news: DeepSeek validation, vs_forecast fix, /status source health

Validation (llm_news_eval.py harness + 78-headline golden CSV + verdict doc): headline scoring 78/78 vs keyword baseline 24/78 — validated; event gold-direction calls misread close K/M numeric comparisons ~11% of the time (both claims cases in sample, JOLTS 8.79M vs 8.84M). Fix: event payload now carries code-computed "vs_forecast" (from EconomicEvent.surprise) and the prompt gains an explicit inverted- metrics rule + worked claims example — iteration 1 (field alone) fixed the arithmetic but exposed a masked inversion error. 18/18 on live re-verification (4 misreads fixed, 11 regression cases clean, 3 fresh both-side probes). /status: new "News & AI" section — per-source health for the calendar chain (which of FF/MyFXBook served), each RSS feed, and DeepSeek (daily count, last error kind), written passively by the existing fetch paths under _cache_lock. Failures WARN once on transition (repeats DEBUG, recovery INFO) so the rotating log carries outage stories without 10-min spam. test_all.py: +5 Module 19 tests (351 total); also carries the Module 22 fixed-RR-TP tests from the prior research session (shared file). Replay artifacts dir gitignored like commit_ladder. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017mbXs9HRawZhD1uZ3ze7zQ
2026-08-06 22:21:02 -04:00
parent daaed8e6c7
commit b831001cfb
8 changed files with 1315 additions and 9 deletions
@@ -4829,6 +4829,8 @@ def _():
 print("\n═══ Module 19: DeepSeek news enrichment ═══")

 import datetime as _dt_m19
+import logging
+import requests
 import llm_client as lc
 from llm_client import DeepSeekClient
 from news_engine import (EconomicEvent, FundamentalContext, NewsEngine,
@@ -4906,10 +4908,12 @@ class _StubLLM:
    def __init__(self, reply):
        self.reply = reply
        self.calls = 0
+        self.last_payload = None
    available = True

    def chat_json(self, system_prompt, user_payload):
        self.calls += 1
+        self.last_payload = user_payload
        return self.reply


@@ -5029,6 +5033,10 @@ def _():
        risk, desc = ne.get_event_risk_level("XAU/USD", ctx)
        assert risk == "caution"
        assert desc.endswith("— gold bearish: strong jobs lift yields"), desc
+        # payload carries the precomputed comparison (300K > 200K) so the
+        # model never re-compares the raw numbers (2026-08-06 validation)
+        assert lc._client.last_payload["vs_forecast"] == "above forecast", \
+            lc._client.last_payload
        # bogus direction → numeric fallback
        NewsEngine._llm_event_cache = {}
        lc._client = _StubLLM({"gold_direction": "sideways"})
@@ -5094,6 +5102,171 @@ def _():
    assert cfg.sentiment_affects_signals is False


+@test("llm_client: health stamps track last success and failure")
+def _():
+    client = DeepSeekClient(api_key="k", daily_cap=10)
+    assert client.last_ok == 0.0 and client.last_error is None
+    client._session = _counting_session(_chat_body({"a": 1}))
+    assert client.chat_json("sys", {}) == {"a": 1}
+    assert client.last_ok > 0 and client.last_error is None
+    client._session = _counting_session({"detail": "quota"}, status=429)
+    assert client.chat_json("sys", {}) is None
+    assert client.last_error is not None
+    ts, kind = client.last_error
+    assert kind == "http_429" and ts >= client.last_ok
+
+
+class _LogCapture(logging.Handler):
+    def __init__(self):
+        super().__init__()
+        self.records = []
+
+    def emit(self, record):
+        self.records.append(record)
+
+
+@test("NewsEngine: _record_source logs transitions only, tracks state")
+def _():
+    import news_engine as ne_mod
+    cap = _LogCapture()
+    ne_mod.logger.addHandler(cap)
+    old_level = ne_mod.logger.level
+    ne_mod.logger.setLevel(logging.INFO)
+    saved = NewsEngine._source_health
+    NewsEngine._source_health = {}
+    try:
+        NewsEngine._record_source("t:src", False, "HTTP 503")
+        warns = [r for r in cap.records if r.levelno == logging.WARNING]
+        assert len(warns) == 1 and "HTTP 503" in warns[0].getMessage()
+        assert NewsEngine.source_is_failing(
+            NewsEngine._source_health["t:src"])
+        # repeat failure must NOT warn again (DEBUG only)
+        NewsEngine._record_source("t:src", False, "HTTP 503")
+        warns = [r for r in cap.records if r.levelno == logging.WARNING]
+        assert len(warns) == 1
+        # recovery logs INFO and clears the failing state
+        NewsEngine._record_source("t:src", True, extra="42 kept")
+        infos = [r for r in cap.records if r.levelno == logging.INFO]
+        assert any("recovered" in r.getMessage() for r in infos)
+        entry = NewsEngine._source_health["t:src"]
+        assert not NewsEngine.source_is_failing(entry)
+        assert entry["extra"] == "42 kept"
+    finally:
+        ne_mod.logger.removeHandler(cap)
+        ne_mod.logger.setLevel(old_level)
+        NewsEngine._source_health = saved
+
+
+@test("NewsEngine: calendar fallback source is visible in health state")
+def _():
+    ne = NewsEngine(NewsConfig())
+    saved = NewsEngine._source_health
+    NewsEngine._source_health = {}
+    now = _dt_m19.datetime.now(_dt_m19.timezone.utc)
+    ev = EconomicEvent(time=now, currency="USD", impact="high", event="X")
+    try:
+        ne._try_fetch_calendar_api = lambda: []
+        ne._try_fetch_myfxbook_rss = lambda: [ev]
+        assert ne._fetch_calendar_raw() == [ev]
+        sh = NewsEngine._source_health
+        assert NewsEngine.source_is_failing(sh["calendar:forexfactory"])
+        assert not NewsEngine.source_is_failing(sh["calendar"])
+        assert "MyFXBook" in sh["calendar"]["extra"]
+        # both dead → calendar itself failing
+        ne._try_fetch_myfxbook_rss = lambda: []
+        assert ne._fetch_calendar_raw() == []
+        assert NewsEngine.source_is_failing(sh["calendar"])
+    finally:
+        NewsEngine._source_health = saved
+
+
+@test("NewsEngine: per-feed RSS health recorded, aggregate survives")
+def _():
+    cfg = NewsConfig()
+    cfg.rss_feeds = ["https://good.example/rss", "https://bad.example/rss"]
+    ne = NewsEngine(cfg)
+    saved = NewsEngine._source_health
+    NewsEngine._source_health = {}
+    rss = (b"<rss><channel><item>"
+           b"<title>Fed signals rate cuts as dollar slides</title>"
+           b"</item></channel></rss>")
+
+    class _Sess:
+        def get(self, url, timeout=None):
+            if "bad.example" in url:
+                raise requests.ConnectionError("refused")
+            return type("R", (), {"status_code": 200, "content": rss})()
+
+    ne._session = _Sess()
+    try:
+        heads = ne._fetch_news_sentiment_raw()
+        assert len(heads) == 1, heads
+        sh = NewsEngine._source_health
+        assert not NewsEngine.source_is_failing(sh["rss:good.example"])
+        assert NewsEngine.source_is_failing(sh["rss:bad.example"])
+        assert "refused" in sh["rss:bad.example"]["detail"]
+        agg = sh["headlines"]
+        assert not NewsEngine.source_is_failing(agg)
+        assert "1/2 feeds" in agg["extra"], agg
+    finally:
+        NewsEngine._source_health = saved
+
+
+@test("/status renders news/calendar/AI health with ⚠️ degradations")
+def _():
+    import telegram_bot as tb
+    saved = NewsEngine._source_health
+    t0 = time.time()
+    NewsEngine._source_health = {
+        "calendar": {"ok": t0, "err": None, "detail": "",
+                     "extra": "42 events via MyFXBook"},
+        "calendar:forexfactory": {"ok": None, "err": t0,
+                                  "detail": "HTTP 503", "extra": ""},
+        "headlines": {"ok": t0, "err": None, "detail": "",
+                      "extra": "38 headlines from 3/4 feeds"},
+        "rss:fxstreet.com": {"ok": None, "err": t0,
+                             "detail": "timeout", "extra": ""},
+    }
+    stub = _StubLLM(None)
+    stub.last_ok = 0.0
+    stub.last_error = (t0, "http_429")
+    stub._count = 87
+    stub._daily_cap = 500
+    stub._count_day = _dt_m19.datetime.now(_dt_m19.timezone.utc).date()
+    lc._client = stub
+    try:
+        text = tb._status_text()
+        assert "⚠️ Calendar: 42 events via MyFXBook — ForexFactory" in text
+        assert "HTTP 503" in text
+        assert "38 headlines from 3/4 feeds" in text
+        assert "⚠️ fxstreet.com: timeout" in text
+        assert "⚠️ DeepSeek: last call failed (http_429" in text
+        # healthy path: FF serving, DeepSeek clean → no warnings
+        NewsEngine._source_health = {
+            "calendar": {"ok": t0, "err": None, "detail": "",
+                         "extra": "42 events via ForexFactory"},
+            "headlines": {"ok": t0, "err": None, "detail": "",
+                          "extra": "38 headlines from 4/4 feeds"},
+        }
+        stub.last_ok = t0
+        text = tb._status_text()
+        assert "✅ Calendar: 42 events via ForexFactory" in text
+        assert "✅ DeepSeek: 87/500 calls today" in text
+        assert "⚠️" not in text.split("*News & AI:*")[1].split("*Alert")[0]
+        # empty state (pre-first-scan) renders the neutral line
+        NewsEngine._source_health = {}
+        assert "no news cycle yet" in tb._status_text()
+        # yesterday's exhausted counter must NOT show as cap-exhausted
+        # after the UTC date roll (counter resets lazily on next call)
+        stub._count = 500
+        stub._count_day = stub._count_day - _dt_m19.timedelta(days=1)
+        text = tb._status_text()
+        assert "cap" not in text and "0/500" in text, text
+    finally:
+        NewsEngine._source_health = saved
+        lc._client = None
+
+
 # ══════════════════════════════════════════════════════════════════════
 #  MODULE 20: dashboard_publisher.py (static web dashboard)
 # ══════════════════════════════════════════════════════════════════════
@@ -6456,6 +6629,92 @@ def _():
        os.environ.pop("FB2_EMA_LO", None)


+# ══════════════════════════════════════════════════════════════════════
+#  MODULE 22: fixed-RR TP research flag (--fixed-rr-tp)
+# ══════════════════════════════════════════════════════════════════════
+print("\n═══ Module 22: fixed-RR TP research flag ═══")
+
+
+@test("fixed-RR TP: tp1 = entry ± mult×risk (BUY+SELL), tp2=tp1, trail zeroed")
+def _():
+    import gpu_backtest as _gb
+    buy = _m21_sim_trade()                                       # 2400 / sl 2390
+    buy.trail_arm_r, buy.trail_dist_r = 0.8, 0.75                # PB-style stamp
+    sell = _m21_sim_trade(direction="SELL", entry=2400.0, sl=2410.0, tp1=2390.0)
+    _gb._apply_fixed_rr_tp([buy, sell], 2.0)
+    assert abs(buy.tp1 - 2420.0) < 1e-9, f"BUY tp1 must be 2420, got {buy.tp1}"
+    assert abs(sell.tp1 - 2380.0) < 1e-9, f"SELL tp1 must be 2380, got {sell.tp1}"
+    for t in (buy, sell):
+        assert t.tp2 == t.tp1, "tp2 must mirror the hard target"
+        assert t.trail_arm_r == 0.0 and t.trail_dist_r == 0.0, \
+            "per-trade trail params must be zeroed"
+        assert t.rr_planned == 2.0
+        assert abs(t.reward_pips - 200.0) < 1e-9   # 2 × $10 risk / 0.1 pip
+
+
+@test("fixed-RR TP: 2R target fills on the static path, spread charged")
+def _():
+    import gpu_backtest as _gb
+    t = _m21_sim_trade()
+    _gb._apply_fixed_rr_tp([t], 2.0)
+    sim = _m21_sim_df([(2400, 2405, 2395, 2404), (2404, 2421, 2402, 2418)])
+    _gb.simulate_trades([t], sim, max_bars=5, spread=0.30, trail_arm_r=0.0)
+    assert t.exit_reason == "TP1", f"got {t.exit_reason}"
+    assert abs(t.exit_price - 2420.0) < 1e-9
+    assert abs(t.r_multiple - (20.0 - 0.30) / 10.0) < 1e-9, \
+        "R must be (2R target - round-trip spread) / risk"
+
+
+@test("fixed-RR TP: same-bar SL+TP tie still resolves SL-first (rule 6)")
+def _():
+    import gpu_backtest as _gb
+    t = _m21_sim_trade()
+    _gb._apply_fixed_rr_tp([t], 2.0)
+    sim = _m21_sim_df([(2400, 2421, 2389, 2400)])   # both barriers inside bar 0
+    _gb.simulate_trades([t], sim, max_bars=5, spread=0.0, trail_arm_r=0.0)
+    assert t.exit_reason == "SL", f"tie must stay SL-first, got {t.exit_reason}"
+    assert abs(t.r_multiple - (-1.0)) < 1e-9
+
+
+@test("fixed-RR TP: per-trade PB trail params cannot resurrect the trail")
+def _():
+    import gpu_backtest as _gb
+    t = _m21_sim_trade()
+    t.trail_arm_r, t.trail_dist_r = 0.8, 0.75       # XAU_PB per-trade override
+    _gb._apply_fixed_rr_tp([t], 2.0)
+    # bar0 runs to +1.2R (a live 0.8R-arm trail would ratchet to 2404.5),
+    # bar1 retraces through that would-be trail stop, bar2 reaches 2R.
+    sim = _m21_sim_df([(2400, 2412, 2398, 2410),
+                       (2410, 2411, 2402, 2403),
+                       (2403, 2421, 2402, 2420)])
+    _gb.simulate_trades([t], sim, max_bars=5, spread=0.0, trail_arm_r=0.0)
+    assert t.exit_reason == "TP1", \
+        f"trail must be fully disabled under the flag, got {t.exit_reason}"
+    assert abs(t.exit_price - 2420.0) < 1e-9 and t.bars_held == 3
+    assert abs(t.r_multiple - 2.0) < 1e-9
+
+
+@test("fixed-RR TP: CLI parses the flag; trail-override combinations refused")
+def _():
+    import gpu_backtest as _gb
+    assert _gb._parse_args(
+        ["--pair", "XAUUSD", "--fixed-rr-tp", "2.0"]).fixed_rr_tp == 2.0
+    assert _gb._parse_args(["--pair", "XAUUSD"]).fixed_rr_tp is None
+    for bad in (["--trail-arm", "0.5"], ["--trail-dist", "0.8"],
+                ["--set", "signals.trail_arm_r=0.5"],
+                ["--set", "signals.pb_trail_dist_r=0.5"]):
+        try:
+            _gb.main(["--pair", "XAUUSD", "--fixed-rr-tp", "2.0"] + bad)
+            assert False, f"conflict {bad} must raise SystemExit"
+        except SystemExit as e:
+            assert "conflicts" in str(e)
+    try:
+        _gb.main(["--pair", "XAUUSD", "--fixed-rr-tp", "-1"])
+        assert False, "non-positive multiple must raise SystemExit"
+    except SystemExit as e:
+        assert "must be > 0" in str(e)
+
+
 # ══════════════════════════════════════════════════════════════════════
 #  SUMMARY
 # ══════════════════════════════════════════════════════════════════════