Recommendations

Confluence-ranked feed of setups across your watchlist + portfolio. The “what should I look at right now” surface.

What it does

• Ranks symbols by hit_rate × log(1 + sample_size) so high-hit-rate / large-sample setups rise to the top.
• Each row exposes the layer-by-layer breakdown: which fired, what the historical hit rate is on this symbol, sample size, experimental flag.
• Drill into any row to open the full per-symbol detail with the detection’s chart timestamp pre-anchored.
• Dismiss / watch controls.

How to access

/trading-intel/recommendations — also reachable from the hub’s left rail.

Key controls

Sort

Default: ranking score (hit_rate weighted by sample). Alternative sorts: confluence score raw, recency, alphabetical.

Filters

• Layer — structure / PA / Smart Money / any
• Direction — bull / bear / both
• Timeframe — multiselect (1h, 4h, 1d default)
• Experimental — show / hide (default: show with warning)

Drill-in

Click any row → opens /trading-intel/analysis?symbol=...&t=... with the chart anchored on the detection.

Dismiss / watch

• Dismiss: hides the row for 24h.
• Watch: pins the row to your “Watching” subset for monitoring.

Common workflows

”Top 3 for the morning”

1. Default sort + default filters.
2. Top 3 rows.
3. Drill into each in sequence.

”Only bullish, only large-sample”

1. Direction filter → bull.
2. Sort by sample size descending.
3. Skip experimental rows.

”I keep dismissing AAPL”

If you dismiss the same symbol repeatedly, BlitzPulse remembers the pattern but doesn’t suppress beyond 24h. To silence permanently, remove the symbol from your watchlist or unsubscribe from recommendation alerts on it.

Tips & best practices

• Sample size matters. A pattern with 80% hit rate on 5 samples is luck; the same pattern on 200 samples is signal. Our ranking metric handles this — but read it yourself.
• Experimental ≠ wrong. It means the historical edge is weak or unconfirmed. If you have a fundamental thesis, the experimental detection is one data point, not the whole story.
• Dismiss is cheap. Use it; the model doesn’t punish you.

Plan tier matrix

Troubleshooting

• List looks identical day to day — your watchlist is small; recommendations dedup against same-symbol+type+direction in 24h.
• Sample size = 0 on most rows — backtest job hasn’t run since the symbol was added; new symbols populate stats overnight.
• A favorite pattern never appears — the pattern’s hit rate on your symbol is below 55%. The detection still fires (visible on /trading-intel/analysis); it just doesn’t earn a slot in recommendations.

See also: Smart Money, Alerts.