feat(ai): asset routing — categorize post → trade the right perp

Why
  Trading BTC on every actionable signal leaves alpha on the table.
  The same Trump post moves different assets very differently:
    • "Strategic Reserve including SOL" → BTC +8%, SOL +33%
    • "Tokenize Treasuries on ETH"      → ETH +5%, BTC +2%
    • "$TRUMP coin is GREATEST"         → TRUMP +50%+, BTC ~0%
  Picking the wrong asset is silent alpha leak.

What changed
  • SYSTEM_PROMPT: new ASSET ROUTING section with 6 named categories
    (direct_named / crypto_policy / macro_risk_on / macro_risk_off /
     defi_thematic / meme_named) + asset-pick rules per category.
  • Few-shot examples now show category + target_asset + expected_move.
  • Output JSON adds: category, target_asset, target_chain,
    expected_move_pct.
  • Python normalization:
    - HL_PERPS whitelist (current Hyperliquid perp universe).
    - CHAIN_FALLBACK map: meme not on HL → trade chain native
      (e.g. $FLOKI on Solana → SOL; ETH-chain meme not on HL → ETH).
    - Safe default → BTC if everything else fails.
    - expected_move_pct < 0.8 → coerce to hold (not worth fees).
    - Legacy `asset` field kept aligned to BTC/ETH for the existing
      price_impact tracker; alts leave it null until tracker is upgraded.

Note
  Bot routing (bot_engine.py) NOT yet updated — it still trades whatever
  `price_impact_asset` is, which is BTC/ETH only. The new fields are
  emitted and stored in DB but consumed downstream in a follow-up.
  This commit is "AI says the right thing" — making the bot ACT on it
  is the next step.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

app/services/analysis.py

@@ -116,6 +116,67 @@ CONFIDENCE CALIBRATION (read before assigning a number)
            below 80 produces zero downstream action. Do NOT pad to 70-79
            as a "soft signal" — that wastes your output. Just emit HOLD.
 
+═══════════════════════════════════════════════════════════════════════
+ASSET ROUTING — which coin to actually trade
+═══════════════════════════════════════════════════════════════════════
+Different posts move different assets DIFFERENTLY. Trading BTC on every
+signal leaves alpha on the table. Classify the post into a CATEGORY,
+then pick the asset that this category moves MOST.
+
+CATEGORIES (pick exactly one):
+
+  "direct_named"   — post explicitly names a specific coin/token
+                     (Bitcoin, Solana, XRP, $TRUMP, Dogecoin, Ethereum,
+                     SUI, AVAX, etc.). The named asset reacts most.
+                     Example move: post says "SOL in reserve" →
+                     BTC +8%, SOL +33%. Trade SOL.
+
+  "crypto_policy"  — generic crypto policy without a specific coin
+                     (SEC enforcement, executive order on crypto,
+                     "Strategic Reserve" without naming). BTC is the
+                     cleanest beta vehicle. Trade BTC.
+
+  "macro_risk_on"  — risk appetite rises (ceasefire, deal signed,
+                     rate cut signal, tariff withdrawal). All risk
+                     assets up; alts amplify BTC's move. Trade SOL
+                     for amplification, BTC if HL liquidity concern.
+
+  "macro_risk_off" — risk appetite falls (war escalation, tariff
+                     enacted, sanctions, banking shock). All risk
+                     assets down; alts amplify. Trade SOL short for
+                     amplification, BTC short for liquidity.
+
+  "defi_thematic"  — tokenization, DeFi, on-chain settlement, staking,
+                     yield products. ETH is the primary beneficiary.
+                     Trade ETH.
+
+  "meme_named"     — Trump explicitly references a meme coin ($TRUMP,
+                     $WIF, $PEPE, $DOGE, etc.). Trade THAT meme. If
+                     not on Hyperliquid, trade its CHAIN'S native
+                     token (Solana memes → SOL, ETH memes → ETH,
+                     BNB memes → BNB).
+
+TARGET ASSET (the actual perp to trade):
+  Pick from this Hyperliquid universe, in priority order based on
+  liquidity and the category above:
+    BTC, ETH, SOL, AVAX, ARB, OP, DOGE, SHIB, PEPE, WIF, TRUMP,
+    SUI, APT, INJ, ATOM, ADA, MATIC, FARTCOIN
+
+  Fallback rules if your preferred asset isn't on HL:
+    • Solana-based meme not on HL → SOL
+    • ETH-chain meme not on HL → ETH
+    • BNB-chain meme not on HL → BNB (if perp exists, else BTC)
+    • Unknown chain or ambiguous → BTC (always-safe default)
+
+EXPECTED MOVE (% in 1h, side-adjusted toward the signal direction):
+  Realistic ranges based on past Trump-event reactions:
+    BTC:    0.5 – 2%   for strong direct catalyst, max ~5%
+    ETH:    0.7 – 3%   slightly amplified
+    SOL:    1.5 – 5%   ~3× BTC's beta
+    Memes:  3 – 15%   wildly variable, asymmetric upside
+  If your honest expected move is < 0.8% on the chosen asset, the trade
+  isn't worth fees + slippage → DOWNGRADE TO HOLD.
+
 ═══════════════════════════════════════════════════════════════════════
 EXAMPLES — calibrate against these
 ═══════════════════════════════════════════════════════════════════════
@@ -124,21 +185,43 @@ EXAMPLES — calibrate against these
 
 POST: "Today I am announcing a Strategic Crypto Reserve including Bitcoin,
        Solana, XRP, Ethereum. Effective immediately by Executive Order."
-ANSWER: signal=buy, conf=95
+ANSWER: signal=buy, conf=95, category=direct_named, target_asset=SOL,
-WHY: (a) DIRECT_CRYPTO. Specific entities, immediate execution, concrete
+        expected_move=4
-     EO. All 7 checklist items pass strongly. This is the textbook hit.
+WHY: (a) DIRECT_CRYPTO. Smaller-cap named coin (SOL) outperforms BTC
+     dramatically on direct mention — historical precedent shows
+     +30% range. Trade SOL not BTC for amplification.
 
 POST: "Effective midnight Eastern time, 25% tariffs on all Chinese imports.
        Signed today. No exceptions."
-ANSWER: signal=short, conf=88
+ANSWER: signal=short, conf=88, category=macro_risk_off, target_asset=SOL,
-WHY: (b) USD_RATES. Concrete order, immediate execution, named country and
+        expected_move=3
-     percentage. Strong USD bid + risk-off → BTC down.
+WHY: (b) USD_RATES. Risk-off. SOL has higher beta than BTC, so short
+     SOL captures 1.5-2× the move with same conviction.
 
 POST: "Just spoke with Israel and Iran. Ceasefire signed and effective in
        6 hours. All hostilities to cease."
-ANSWER: signal=buy, conf=85
+ANSWER: signal=buy, conf=85, category=macro_risk_on, target_asset=SOL,
-WHY: (c) GEOPOLITICAL. Named parties, signed (not "discussing"), hard
+        expected_move=2.5
-     deadline. Risk-on unwind → BTC up.
+WHY: (c) GEOPOLITICAL. Risk-on unwind. SOL amplifies BTC's beta.
+
+POST: "Just signed Executive Order: Tokenization of US Treasury bills
+       on Ethereum-based protocols. Effective Q1."
+ANSWER: signal=buy, conf=82, category=defi_thematic, target_asset=ETH,
+        expected_move=3
+WHY: (d) REGULATORY. ETH primary beneficiary of on-chain tokenization
+     narrative. BTC reaction muted compared to ETH on this theme.
+
+POST: "$TRUMP coin is the GREATEST. Buying NOW. America First!"
+ANSWER: signal=buy, conf=82, category=meme_named, target_asset=TRUMP,
+        expected_move=15
+WHY: meme_named. Direct meme reference. TRUMP perp exists on HL.
+     Asymmetric upside, expect 10-30% range historically.
+
+POST: "Big things coming for $WIF holders. Solana season is back!"
+ANSWER: signal=buy, conf=80, category=meme_named, target_asset=SOL,
+        expected_move=4
+WHY: meme_named WIF, but if WIF perp not on HL, route to chain native
+     SOL. Captures the Solana-meme rotation effect.
 
 [NOT A CATALYST — must be HOLD]
 
@@ -190,13 +273,24 @@ OUTPUT FORMAT (strict JSON, no markdown)
     "is_rt":        <true|false>,
     "surprise":     "<'headline' | 'noise'>"
   },
-  "relevant":   <true|false>,
+  "relevant":     <true|false>,
-  "asset":      "BTC" | "ETH" | null,
+  "asset":        "BTC" | "ETH" | null,
-  "sentiment":  "bullish" | "bearish" | "neutral",
+                  ← legacy field for price_impact tracking; set to BTC if
-  "signal":     "buy" | "short" | "hold",
+                    target_asset is not BTC/ETH. Used only for backtest tables.
-  "confidence": <0-100>,
+  "category":     "direct_named" | "crypto_policy" | "macro_risk_on" |
-  "reasoning":  "<EITHER the ≤15-word transmission chain (if buy/short),
+                  "macro_risk_off" | "defi_thematic" | "meme_named" | "none",
-                 OR which checklist item failed (if hold). One sentence.>"
+  "target_asset": "<HL ticker for the actual trade — e.g. BTC, SOL, TRUMP>"
+                  | null   ← null if signal is hold
+  "target_chain": "<chain native token if target_asset is a meme — e.g.
+                   SOL for $WIF; ETH for $PEPE>" | null
+  "expected_move_pct": <0-50, your honest 1h move estimate on target_asset
+                        in the signal direction>,
+  "sentiment":    "bullish" | "bearish" | "neutral",
+  "signal":       "buy" | "short" | "hold",
+  "confidence":   <0-100>,
+  "reasoning":    "<EITHER the ≤15-word transmission chain + asset
+                   choice rationale (if buy/short), OR which checklist
+                   item failed (if hold). One sentence.>"
 }
 
 HARD CONSTRAINTS on the JSON:
@@ -205,6 +299,8 @@ HARD CONSTRAINTS on the JSON:
   • If transmission == "none" OR domestic == true OR is_rt == true OR
     execution == "future" OR execution == "none", then signal MUST be
     "hold" and confidence MUST be 0.
+  • If signal != "hold", target_asset MUST be a real ticker.
+  • If expected_move_pct < 0.8, signal MUST be "hold" (not worth fees).
   • Never emit "sell" — that signal type does not exist anymore.
 """
 
@@ -228,6 +324,10 @@ decide. Default to HOLD. Only buy/short if every gate passes."""
 _FALLBACK = {
     "relevant": False,
     "asset": None,
+    "category": "none",
+    "target_asset": None,
+    "target_chain": None,
+    "expected_move_pct": 0.0,
     "sentiment": "neutral",
     "signal": "hold",
     "confidence": 0,
@@ -236,6 +336,35 @@ _FALLBACK = {
     "analysis_version": ANALYSIS_VERSION,
 }
 
+# Hyperliquid perp universe (as of v5).  Used to validate target_asset and
+# fall back to BTC when the model picks something we can't actually trade.
+# Update this list when HL adds new perps.
+HL_PERPS = {
+    "BTC", "ETH", "SOL", "AVAX", "ARB", "OP", "DOGE", "SHIB", "PEPE",
+    "WIF", "TRUMP", "SUI", "APT", "INJ", "ATOM", "ADA", "MATIC",
+    "FARTCOIN", "BNB", "LINK", "LTC", "XRP",
+}
+
+# Chain → its native token's HL ticker. Used when a meme isn't on HL —
+# we route to the chain's native instead.
+CHAIN_FALLBACK = {
+    "SOL":     "SOL",
+    "SOLANA":  "SOL",
+    "ETH":     "ETH",
+    "ETHEREUM":"ETH",
+    "BNB":     "BNB",
+    "BSC":     "BNB",
+    "AVAX":    "AVAX",
+    "BASE":    "ETH",      # Base is ETH L2
+    "ARB":     "ARB",
+    "OP":      "OP",
+}
+
+VALID_CATEGORIES = {
+    "direct_named", "crypto_policy", "macro_risk_on", "macro_risk_off",
+    "defi_thematic", "meme_named", "none",
+}
+
 
 def _fallback(prefilter: Optional[str] = None, reasoning: str = "") -> dict:
     out = dict(_FALLBACK)
@@ -369,17 +498,74 @@ async def analyze_post(text: str) -> dict:
 
         relevant = bool(result.get("relevant", False)) or signal in ("buy", "short")
 
-        asset = result.get("asset")
+        # ── Asset routing — pick the actual perp to trade ─────────────
-        if asset == "BOTH":
+        # `asset` (BTC/ETH legacy) is kept for backward-compat with the
-            asset = "BTC"
+        # price_impact tracker. `target_asset` is the new field that the
-        if asset not in ("BTC", "ETH", None):
+        # bot router should consume.
-            asset = None
+        asset_raw = (result.get("asset") or "").upper().strip() or None
+        if asset_raw == "BOTH":
+            asset_raw = "BTC"
+        if asset_raw not in ("BTC", "ETH", None):
+            asset_raw = None
+
+        category = (result.get("category") or "none").strip().lower()
+        if category not in VALID_CATEGORIES:
+            category = "none"
+
+        target_raw = (result.get("target_asset") or "").upper().strip() or None
+        target_chain = (result.get("target_chain") or "").upper().strip() or None
+
+        # Resolve target_asset to a real HL perp.
+        # 1. If model picked a real HL ticker → use it.
+        # 2. If not on HL but chain is given → fall back to chain native.
+        # 3. Otherwise → safe default BTC (only if we have a signal at all).
+        target_asset: Optional[str] = None
+        if signal in ("buy", "short"):
+            if target_raw and target_raw in HL_PERPS:
+                target_asset = target_raw
+            elif target_chain and CHAIN_FALLBACK.get(target_chain) in HL_PERPS:
+                target_asset = CHAIN_FALLBACK[target_chain]
+                logger.info("Routed %s (not on HL) → chain fallback %s",
+                            target_raw, target_asset)
+            else:
+                target_asset = "BTC"
+                logger.info("Routed %s/%s → safe default BTC",
+                            target_raw, target_chain)
+
+        # expected_move_pct gate: if model itself says <0.8%, kill the trade
+        try:
+            expected_move = float(result.get("expected_move_pct") or 0)
+        except (TypeError, ValueError):
+            expected_move = 0.0
+        expected_move = max(0.0, min(50.0, expected_move))
+        if signal in ("buy", "short") and expected_move < 0.8:
+            logger.info("Coerced to hold: expected_move=%.2f < 0.8%%", expected_move)
+            signal = "hold"
+            confidence = 0
+            target_asset = None
+
+        # Keep legacy `asset` field aligned with target_asset when possible
+        # so existing price_impact_monitor (which only tracks BTC/ETH) still
+        # sees something meaningful for those two assets. For SOL/MEME, set
+        # asset=None — the new router will decide what to do.
+        if target_asset in ("BTC", "ETH"):
+            asset_raw = target_asset
+        elif target_asset is None:
+            asset_raw = None  # hold case
+        else:
+            # Trade routed to alt — leave legacy asset null so the existing
+            # BTC/ETH-only price_impact tracker doesn't get confused.
+            asset_raw = None
 
         reasoning = str(result.get("reasoning", ""))[:1200]
 
         return {
             "relevant": relevant,
-            "asset": asset,
+            "asset": asset_raw,
+            "category": category,
+            "target_asset": target_asset,
+            "target_chain": target_chain,
+            "expected_move_pct": round(expected_move, 2),
             "sentiment": sentiment,
             "signal": signal,
             "confidence": confidence,