diff --git a/backend/alembic/versions/a99f438e0010_аллокация_портфеля.py b/backend/alembic/versions/a99f438e0010_аллокация_портфеля.py
new file mode 100644
index 0000000..fbb81de
--- /dev/null
+++ b/backend/alembic/versions/a99f438e0010_аллокация_портфеля.py
@@ -0,0 +1,57 @@
+"""аллокация портфеля
+
+Revision ID: a99f438e0010
+Revises: b7424afbb5e2
+Create Date: 2026-09-18 13:48:30.529174
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "a99f438e0010"
+down_revision: str | None = "b7424afbb5e2"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table(
+        "metric_allocation",
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("scope", sa.String(length=32), nullable=False),
+        sa.Column(
+            "dimension",
+            sa.Enum("asset_class", "sector", "country", "currency", name="allocation_dimension"),
+            nullable=False,
+        ),
+        sa.Column("bucket", sa.String(length=64), nullable=False),
+        sa.Column("value_rub", sa.Numeric(precision=24, scale=10), nullable=False),
+        sa.Column("weight", sa.Numeric(precision=24, scale=10), nullable=False),
+        sa.Column("holding_count", sa.Integer(), nullable=False),
+        sa.Column(
+            "computed_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_metric_allocation")),
+        sa.UniqueConstraint(
+            "scope", "dimension", "bucket", name=op.f("uq_metric_allocation_scope_dimension_bucket")
+        ),
+    )
+    op.create_index(
+        op.f("ix_metric_allocation_scope"), "metric_allocation", ["scope"], unique=False
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_index(op.f("ix_metric_allocation_scope"), table_name="metric_allocation")
+    op.drop_table("metric_allocation")
+    # ### end Alembic commands ###
diff --git a/backend/src/fintracker/analytics/__init__.py b/backend/src/fintracker/analytics/__init__.py
index 2a2c131..eee6350 100644
--- a/backend/src/fintracker/analytics/__init__.py
+++ b/backend/src/fintracker/analytics/__init__.py
@@ -101,6 +101,7 @@ def register_steps() -> None:
     _registered = True
 
     from fintracker.analytics import (
+        allocation,
         cashflow,
         classify,
         networth,
@@ -120,6 +121,7 @@ def register_steps() -> None:
     # valuation prices the positions the lots describe; returns reads the series it writes
     register_step("valuation", valuation.rebuild_valuation)
     register_step("returns", returns.rebuild_returns)
+    register_step("allocation", allocation.rebuild_allocation)
     register_step("networth", networth.rebuild_net_worth_daily)
     register_step("cashflow", cashflow.rebuild_cash_flow_monthly)
     register_step("spending", spending.rebuild_spending_by_category)
diff --git a/backend/src/fintracker/analytics/allocation.py b/backend/src/fintracker/analytics/allocation.py
new file mode 100644
index 0000000..a28d19d
--- /dev/null
+++ b/backend/src/fintracker/analytics/allocation.py
@@ -0,0 +1,215 @@
+"""How the portfolio is split — by asset class, sector, country and currency (plan §3).
+
+Every dimension covers the same total: the securities that could be valued, plus cash. That
+is the whole point of slicing — four pies of the same portfolio must be the same size, or
+the screens contradict each other. So cash is a bucket in each of them, not something left
+out of the ones where it does not obviously belong.
+
+A position nobody quotes is excluded rather than counted as zero, for the reason it always
+is here: a zero would silently shrink every other weight. `holding_without_price` in
+`metric_data_quality` already names those instruments, so this step adds no second remark.
+
+The input is `metric_holding`, which `valuation.py` has just built — instrument attributes
+are joined on top of it, nothing is re-derived from the ledger. Cash is the exception: it
+has no holding row, so it comes from the same derived balances the reconciliation uses.
+"""
+
+from __future__ import annotations
+
+import logging
+from collections import defaultdict
+from collections.abc import Mapping, Sequence
+from dataclasses import dataclass
+from decimal import Decimal
+
+from sqlalchemy import delete, insert, select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from fintracker.analytics import FINDINGS, today_local
+from fintracker.analytics.valuation import account_scopes, cash_balances
+from fintracker.models import (
+    AllocationDimension,
+    Instrument,
+    MetricAllocation,
+    MetricHolding,
+)
+from fintracker.pricing.fx import FxTable
+
+log = logging.getLogger(__name__)
+
+ZERO = Decimal(0)
+
+CASH = "cash"
+"""Bucket for money. A literal, so the client can label it without parsing anything."""
+
+UNKNOWN = "unknown"
+"""Bucket for an instrument whose sector or country nobody filled in — not a guess."""
+
+
+@dataclass(frozen=True)
+class Holding:
+    """What allocation needs from a valued position."""
+
+    instrument_id: int
+    value_rub: Decimal
+    asset_class: str
+    sector: str | None
+    country: str | None
+    currency: str
+
+
+def bucket_of(holding: Holding, dimension: AllocationDimension) -> str:
+    match dimension:
+        case AllocationDimension.asset_class:
+            return holding.asset_class
+        case AllocationDimension.sector:
+            return holding.sector or UNKNOWN
+        case AllocationDimension.country:
+            return holding.country or UNKNOWN
+        case AllocationDimension.currency:
+            return holding.currency.upper()
+
+
+def split(
+    holdings: Sequence[Holding], cash_rub: Mapping[str, Decimal], dimension: AllocationDimension
+) -> list[tuple[str, Decimal, int]]:
+    """One dimension's buckets as (bucket, value, holding count), largest first.
+
+    Cash lands in its own bucket everywhere except the currency dimension, where money in a
+    currency belongs with the papers denominated in it — that is the question the currency
+    chart is actually asked: how much of me is exposed to the dollar.
+    """
+    values: dict[str, Decimal] = defaultdict(lambda: ZERO)
+    counts: dict[str, int] = defaultdict(int)
+    for holding in holdings:
+        key = bucket_of(holding, dimension)
+        values[key] += holding.value_rub
+        counts[key] += 1
+    for ccy, amount in cash_rub.items():
+        if amount == ZERO:
+            continue
+        key = ccy.upper() if dimension is AllocationDimension.currency else CASH
+        values[key] += amount
+    return sorted(
+        ((k, v, counts[k]) for k, v in values.items()),
+        key=lambda row: (-row[1], row[0]),
+    )
+
+
+def weigh(buckets: Sequence[tuple[str, Decimal, int]]) -> list[tuple[str, Decimal, int, Decimal]]:
+    """Add each bucket's share of the total.
+
+    The base is the sum of the POSITIVE buckets. A short position or a negative cash balance
+    is real and keeps its own value, but letting it shrink the denominator would push the
+    long side above 100 %, which means nothing on a pie chart.
+    """
+    base = sum((value for _, value, _ in buckets if value > ZERO), start=ZERO)
+    return [
+        (bucket, value, count, value / base if base > ZERO else ZERO)
+        for bucket, value, count in buckets
+    ]
+
+
+async def rebuild_allocation(session: AsyncSession) -> None:
+    """Replace `metric_allocation` for every scope and dimension."""
+    await session.execute(delete(MetricAllocation))
+
+    rows = (
+        await session.execute(
+            select(
+                MetricHolding.scope,
+                MetricHolding.instrument_id,
+                MetricHolding.value_rub,
+                Instrument.asset_class,
+                Instrument.sector,
+                Instrument.country,
+                Instrument.currency,
+            )
+            .join(Instrument, Instrument.id == MetricHolding.instrument_id)
+            .where(MetricHolding.value_rub.is_not(None))
+        )
+    ).all()
+    per_scope: dict[str, list[Holding]] = defaultdict(list)
+    for r in rows:
+        per_scope[r.scope].append(
+            Holding(
+                instrument_id=r.instrument_id,
+                value_rub=Decimal(r.value_rub),
+                asset_class=str(r.asset_class),
+                sector=r.sector,
+                country=r.country,
+                currency=r.currency,
+            )
+        )
+
+    cash = await _cash_by_scope(session)
+    scopes = sorted(set(per_scope) | set(cash))
+    if not scopes:
+        return
+
+    out: list[dict[str, object]] = []
+    for scope in scopes:
+        for dimension in AllocationDimension:
+            for bucket, value, count, weight in weigh(
+                split(per_scope.get(scope, []), cash.get(scope, {}), dimension)
+            ):
+                out.append(
+                    {
+                        "scope": scope,
+                        "dimension": dimension,
+                        "bucket": bucket,
+                        "value_rub": value,
+                        "weight": weight,
+                        "holding_count": count,
+                    }
+                )
+    if out:
+        await session.execute(insert(MetricAllocation), out)
+    _report_unknown(per_scope.get("all", []))
+    log.info("allocation: %s scopes, %s rows", len(scopes), len(out))
+
+
+def _report_unknown(holdings: Sequence[Holding]) -> None:
+    """A dimension that is mostly `unknown` is a missing attribute, not a portfolio shape.
+
+    Sector is the live case: the instrument sync reads `GetInstrumentBy`, whose response has
+    no sector field at all — it would take the per-type Shares/Bonds calls. Until then the
+    chart is honestly empty, and says so here instead of looking broken.
+    """
+    if not holdings:
+        return
+    for dimension in (AllocationDimension.sector, AllocationDimension.country):
+        missing = [h for h in holdings if bucket_of(h, dimension) == UNKNOWN]
+        if not missing:
+            continue
+        name = "сектор" if dimension is AllocationDimension.sector else "страна"
+        FINDINGS.add(
+            f"instrument_without_{dimension.value}",
+            "info",
+            f"У {len(missing)} из {len(holdings)} инструментов не заполнен {name} — "
+            f"аллокация по этому измерению неполная",
+            count=len(missing),
+            ref={"instruments": sorted(h.instrument_id for h in missing)},
+        )
+
+
+async def _cash_by_scope(session: AsyncSession) -> dict[str, dict[str, Decimal]]:
+    """Derived cash per scope and currency, in RUB at today's rate."""
+    balances = await cash_balances(session)
+    if not balances:
+        return {}
+    fx = await FxTable.load(session)
+    as_of = today_local()
+    scopes = await account_scopes(session, {account_id for account_id, _ in balances})
+
+    out: dict[str, dict[str, Decimal]] = defaultdict(lambda: defaultdict(Decimal))
+    for scope, account_ids in scopes.items():
+        for (account_id, ccy), amount in balances.items():
+            if account_id not in account_ids or amount == ZERO:
+                continue
+            rub = fx.to_rub(amount, ccy, as_of)
+            # no rate today: the money is real but unconvertible, and a substituted number
+            # would be worse than a bucket that quietly omits it (see `missing_fx`)
+            if rub is not None:
+                out[scope][ccy.upper()] += rub
+    return {scope: dict(by_ccy) for scope, by_ccy in out.items()}
diff --git a/backend/src/fintracker/analytics/valuation.py b/backend/src/fintracker/analytics/valuation.py
index ce57a8a..facf31c 100644
--- a/backend/src/fintracker/analytics/valuation.py
+++ b/backend/src/fintracker/analytics/valuation.py
@@ -461,6 +461,27 @@ async def _load_deltas(session: AsyncSession, prices: PriceTable) -> tuple[Delta
     return Deltas(positions, cash, flows, instrument_cash), rows[0].trade_date
 
 
+async def cash_balances(session: AsyncSession) -> dict[tuple[int, str], Decimal]:
+    """Derived cash per (account, currency): the sum of every confirmed event's amount.
+
+    Card-funded trades are left out — the money came from a linked card and the account's
+    balance never saw it, so counting the payment would show an overdraft the broker does not
+    report. This is the one definition of "our cash", used by the reconciliation here and by
+    `analytics/allocation.py`.
+    """
+    rows = (
+        await session.execute(
+            select(Event.account_id, Event.currency, func.sum(Event.amount))
+            .where(
+                Event.status == EventStatus.confirmed,
+                Event.meta["card_funded"].as_string().is_distinct_from("true"),
+            )
+            .group_by(Event.account_id, Event.currency)
+        )
+    ).all()
+    return {(r[0], (r[1] or RUB).upper()): Decimal(r[2] or 0) for r in rows}
+
+
 async def account_scopes(session: AsyncSession, ledger_accounts: set[int]) -> dict[str, set[int]]:
     """Every set of accounts the metrics are reported for: all, each one, each portfolio."""
     scopes: dict[str, set[int]] = {"all": set(ledger_accounts)}
@@ -767,8 +788,7 @@ async def rebuild_valuation(session: AsyncSession) -> None:
     income, income_missing_fx = await _income_by_instrument(session, fx)
     await _write_holdings(session, scopes, positions, realized, income, prices, fx, as_of)
 
-    cash_derived = {key: sum(by_day.values(), start=ZERO) for key, by_day in deltas.cash.items()}
-    await _reconcile(session, cash_derived)
+    await _reconcile(session, await cash_balances(session))
     if income_missing_fx:
         FINDINGS.add(
             "income_missing_fx",
diff --git a/backend/src/fintracker/models/__init__.py b/backend/src/fintracker/models/__init__.py
index 31f45ce..cbce08d 100644
--- a/backend/src/fintracker/models/__init__.py
+++ b/backend/src/fintracker/models/__init__.py
@@ -22,6 +22,8 @@ from fintracker.models.ledger import (
     LotDisposal,
 )
 from fintracker.models.metrics import (
+    AllocationDimension,
+    MetricAllocation,
     MetricCashFlowMonthly,
     MetricDataQuality,
     MetricHolding,
@@ -79,6 +81,7 @@ __all__ = [
     "AccountKind",
     "AccountLink",
     "AccountRole",
+    "AllocationDimension",
     "AppUser",
     "AssetClass",
     "Broker",
@@ -101,6 +104,7 @@ __all__ = [
     "Lot",
     "LotDisposal",
     "Merchant",
+    "MetricAllocation",
     "MetricCashFlowMonthly",
     "MetricDataQuality",
     "MetricHolding",
diff --git a/backend/src/fintracker/models/metrics.py b/backend/src/fintracker/models/metrics.py
index 8faf2c4..3d0486a 100644
--- a/backend/src/fintracker/models/metrics.py
+++ b/backend/src/fintracker/models/metrics.py
@@ -3,6 +3,7 @@ wholesale inside one transaction by metrics/refresh.py; nothing else writes here
 
 from __future__ import annotations
 
+import enum
 from datetime import date, datetime
 from decimal import Decimal
 from typing import Any
@@ -10,7 +11,7 @@ from typing import Any
 from sqlalchemy import ForeignKey, Integer, String, Text, UniqueConstraint, func
 from sqlalchemy.orm import Mapped, mapped_column
 
-from fintracker.db.base import Base
+from fintracker.db.base import Base, db_enum
 
 
 class MetricNetWorthDaily(Base):
@@ -214,3 +215,39 @@ class MetricReturns(Base):
     """Days left out of the chain because the portfolio could not be valued in full on them.
     Non-zero means `twr` covers only part of the period."""
     computed_at: Mapped[datetime] = mapped_column(server_default=func.now())
+
+
+class AllocationDimension(enum.StrEnum):
+    """How a portfolio can be sliced. Every dimension covers the SAME total — securities
+    plus cash — so the weights of any one of them add up to 1 and the charts agree."""
+
+    asset_class = "asset_class"
+    sector = "sector"
+    country = "country"
+    currency = "currency"
+
+
+class MetricAllocation(Base):
+    """Portfolio split per (scope, dimension, bucket). Target weights arrive in phase 4.
+
+    `bucket` is a stable key, not a label: an asset class as stored, a sector or country as
+    the source spells it, a currency code, plus two literals — `cash` for money and `unknown`
+    for an instrument whose attribute nobody filled in. The client decides how to say those
+    in Russian; inventing a label here would bake one language into the data.
+    """
+
+    __tablename__ = "metric_allocation"
+    __table_args__ = (UniqueConstraint("scope", "dimension", "bucket"),)
+
+    id: Mapped[int] = mapped_column(primary_key=True)
+    scope: Mapped[str] = mapped_column(String(32), index=True)
+    dimension: Mapped[AllocationDimension] = mapped_column(
+        db_enum(AllocationDimension, "allocation_dimension")
+    )
+    bucket: Mapped[str] = mapped_column(String(64))
+    value_rub: Mapped[Decimal]
+    weight: Mapped[Decimal]
+    """Share of the scope's valued total; the weights of one dimension add up to 1."""
+    holding_count: Mapped[int] = mapped_column(Integer, default=0)
+    """Instruments in this bucket; 0 for the cash bucket."""
+    computed_at: Mapped[datetime] = mapped_column(server_default=func.now())
diff --git a/backend/tests/analytics/test_allocation.py b/backend/tests/analytics/test_allocation.py
new file mode 100644
index 0000000..729a2a4
--- /dev/null
+++ b/backend/tests/analytics/test_allocation.py
@@ -0,0 +1,88 @@
+"""Allocation rules on synthetic holdings — no database."""
+
+from decimal import Decimal
+
+from fintracker.analytics.allocation import CASH, UNKNOWN, Holding, bucket_of, split, weigh
+from fintracker.models import AllocationDimension
+
+D = Decimal
+DIM = AllocationDimension
+
+
+def holding(
+    instrument_id: int,
+    value: str,
+    *,
+    asset_class: str = "share",
+    sector: str | None = "energy",
+    country: str | None = "RU",
+    currency: str = "RUB",
+) -> Holding:
+    return Holding(
+        instrument_id=instrument_id,
+        value_rub=D(value),
+        asset_class=asset_class,
+        sector=sector,
+        country=country,
+        currency=currency,
+    )
+
+
+def test_buckets_are_summed_per_dimension_and_sorted_by_value():
+    rows = split(
+        [
+            holding(1, "100", asset_class="share"),
+            holding(2, "300", asset_class="bond"),
+            holding(3, "50", asset_class="share"),
+        ],
+        {},
+        DIM.asset_class,
+    )
+    assert rows == [("bond", D(300), 1), ("share", D(150), 2)]
+
+
+def test_cash_is_its_own_bucket_everywhere_but_the_currency_chart():
+    holdings = [holding(1, "900", currency="RUB")]
+    for dimension in (DIM.asset_class, DIM.sector, DIM.country):
+        assert (CASH, D(100), 0) in split(holdings, {"RUB": D(100)}, dimension)
+    # the currency question is "how exposed am I to this money", and cash is exposure too
+    assert split(holdings, {"RUB": D(100)}, DIM.currency) == [("RUB", D(1000), 1)]
+
+
+def test_an_unfilled_attribute_becomes_its_own_bucket_not_a_guess():
+    rows = split([holding(1, "100", sector=None, country=None)], {}, DIM.sector)
+    assert rows == [(UNKNOWN, D(100), 1)]
+    assert bucket_of(holding(1, "1", country=None), DIM.country) == UNKNOWN
+
+
+def test_every_dimension_covers_the_same_total():
+    holdings = [
+        holding(1, "600", asset_class="bond", sector="gov", country="RU", currency="RUB"),
+        holding(2, "400", asset_class="etf", sector=None, country="US", currency="USD"),
+    ]
+    cash = {"RUB": D(200)}
+    totals = {
+        dimension: sum(value for _, value, _ in split(holdings, cash, dimension))
+        for dimension in AllocationDimension
+    }
+    assert set(totals.values()) == {D(1200)}
+
+
+def test_weights_add_up_to_one():
+    weighted = weigh(
+        split([holding(1, "300"), holding(2, "100")], {"RUB": D(100)}, DIM.asset_class)
+    )
+    assert sum(w for _, _, _, w in weighted) == D(1)
+
+
+def test_a_short_keeps_its_value_but_does_not_inflate_the_longs():
+    weighted = weigh([("share", D(100), 1), ("bond", D(-50), 1)])
+    shares = next(row for row in weighted if row[0] == "share")
+    shorts = next(row for row in weighted if row[0] == "bond")
+    assert shares[3] == D(1)
+    assert shorts[1] == D(-50)
+
+
+def test_an_empty_portfolio_has_no_buckets():
+    assert split([], {}, DIM.asset_class) == []
+    assert weigh([]) == []