Vanguard Signal Framework
Confluence • Regime Awareness • Conflict Suppression
Three weighted scores built from Vanguard Agent — a Python bot that logs hourly SUI market data. All weights are data-driven from a 1,679-entry backtest (March–June 2026). Forward returns measured at +8h, +24h, and +72h. Every condition, every number is documented below. No black box.
Baseline context
The unconditional average across all 1,679 hourly entries. This is what you'd get by being in the market at a random hour. Every signal is evaluated against this floor.
| Condition | n | +8h avg | +24h avg | +72h avg | Win% 24h | Win% 72h |
|---|---|---|---|---|---|---|
| All candles (no filter) | 1,679 | -0.01% | -0.09% | +0.11% | 45% | 44% |
Bullish signals
Value & StructureConditions with positive forward-return edge vs. the baseline. Sorted by 24h average return.
Price below both the weekly and daily volume-weighted average — structurally oversold relative to where volume has transacted. The most reliable bullish setup in the dataset. Strong at every horizon, improves further out.
| Signal | n | +8h avg | +24h avg | +72h avg | Win% 24h | Win% 72h |
|---|---|---|---|---|---|---|
| Below VWAP Weekly + Daily | 144 | +0.76% | +2.54% | +4.38% | 76% | 80% |
| Below VWAP Weekly (alone) | 205 | +0.52% | +1.96% | +3.60% | 70% | 74% |
Fires when the 1H Bollinger Bands were in a squeeze (compressed) last hour and have now expanded, with price above VWAP Daily. Highest per-occurrence edge in the dataset. The 3 losses (out of 11) all had RSI 14H ≥ 70 or price already extended above VWAP Weekly at fire time — overextension kills the signal.
| Signal | n | +8h avg | +24h avg | +72h avg | Win% 24h | Win% 72h |
|---|---|---|---|---|---|---|
| BB Squeeze Break (all) | 11 | +1.12% | +4.62% | +5.18% | 73% | 64% |
Hourly RSI in isolation has weakened as the dataset grew through a prolonged downtrend — oversold can keep getting more oversold. Adding daily RSI context (<50) significantly improves the signal. The RSI 14H + 14D combo is the version worth tracking.
| Signal | n | +8h avg | +24h avg | +72h avg | Win% 24h | Win% 72h |
|---|---|---|---|---|---|---|
| RSI 14H <30 + RSI 14D <50 | 69 | +0.28% | +1.55% | +1.78% | 67% | 55% |
| RSI 14H 30–40 | 269 | -0.02% | +0.40% | +0.75% | 51% | 46% |
| RSI 14H 20–30 | 100 | -0.23% | +0.22% | +0.12% | 53% | 49% |
| RSI 14D 40–50 | 797 | +0.08% | +0.14% | +0.31% | 53% | 53% |
The 4H timeframe squeeze is a different signal than the 1H squeeze — the 1H squeeze alone is slightly bearish in this dataset (-0.68% at 24h, 42% win). The 4H squeeze, especially with price already in the upper half of the band (position 0.50–0.75), has consistent edge that grows at 72h. Combining both timeframes squeezed simultaneously improves the 72h outcome further.
| Signal | n | +8h avg | +24h avg | +72h avg | Win% 24h | Win% 72h |
|---|---|---|---|---|---|---|
| 4H squeeze + pos 0.50–0.75 | 58 | +0.47% | +1.01% | +3.85% | 71% | 71% |
| Both 1H + 4H squeeze, pos 0.50–0.75 | 53 | +0.44% | +0.76% | +3.57% | 68% | 69% |
| 4H squeeze (any position) | 147 | +0.07% | +0.36% | +1.76% | 51% | 55% |
| Both 1H + 4H squeeze | 128 | +0.05% | +0.29% | +1.92% | 53% | 58% |
| 1H squeeze alone (no 4H) | 439 | -0.35% | -0.68% | -1.43% | 42% | 38% |
Low-volatility BTC and quiet SUI volume both correlate with positive forward returns, especially at 72h. The BTC calm threshold tightened from <50% to <30% realized vol — the 30–50% bucket is essentially neutral. Quiet SUI volume (<20th percentile rank) is a slow accumulation signal, weakest at 8h, strongest at 72h.
| Signal | n | +8h avg | +24h avg | +72h avg | Win% 24h | Win% 72h |
|---|---|---|---|---|---|---|
| BTC realized vol <30% | 710 | +0.14% | +0.65% | +2.25% | 44% | 52% |
| SUI vol rank <20 (quiet) | 593 | +0.28% | +0.39% | +1.62% | 53% | 57% |
| BTC realized vol <50% (old threshold) | 1,552 | -0.01% | -0.06% | +0.07% | 44% | 44% |
Bearish / distribution signals
DistributionConditions with negative forward-return edge. Three of these were previously classified as bullish based on an earlier, shorter backtest window — the full dataset reversed the sign on all three.
An earlier ~400-entry backtest showed these three as bullish. The full 1,679-entry dataset reverses all three. In combination they're the strongest bearish signal in the log — when both funding and underperformance fire together, SUI is lower 75% of the time at 24h and 85% of the time at 72h.
| Signal | n | +8h avg | +24h avg | +72h avg | Win% 24h | Win% 72h |
|---|---|---|---|---|---|---|
| Funding elevated + SUI underperforming | 114 | -1.19% | -2.65% | -5.22% | 25% | 15% |
| RVOL spike + SUI underperforming | 173 | -0.87% | -2.19% | -5.04% | 30% | 16% |
| Funding rate elevated (alone) | 286 | -0.92% | -2.11% | -4.64% | 32% | 18% |
| SUI underperforming BTC + SOL | 215 | -0.73% | -1.98% | -4.66% | 28% | 18% |
| Volume spike (rank >60th pct) | 533 | -0.40% | -0.97% | -3.19% | 29% | 27% |
RSI 80+ is a strong distribution signal at 24h. RSI 70–80 is noisier — the average is positive but the median is -1.60% and win rate is only 35%, suggesting a few large outliers (the May 10 rally) are skewing the mean. Price above VWAP Weekly is neutral to slightly negative; it only becomes meaningful as a distribution signal in combination with elevated funding or RSI.
| Signal | n | +8h avg | +24h avg | +72h avg | Win% 24h | Win% 72h |
|---|---|---|---|---|---|---|
| RSI 14H 80–100 | 13 | -2.13% | -3.22% | -0.06% | 8% | 38% |
| RSI 14H 70–80 | 51 | +1.24% | +0.17% | +4.88% | 35% | 65% |
| RSI 14D 50–60 | 874 | -0.06% | -0.25% | +0.04% | 38% | 37% |
| Price above VWAP Weekly | 764 | +0.09% | -0.02% | +1.00% | 41% | 44% |
The three scores
Vanguard Agent computes three weighted scores each hour. Conditions are assigned to the score that reflects their market phase. Weights are proportional to backtested edge — stronger conditions get heavier weights.
Measures structural compression — how far price has pulled back relative to where volume transacted. Context signal, not an entry trigger on its own.
Price below VWAP Weekly
205x · +24h avg +1.96% · +72h avg +3.60% · win 24h 70% · win 72h 74%
Price below VWAP Daily
Intraday compression. Adds to VWAP Weekly signal — the combo (n=144) hits +2.54% at 24h, 76% win.
RSI 14H <30 + RSI 14D <50
69x · +24h avg +1.55% · +72h avg +1.78% · win 24h 67%. Hourly oversold with daily RSI not yet in recovery — the paired version. RSI 14H <30 alone is much weaker; daily context is what makes it predictive.
BB position <0.25 (lower band)
193x · +24h avg +0.27% · win 24h 54%. Price near the lower Bollinger Band — additional compression confirmation.
Momentum and regime conditions. Entry signal fires when Perp Long ≥ 3.0 with ≥ 2 conditions contributing — multi-condition confluence required.
BB Squeeze Break above VWAP PROVISIONAL
11x · +24h avg +4.62% · +72h avg +5.18% · win 24h 73%. Highest per-occurrence edge. Provisional pending more data — n=11 is too small for full confidence. Loses edge when RSI ≥ 70 or price already above VWAP Weekly.
4H BB Squeeze (pos 0.50–0.75)
58x · +24h avg +1.01% · +72h avg +3.85% · win 24h 71% · win 72h 71%. 4H squeeze with price in the upper half of the band — compression with momentum direction already established. The slow-burn version of the squeeze signal.
BTC realized vol <30%
710x · +24h avg +0.65% · +72h avg +2.25% · win 72h 52%. Clean macro environment. Threshold tightened from <50% — the 30–50% bucket is neutral (+0.07% at 24h).
SUI vol rank <20 (quiet volume)
593x · +24h avg +0.39% · +72h avg +1.62% · win 72h 57%. Low-volume drift = accumulation, not weakness. Slow-burn signal — strongest at 72h.
VWAP Reclaim
Price crossed from below to above VWAP Daily in the last hour. Weak standalone (-0.07% at 24h) — used as a confluence signal to confirm directional intent when other conditions are firing.
When this score ≥ 2.5, the market is signalling distribution. When both Perp Long ≥ 3.0 and Distribution ≥ 2.5 simultaneously, the entry signal is suppressed — contradictory signals, wait for resolution.
Funding rate elevated (>0.0005%)
286x · +24h avg -2.11% · +72h avg -4.64% · win 24h 32%. Previously misclassified as bullish in an earlier backtest. Crowded longs paying elevated funding = vulnerable to liquidation cascade, not sustained demand.
SUI underperforming BTC and SOL
215x · +24h avg -1.98% · +72h avg -4.66% · win 24h 28%. Previously misclassified as bullish. SUI lagging both peers predicts continued underperformance, not mean reversion.
Volume spike (rank >60th percentile)
533x · +24h avg -0.97% · +72h avg -3.19% · win 24h 29%. Previously misclassified as bullish. In this dataset, volume spikes accompany distribution and panic — not breakouts. Combined with underperformance: -2.19% at 24h, -5.04% at 72h.
RSI 14H 80–100
13x · +24h avg -3.22% · win 24h 8%. Extreme overbought on the hourly — strongest distribution reading in the dataset. Small sample but consistent.
Price above VWAP Weekly
764x · +24h avg -0.02% · win 24h 41%. Price extended above where volume has transacted on the week — structural distribution zone. The mirror of the strongest bullish condition.
RSI 14H building (55–75)
338x · +24h avg -0.68% · win 24h 41%. Upper-momentum zone without compression. Consistent distribution pattern — momentum without a value entry.
Entry signal & conflict logic
Entry Signal
Perp Long ≥ 3.0 AND ≥ 2 conditions contributing
The 2-condition floor prevents a single heavy condition from solo-triggering in a choppy environment.
Signal Conflict — entry suppressed
Distribution ≥ 2.5 AND Perp Long ≥ 3.0
Both sides elevated simultaneously. Market sending contradictory signals. Entry suppressed until one side resolves.
Data & methodology
Source: Vanguard Agent — a Python bot on a Mac mini, fetching CoinGecko price, Binance US OHLC candles, and TradingView indicators via MCP every hour.
History: 1,679 hourly entries (March 28 – June 3 2026). 4H BB data available from April 9 onward (1,167 entries). Log continues to grow — weights will be recalibrated as data accumulates.
Validation: Each entry paired with forward price returns at +8h, +24h, +72h. Win rate and average return computed per condition and score bucket. Baseline across full dataset: -0.09% at 24h, 45% win rate.
Reclassifications (June 2026): Funding rate elevated, volume spike, and SUI underperforming were all previously scored as bullish based on a ~400-entry early window. The full 1,679-entry dataset reversed all three. These now contribute to Distribution score, not Perp Long.
Weights: Proportional to backtested edge, not from a pure regression — the dataset is too short for reliable regression coefficients. Will be updated as data accumulates.
These signals are probabilistic market context, not trade recommendations. · Back to dashboard