utils.py

@@ -22,6 +22,39 @@ def valid_email(value: str | None) -> bool:
     return bool(value and _EMAIL_RE.match(value))
 
 
+# Gmail (and its googlemail.com alias) delivers to one mailbox regardless of dots in the local part
+# and of any `+tag`; nobody else is guaranteed to. So the canonical form below strips those for
+# Gmail only, and merely lowercases everything else.
+_GMAIL_DOMAINS = {"gmail.com", "googlemail.com"}
+
+
+def canonical_email(value: str | None) -> str:
+    """The form used to decide whether two addresses are the SAME MAILBOX. Never used for sending —
+    mail always goes to the address as the user typed it.
+
+    Why this exists (observed on prod, 2026-09-03): someone registered as `npeter83.temu@gmail.com`
+    and was approved, but Google's OAuth `email` claim came back as the account's stored primary,
+    `npeter83temu@gmail.com`. The invite gate only lowercased, found no approved row for that literal
+    string, and turned an approved user's sign-in into a silent SECOND access request — plus a second
+    User row. The `+tag` case is worse and fails every time: a Google account cannot contain `+`, so
+    OAuth can only ever return the base address, meaning anyone who registers with
+    `user+siftlode@gmail.com` (a common throwaway habit) is guaranteed to be denied after approval.
+
+    Scope is deliberately narrow. The dot rule does NOT hold for Google Workspace custom domains
+    unless the admin turned it on, and dots are significant on most providers, so anything outside
+    gmail.com/googlemail.com is only lowercased. Approving one Gmail spelling always did mean
+    approving that mailbox, so collapsing them is the honest reading of an invite, not a widening
+    of it.
+    """
+    email = (value or "").strip().lower()
+    local, sep, domain = email.partition("@")
+    if not sep or domain not in _GMAIL_DOMAINS:
+        return email
+    local = local.partition("+")[0].replace(".", "")
+    # Normalise the alias domain too, so googlemail and gmail spellings meet.
+    return f"{local}@gmail.com" if local else email
+
+
 def now_utc() -> datetime:
     """Current time as a timezone-aware UTC datetime."""
     return datetime.now(timezone.utc)