diff --git a/backend/tests/test_indicator_service.py b/backend/tests/test_indicator_service.py
new file mode 100644
index 0000000..f24e86f
--- /dev/null
+++ b/backend/tests/test_indicator_service.py
@@ -0,0 +1,277 @@
+"""Tests for app/services/indicator_service.py — the pure-math indicator
+library that every algorithm in the 13-vote signal system is built on top
+of. Runs against the pure-Python fallback path (this venv has no numpy
+installed), which is the code path actually exercised in a minimal
+deployment of this backend.
+"""
+from __future__ import annotations
+
+import math
+
+import pytest
+
+from app.services.indicator_service import (
+    atr,
+    bollinger_bands,
+    detect_market_regime,
+    ema,
+    macd,
+    mfi,
+    obv,
+    obv_signal,
+    rsi,
+    sma,
+    vwap,
+)
+
+
+def candle(high, low, close, volume=100.0):
+    return {"high": high, "low": low, "close": close, "volume": volume}
+
+
+class TestSma:
+    def test_leading_entries_are_none(self):
+        result = sma([1, 2, 3, 4, 5], period=3)
+        assert result[:2] == [None, None]
+
+    def test_values_match_hand_computed_average(self):
+        result = sma([1, 2, 3, 4, 5], period=3)
+        assert result == [None, None, 2.0, 3.0, 4.0]
+
+    def test_rejects_non_positive_period(self):
+        with pytest.raises(ValueError):
+            sma([1, 2, 3], period=0)
+
+
+class TestEma:
+    def test_seeded_with_sma_then_smoothed(self):
+        # period=3 -> multiplier=0.5; seed = mean([1,2,3])=2.0
+        result = ema([1, 2, 3, 4, 5], period=3)
+        assert result[2] == pytest.approx(2.0)
+        assert result[3] == pytest.approx((4 - 2.0) * 0.5 + 2.0)  # 3.0
+        assert result[4] == pytest.approx((5 - 3.0) * 0.5 + 3.0)  # 4.0
+
+
+class TestRsi:
+    def test_all_gains_yields_high_rsi_using_sentinel_rs(self):
+        prices = list(range(1, 17))  # strictly increasing, 15 deltas, all +1
+        result = rsi(prices, period=14)
+        assert all(v is None for v in result[:14])
+        # avg_loss stays 0 -> code uses RS=100 sentinel -> RSI = 100 - 100/101
+        expected = 100.0 - 100.0 / 101.0
+        assert result[14] == pytest.approx(expected)
+
+    def test_all_losses_yields_rsi_near_zero(self):
+        prices = list(range(16, 0, -1))  # strictly decreasing
+        result = rsi(prices, period=14)
+        assert result[14] == pytest.approx(0.0)
+
+    def test_flat_prices_do_not_yield_neutral_50(self):
+        """Documents a discovered quirk (not fixed here — flagged for the
+        team to decide on): when there's truly zero price movement, the
+        code sets an internal `rs = 50.0` sentinel intending "neutral", but
+        that value is still run through the RSI formula
+        (100 - 100/(1+rs)), which maps rs=50 to RSI≈98.04, not the
+        conventionally-expected neutral RSI of 50. A perfectly flat run
+        (e.g. an illiquid pair or stablecoin) would misreport as
+        near-overbought instead of neutral."""
+        prices = [10.0] * 16
+        result = rsi(prices, period=14)
+        assert result[14] == pytest.approx(100.0 - 100.0 / 51.0)
+
+    def test_insufficient_data_returns_all_none(self):
+        result = rsi([1, 2, 3], period=14)
+        assert result == [None, None, None]
+
+
+class TestBollingerBands:
+    def test_zero_variance_collapses_bands_to_middle(self):
+        bands = bollinger_bands([10.0] * 5, period=5, std_dev=2.0)
+        assert bands["middle"][-1] == pytest.approx(10.0)
+        assert bands["upper"][-1] == pytest.approx(10.0)
+        assert bands["lower"][-1] == pytest.approx(10.0)
+
+    def test_hand_computed_bands_for_linear_prices(self):
+        bands = bollinger_bands([1.0, 2.0, 3.0, 4.0, 5.0], period=5, std_dev=2.0)
+        sd = math.sqrt(2.0)  # population variance of [1..5] around mean 3 = 2.0
+        assert bands["middle"][-1] == pytest.approx(3.0)
+        assert bands["upper"][-1] == pytest.approx(3.0 + 2 * sd)
+        assert bands["lower"][-1] == pytest.approx(3.0 - 2 * sd)
+        assert bands["upper_1"][-1] == pytest.approx(3.0 + sd)
+        assert bands["lower_1"][-1] == pytest.approx(3.0 - sd)
+
+    def test_upper_always_above_lower_when_present(self):
+        bands = bollinger_bands([5, 3, 8, 1, 9, 2, 7, 4, 6, 10.0], period=5)
+        for u, l in zip(bands["upper"], bands["lower"]):
+            if u is not None and l is not None:
+                assert u >= l
+
+
+class TestMacd:
+    def test_rejects_fast_not_less_than_slow(self):
+        with pytest.raises(ValueError):
+            macd([1.0] * 30, fast=26, slow=12)
+
+    def test_returns_expected_keys_with_matching_length(self):
+        prices = [float(i % 7 + 10) for i in range(40)]
+        result = macd(prices, fast=5, slow=10, signal=3)
+        assert set(result.keys()) == {"macd_line", "signal_line", "histogram"}
+        assert len(result["macd_line"]) == len(prices)
+        assert len(result["signal_line"]) == len(prices)
+        assert len(result["histogram"]) == len(prices)
+
+    def test_macd_line_is_ema_fast_minus_ema_slow(self):
+        prices = [float(i) for i in range(1, 41)]
+        result = macd(prices, fast=5, slow=10, signal=3)
+        ema_fast = ema(prices, 5)
+        ema_slow = ema(prices, 10)
+        for i in range(9, len(prices)):
+            assert result["macd_line"][i] == pytest.approx(ema_fast[i] - ema_slow[i])
+
+
+class TestAtr:
+    def test_constant_true_range_gives_exact_atr(self):
+        # high=105, low=95 always -> TR=10 for every bar (including vs prev close=100)
+        candles = [candle(105, 95, 100) for _ in range(4)]
+        result = atr(candles, period=3)
+        assert result[:2] == [None, None]
+        assert result[2] == pytest.approx(10.0)
+        assert result[3] == pytest.approx(10.0)
+
+    def test_insufficient_candles_returns_all_none(self):
+        assert atr([candle(1, 1, 1)], period=14) == [None]
+
+
+class TestVwap:
+    def test_cumulative_volume_weighted_average(self):
+        candles = [candle(10, 8, 9, volume=100), candle(12, 10, 11, volume=50)]
+        result = vwap(candles)
+        assert result[0] == pytest.approx(9.0)  # typical=(10+8+9)/3=9, only candle so far
+        assert result[1] == pytest.approx((9.0 * 100 + 11.0 * 50) / 150)
+
+    def test_empty_input_returns_empty_list(self):
+        assert vwap([]) == []
+
+
+class TestObv:
+    def test_hand_computed_cumulative_volume(self):
+        candles = [
+            candle(0, 0, 10, volume=100),
+            candle(0, 0, 11, volume=200),  # up -> +200
+            candle(0, 0, 10, volume=150),  # down -> -150
+            candle(0, 0, 12, volume=300),  # up -> +300
+        ]
+        assert obv(candles) == [100.0, 300.0, 150.0, 450.0]
+
+    def test_unchanged_close_keeps_obv_flat(self):
+        candles = [candle(0, 0, 10, volume=100), candle(0, 0, 10, volume=50)]
+        assert obv(candles) == [100.0, 100.0]
+
+
+class TestObvSignal:
+    def test_insufficient_data_returns_all_none(self):
+        crossovers, sma_vals = obv_signal([1.0, 2.0], period=3)
+        assert crossovers == [None, None]
+        assert sma_vals == [None, None]
+
+    def test_detects_bullish_crossover(self):
+        # OBV sits at/below its SMA (all zero) then jumps above it
+        crossovers, obv_sma = obv_signal([0, 0, 0, 0, 10], period=3)
+        assert crossovers[4] is True
+
+    def test_detects_bearish_crossover(self):
+        crossovers, _ = obv_signal([10, 10, 10, 10, 0], period=3)
+        assert crossovers[4] is False
+
+
+class TestMfi:
+    def test_insufficient_candles_returns_all_none(self):
+        assert mfi([candle(1, 1, 1)], period=14) == [None]
+
+    def test_overbought_when_no_negative_flow(self):
+        # typical price strictly increasing -> every period contributes only
+        # positive flow -> neg_flow == 0 -> MFI defined as 100.0.
+        # Index 4 (not 3) is asserted because index 3 is the very first
+        # computed value and hits the negative-indexing quirk below.
+        candles = [candle(10 + i, 10 + i, 10 + i, volume=100) for i in range(5)]
+        result = mfi(candles, period=3)
+        assert result[4] == pytest.approx(100.0)
+
+    def test_first_computed_value_has_a_wraparound_indexing_quirk(self):
+        """Documents a discovered quirk (not fixed here — flagged for the
+        team to decide on): for the first computed MFI value in a series,
+        the loop compares `typical_prices[j-1]` with `j=0`, which in Python
+        wraps around to `typical_prices[-1]` (the LAST candle in the whole
+        series) instead of having no prior candle to compare against. This
+        spuriously injects one bogus flow-direction comparison. In practice
+        this only taints the single oldest computed value in a long series
+        (never the latest, which is what signal_service.py actually reads),
+        so real-world impact is negligible — but it is objectively wrong."""
+        candles = [candle(10 + i, 10 + i, 10 + i, volume=100) for i in range(5)]
+        result = mfi(candles, period=3)
+        # Without the quirk this would also be 100.0 (strictly increasing,
+        # no real negative flow) — the quirk drags it down to ~69.7.
+        assert result[3] == pytest.approx(69.69696969696969)
+
+
+class TestDetectMarketRegime:
+    def _adx(self, value):
+        return {"adx": [value]}
+
+    def test_squeeze_with_volume_spike_is_breakout(self):
+        bb = {
+            "upper": [110.0] * 20,
+            "lower": [90.0] * 20,
+        }
+        # Force squeeze: last width == min width
+        regime = detect_market_regime(
+            self._adx(15), bb, atr_pct=1.0, volume_data=[False, True],
+        )
+        assert regime == "breakout"
+
+    def test_squeeze_without_volume_spike_is_squeeze(self):
+        bb = {"upper": [110.0] * 20, "lower": [90.0] * 20}
+        regime = detect_market_regime(
+            self._adx(15), bb, atr_pct=1.0, volume_data=[False, False],
+        )
+        assert regime == "squeeze"
+
+    def test_high_atr_without_squeeze_is_volatile(self):
+        bb = {"upper": [], "lower": []}
+        regime = detect_market_regime(
+            self._adx(15), bb, atr_pct=10.0, volume_data=None,
+        )
+        assert regime == "volatile"
+
+    def test_high_adx_alone_is_not_enough_to_confirm_trending(self):
+        """High ADX alone yields 'neutral', not 'trending' — the function
+        also requires Efficiency-Ratio/Choppiness confirmation (which needs
+        `prices`/`highs`/`lows`), matching its documented "multi-factor"
+        design rather than being a pure ADX threshold classifier."""
+        bb = {"upper": [], "lower": []}
+        regime = detect_market_regime(
+            self._adx(30), bb, atr_pct=1.0, volume_data=None,
+        )
+        assert regime == "neutral"
+
+    def test_high_adx_with_confirming_efficiency_ratio_is_trending(self):
+        bb = {"upper": [], "lower": []}
+        prices = [float(i) for i in range(1, 30)]  # strictly trending -> ER ~1.0
+        regime = detect_market_regime(
+            self._adx(30), bb, atr_pct=1.0, volume_data=None, prices=prices,
+        )
+        assert regime == "trending"
+
+    def test_low_adx_is_sideways(self):
+        bb = {"upper": [], "lower": []}
+        regime = detect_market_regime(
+            self._adx(10), bb, atr_pct=1.0, volume_data=None,
+        )
+        assert regime == "sideways"
+
+    def test_mid_range_with_no_data_is_neutral(self):
+        bb = {"upper": [], "lower": []}
+        regime = detect_market_regime(
+            self._adx(22), bb, atr_pct=1.0, volume_data=None,
+        )
+        assert regime == "neutral"
diff --git a/backend/tests/test_signal_service_async.py b/backend/tests/test_signal_service_async.py
new file mode 100644
index 0000000..e08f23d
--- /dev/null
+++ b/backend/tests/test_signal_service_async.py
@@ -0,0 +1,233 @@
+"""Tests for the async trade-lifecycle functions in
+app/services/signal_service.py: `close_stale_trades` (the exit-rule engine —
+time limit, stop loss, take profit, trailing stop) and `expire_old_signals`.
+
+Both functions open their own DB session via a module-level
+`async_session_factory` rather than accepting one as a parameter, so these
+tests monkeypatch that name to an in-memory SQLite session factory instead
+of the real Postgres one.
+"""
+from __future__ import annotations
+
+import uuid
+from datetime import datetime, timedelta, timezone
+from decimal import Decimal
+
+import pytest
+from sqlalchemy import select
+
+from app.models import Exchange, HypotheticalTrade, Signal, Symbol, User
+from app.models.candle import Candle
+from app.services import signal_service
+
+
+def make_user(**prefs):
+    return User(
+        id=uuid.uuid4(), username=f"u-{uuid.uuid4().hex[:8]}",
+        email=f"{uuid.uuid4().hex[:8]}@example.com", password_hash="x",
+        role="trader", is_active=True, preferences=prefs,
+    )
+
+
+async def _seed_price(db, *, exchange_name: str, symbol: str, timeframe: str, price: Decimal):
+    exchange = Exchange(name=exchange_name, display_name=exchange_name.upper())
+    db.add(exchange)
+    await db.flush()
+    sym = Symbol(symbol=symbol, base=symbol.split("/")[0], quote=symbol.split("/")[1], exchange_id=exchange.id)
+    db.add(sym)
+    await db.flush()
+    db.add(Candle(
+        symbol_id=sym.id, timeframe=timeframe, timestamp=datetime.now(timezone.utc),
+        open=price, high=price, low=price, close=price, volume=Decimal("1000"),
+    ))
+    await db.flush()
+
+
+class TestCloseStaleTradesTimeLimit:
+    async def test_closes_trade_that_exceeds_max_hold_hours(self, session_factory, monkeypatch):
+        monkeypatch.setattr(signal_service, "async_session_factory", session_factory)
+
+        async with session_factory() as db:
+            user = make_user()
+            db.add(user)
+            await _seed_price(db, exchange_name="mexc", symbol="BTC/USDT", timeframe="1h", price=Decimal("100"))
+            trade = HypotheticalTrade(
+                user_id=user.id, symbol="BTC/USDT", exchange="mexc", timeframe="1h",
+                direction="LONG", entry_price=Decimal("100"),
+                entry_time=datetime.now(timezone.utc) - timedelta(hours=9),  # exceeds default 8h
+                quantity=Decimal("1"), status="OPEN",
+            )
+            db.add(trade)
+            await db.commit()
+            trade_id = trade.id
+
+        await signal_service.close_stale_trades()
+
+        async with session_factory() as db:
+            result = await db.execute(select(HypotheticalTrade).where(HypotheticalTrade.id == trade_id))
+            saved = result.scalar_one()
+            assert saved.status == "CLOSED"
+            assert saved.exit_reason == "TIME_LIMIT"
+
+    async def test_does_not_close_trade_within_hold_window_and_no_sl_tp_hit(self, session_factory, monkeypatch):
+        monkeypatch.setattr(signal_service, "async_session_factory", session_factory)
+
+        async with session_factory() as db:
+            user = make_user()
+            db.add(user)
+            # Small 2% move -- below default 5% SL and 10% TP
+            await _seed_price(db, exchange_name="mexc", symbol="BTC/USDT", timeframe="1h", price=Decimal("102"))
+            trade = HypotheticalTrade(
+                user_id=user.id, symbol="BTC/USDT", exchange="mexc", timeframe="1h",
+                direction="LONG", entry_price=Decimal("100"),
+                entry_time=datetime.now(timezone.utc) - timedelta(hours=1),
+                quantity=Decimal("1"), status="OPEN",
+            )
+            db.add(trade)
+            await db.commit()
+            trade_id = trade.id
+
+        await signal_service.close_stale_trades()
+
+        async with session_factory() as db:
+            result = await db.execute(select(HypotheticalTrade).where(HypotheticalTrade.id == trade_id))
+            saved = result.scalar_one()
+            assert saved.status == "OPEN"
+
+
+class TestCloseStaleTradesStopLossTakeProfit:
+    async def test_closes_long_on_fixed_percent_stop_loss(self, session_factory, monkeypatch):
+        monkeypatch.setattr(signal_service, "async_session_factory", session_factory)
+
+        async with session_factory() as db:
+            user = make_user()
+            db.add(user)
+            # 6% drop -- exceeds default fixed stop_loss_pct=5%
+            await _seed_price(db, exchange_name="mexc", symbol="BTC/USDT", timeframe="1h", price=Decimal("94"))
+            trade = HypotheticalTrade(
+                user_id=user.id, symbol="BTC/USDT", exchange="mexc", timeframe="1h",
+                direction="LONG", entry_price=Decimal("100"),
+                entry_time=datetime.now(timezone.utc) - timedelta(hours=1),
+                quantity=Decimal("1"), status="OPEN",
+            )
+            db.add(trade)
+            await db.commit()
+            trade_id = trade.id
+
+        await signal_service.close_stale_trades()
+
+        async with session_factory() as db:
+            result = await db.execute(select(HypotheticalTrade).where(HypotheticalTrade.id == trade_id))
+            saved = result.scalar_one()
+            assert saved.status == "CLOSED"
+            assert saved.exit_reason == "STOP_LOSS"
+            assert saved.pnl == Decimal("-6")
+
+    async def test_closes_long_on_fixed_percent_take_profit(self, session_factory, monkeypatch):
+        monkeypatch.setattr(signal_service, "async_session_factory", session_factory)
+
+        async with session_factory() as db:
+            user = make_user()
+            db.add(user)
+            # 12% rise -- exceeds default fixed take_profit_pct=10%
+            await _seed_price(db, exchange_name="mexc", symbol="BTC/USDT", timeframe="1h", price=Decimal("112"))
+            trade = HypotheticalTrade(
+                user_id=user.id, symbol="BTC/USDT", exchange="mexc", timeframe="1h",
+                direction="LONG", entry_price=Decimal("100"),
+                entry_time=datetime.now(timezone.utc) - timedelta(hours=1),
+                quantity=Decimal("1"), status="OPEN",
+            )
+            db.add(trade)
+            await db.commit()
+            trade_id = trade.id
+
+        await signal_service.close_stale_trades()
+
+        async with session_factory() as db:
+            result = await db.execute(select(HypotheticalTrade).where(HypotheticalTrade.id == trade_id))
+            saved = result.scalar_one()
+            assert saved.status == "CLOSED"
+            assert saved.exit_reason == "TARGET"
+
+
+class TestCloseStaleTradesTrailingStop:
+    async def test_closes_when_price_drops_through_trailing_stop(self, session_factory, monkeypatch):
+        monkeypatch.setattr(signal_service, "async_session_factory", session_factory)
+
+        async with session_factory() as db:
+            user = make_user(auto_trade_trailing_stops={
+                "BTC/USDT_mexc": {
+                    "direction": "LONG", "best_price": 110.0,
+                    "trailing_stop_price": 104.5, "trailing_pct": 5.0,
+                },
+            })
+            db.add(user)
+            # 4% move -- below the fixed 5% SL threshold, so SL doesn't
+            # preempt; but it's below the pre-set trailing_stop_price=104.5
+            await _seed_price(db, exchange_name="mexc", symbol="BTC/USDT", timeframe="1h", price=Decimal("104"))
+            trade = HypotheticalTrade(
+                user_id=user.id, symbol="BTC/USDT", exchange="mexc", timeframe="1h",
+                direction="LONG", entry_price=Decimal("100"),
+                entry_time=datetime.now(timezone.utc) - timedelta(hours=1),
+                quantity=Decimal("1"), status="OPEN",
+            )
+            db.add(trade)
+            await db.commit()
+            trade_id = trade.id
+
+        await signal_service.close_stale_trades()
+
+        async with session_factory() as db:
+            result = await db.execute(select(HypotheticalTrade).where(HypotheticalTrade.id == trade_id))
+            saved = result.scalar_one()
+            assert saved.status == "CLOSED"
+            assert saved.exit_reason == "TRAILING_STOP"
+
+
+class TestExpireOldSignals:
+    async def test_marks_old_active_signals_as_expired(self, session_factory, monkeypatch):
+        monkeypatch.setattr(signal_service, "async_session_factory", session_factory)
+
+        async with session_factory() as db:
+            old_signal = Signal(
+                symbol="BTC/USDT", exchange="mexc", timeframe="1h",
+                signal_type="BUY", strength="MODERATE", price=Decimal("100"),
+                timestamp=datetime.now(timezone.utc) - timedelta(days=10),
+                status="ACTIVE",
+                created_at=datetime.now(timezone.utc) - timedelta(days=10),
+            )
+            recent_signal = Signal(
+                symbol="BTC/USDT", exchange="mexc", timeframe="1h",
+                signal_type="BUY", strength="MODERATE", price=Decimal("100"),
+                timestamp=datetime.now(timezone.utc) - timedelta(hours=1),
+                status="ACTIVE",
+                created_at=datetime.now(timezone.utc) - timedelta(hours=1),
+            )
+            db.add_all([old_signal, recent_signal])
+            await db.commit()
+            old_id, recent_id = old_signal.id, recent_signal.id
+
+        expired_count = await signal_service.expire_old_signals(max_age_days=7)
+        assert expired_count == 1
+
+        async with session_factory() as db:
+            old_result = await db.execute(select(Signal).where(Signal.id == old_id))
+            recent_result = await db.execute(select(Signal).where(Signal.id == recent_id))
+            assert old_result.scalar_one().status == "EXPIRED"
+            assert recent_result.scalar_one().status == "ACTIVE"
+
+    async def test_already_expired_signals_are_not_recounted(self, session_factory, monkeypatch):
+        monkeypatch.setattr(signal_service, "async_session_factory", session_factory)
+
+        async with session_factory() as db:
+            db.add(Signal(
+                symbol="BTC/USDT", exchange="mexc", timeframe="1h",
+                signal_type="BUY", strength="MODERATE", price=Decimal("100"),
+                timestamp=datetime.now(timezone.utc) - timedelta(days=10),
+                status="EXPIRED",
+                created_at=datetime.now(timezone.utc) - timedelta(days=10),
+            ))
+            await db.commit()
+
+        expired_count = await signal_service.expire_old_signals(max_age_days=7)
+        assert expired_count == 0