feat(manifold): decouple Manifold from edge model (observational-only)

Per-category coverage audit showed coverage_rate=0.0 across every category in the
bot's current universe, so any edge Manifold produced was false edge. Retire it as
an ACTIVE trading signal while keeping the full audit/coverage/cooldown trail for a
future reactivation decision.

Two module-level flags in bayesian.py (read from env):
- MANIFOLD_SIGNAL_ENABLED (default False): when False, Manifold never touches the
  edge model — manifold_log_adj stays 0.0 (no posterior shift), no confidence bump,
  feat_mfld_lo=0.0 (so it can never be the dominant feature), no trade contribution,
  and mfld_audit_id is not propagated so the audit's used_in_trade stays FALSE.
- MANIFOLD_AUDIT_ENABLED (default True): matcher still runs; audit/coverage rows and
  cooldowns are still written. The matcher is only called when a flag is on.

When signal is disabled, logs and reasoning carry "Manifold: observational_only".
Endpoints /api/metrics/manifold-matches and /api/metrics/manifold-coverage,
cooldowns, audit tables and existing trades are unchanged. No code or tables removed.

Other signals, thresholds, exposure, risk manager and the executor are untouched.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

bot/strategy/bayesian.py

@@ -12,6 +12,7 @@ Polymarket might reflect in a slow-moving order book.
 """
 import logging
 import math
+import os
 import uuid
 from dataclasses import dataclass, field
 from datetime import datetime, timedelta, timezone
@@ -61,6 +62,27 @@ NEWS_LOGODDS_WEIGHT = 1.5
 # Weaker than NEWS_LOGODDS_WEIGHT because Manifold can have illiquid/stale markets.
 MANIFOLD_LOGODDS_WEIGHT = 0.6
 
+
+def _env_bool(name: str, default: bool) -> bool:
+    return os.getenv(name, str(default)).strip().lower() in ("1", "true", "yes", "on")
+
+
+# ── Manifold activation flags ──────────────────────────────────────────────────
+# Manifold has been retired as an ACTIVE trading signal: a per-category coverage
+# audit (see /api/metrics/manifold-coverage) showed coverage_rate=0.0 across every
+# category in the bot's current universe, so any edge it produced was false edge.
+#
+# MANIFOLD_SIGNAL_ENABLED (default False): when False, Manifold is observational
+#   only — its probability never touches the edge model: no manifold_log_adj, no
+#   confidence bump, feat_mfld_lo stays 0.0 (so it can never be the dominant
+#   feature), and it never contributes to a trade.
+# MANIFOLD_AUDIT_ENABLED (default True): when True the matcher still runs and
+#   audit/coverage rows + cooldowns are written, preserving the trail so we can
+#   decide later whether to reactivate Manifold in a universe with real coverage.
+# The matcher is only called when at least one flag is on.
+MANIFOLD_SIGNAL_ENABLED = _env_bool("MANIFOLD_SIGNAL_ENABLED", False)
+MANIFOLD_AUDIT_ENABLED  = _env_bool("MANIFOLD_AUDIT_ENABLED", True)
+
 # GNews free tier: 100 req/day.  We limit to 5 queries per trading cycle
 # (politics markets only) and rely on 6 h cache to stay within budget.
 MAX_NEWS_QUERIES_PER_CYCLE = 5
@@ -471,7 +493,8 @@ class BayesianStrategy:
         manifold_result: Optional[ManifoldMatchResult] = None
         audit_id: Optional[str] = None
 
-        if (is_politics or is_tech) and self._manifold is not None:
+        if ((is_politics or is_tech) and self._manifold is not None
+                and (MANIFOLD_AUDIT_ENABLED or MANIFOLD_SIGNAL_ENABLED)):
             # ── Cooldown gate ────────────────────────────────────────────────
             # Skip markets whose Manifold verdict was recently settled to a
             # stable value.  A skip is equivalent to a no-signal: the matcher is
@@ -497,8 +520,10 @@ class BayesianStrategy:
             if not in_cooldown:
                 manifold_result = await self._manifold.get_match(market.question)
 
-                # Persist audit record for ALL outcomes (accepted / rejected / no_results)
-                if self._db is not None:
+                # Persist audit record for ALL outcomes (accepted / rejected / no_results).
+                # Gated by MANIFOLD_AUDIT_ENABLED so the audit/coverage trail and
+                # cooldowns can be kept even while Manifold is observational-only.
+                if MANIFOLD_AUDIT_ENABLED and self._db is not None:
                     if not market.id:
                         log.error(
                             "MANIFOLD_AUDIT: market.id is None/empty — skipping audit save | "
@@ -564,7 +589,10 @@ class BayesianStrategy:
                     "reason":        manifold_result.match_reason,
                 })
 
-                if manifold_result.status == "accepted" and manifold_result.prob_final is not None:
+                if (MANIFOLD_SIGNAL_ENABLED
+                        and manifold_result.status == "accepted"
+                        and manifold_result.prob_final is not None):
+                    # ACTIVE signal path — only when explicitly enabled.
                     manifold_used = True
                     self._manifold_fetched += 1
                     m_clamped = max(0.05, min(0.95, manifold_result.prob_final))
@@ -572,6 +600,16 @@ class BayesianStrategy:
                     p_log = math.log(prior / (1 - prior))
                     manifold_log_adj = (m_log - p_log) * MANIFOLD_LOGODDS_WEIGHT
                     sources.append(f"Manifold:{manifold_result.prob_final:.2f}")
+                elif not MANIFOLD_SIGNAL_ENABLED:
+                    # Observational-only: matched/audited but NEVER fed to the edge
+                    # model.  manifold_log_adj stays 0.0 → no confidence bump,
+                    # feat_mfld_lo=0.0 (cannot be dominant), no trade contribution.
+                    log.info(
+                        "Manifold: observational_only — signal disabled "
+                        "(MANIFOLD_SIGNAL_ENABLED=false) | market=%s status=%s",
+                        market.id, manifold_result.status,
+                    )
+                    sources.append("Manifold: observational_only")
 
         # Confidence cap: macro/politics/tech signals are weaker proxies
         confidence_cap = 0.65 if (is_macro or is_politics or is_tech or is_events) else 0.90
@@ -701,8 +739,12 @@ class BayesianStrategy:
             feat_news_lo=feat_news_lo,
             feat_mfld_lo=feat_mfld_lo,
             feat_btc_dom_lo=feat_btc_dom_lo,
-            # Manifold match audit — propagated through Order → Trade → DB
-            mfld_audit_id=audit_id,
+            # Manifold match audit — propagated through Order → Trade → DB.
+            # mfld_audit_id is the hook main.py uses to flip the audit row's
+            # used_in_trade=TRUE; suppress it when observational so the trail
+            # truthfully shows Manifold drove no trades. The mfld_* fields below
+            # stay as observational record (feat_mfld_lo is already 0.0).
+            mfld_audit_id=(audit_id if MANIFOLD_SIGNAL_ENABLED else None),
             mfld_market_id=manifold_result.market_id if manifold_result else None,
             mfld_market_title=manifold_result.market_title if manifold_result else None,
             mfld_market_url=manifold_result.market_url if manifold_result else None,