Fix ledger context being overridden by prescriptive initial prompt

The 13:20:03 run showed the ledger context was counterproductive: the
initial prompt's "Start with a multi-timeframe trend-following approach"
instruction caused the model to ignore the prior summary and repeat
EMA50-based strategies that produced 0 trades across all 15 iterations.

Two fixes:
- When prior_summary is present, replace the prescriptive starting
  instruction with one that explicitly defers to the ledger: refine the
  best prior strategy or try a different approach if all prior results
  were poor. Prevents the fixed instruction from overriding the context.
- Cap ledger entries per unique strategy at 3. A strategy repeated across
  11 iterations would contribute 33 entries, drowning out other approaches
  in the prior summary. 3 entries (one per instrument) is sufficient.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/agent.rs

@@ -674,10 +674,15 @@ async fn load_prior_summary(ledger: &Path, swym: &SwymClient) -> Option<String>
     // We use the full strategy JSON as the grouping key.
     let mut strategy_groups: std::collections::HashMap<String, Vec<(&LedgerEntry, Option<&RunMetricsSummary>)>> =
         std::collections::HashMap::new();
+    // Cap at 3 entries per unique strategy (one per instrument is enough).
+    // Without this, a strategy repeated across many iterations swamps the summary.
     for entry in &entries {
         let key = serde_json::to_string(&entry.strategy).unwrap_or_default();
-        let m = metrics_map.get(&entry.run_id).copied();
-        strategy_groups.entry(key).or_default().push((entry, m));
+        let group = strategy_groups.entry(key).or_default();
+        if group.len() < 3 {
+            let m = metrics_map.get(&entry.run_id).copied();
+            group.push((entry, m));
+        }
     }
 
     // Compute avg sharpe per strategy group

src/prompts.rs

@@ -499,14 +499,22 @@ pub fn initial_prompt(instruments: &[String], candle_intervals: &[String], prior
         Some(s) => format!("{s}\n\n"),
         None => String::new(),
     };
+    let starting_instruction = if prior_summary.is_some() {
+        "Based on the prior results above: if the best strategy has promising metrics, \
+refine it. If all prior results were poor (0 trades or deeply negative Sharpe), try a \
+clearly different indicator family or candle interval than what was already attempted. \
+Do NOT repeat approaches that consistently produced 0 trades."
+    } else {
+        "Start with a multi-timeframe trend-following approach with proper risk management \
+(stop-loss, time exit, and ATR-based position sizing)."
+    };
     format!(
         r#"{prior_section}Design a trading strategy for crypto spot markets.
 
 Available instruments: {}
 Available candle intervals: {}
 
-Start with a multi-timeframe trend-following approach with proper risk management
-(stop-loss, time exit, and ATR-based position sizing). Use "usdc" as the quote asset.
+{starting_instruction} Use "usdc" as the quote asset.
 
 Respond with ONLY the strategy JSON."#,
         instruments.join(", "),