# BTC-USD perp trade decision (PaperTrade, Hyperliquid mid)

## Decision

{"direction":"long","leverage":750,"take_profit_bp":12,"take_profit_price":81776.51,"bust_distance_bp":8.53,"bust_price_estimate":81608.8,"expected_net_return_on_margin_pct":68.4,"confidence":0.42,"regime":"trending_up","key_levels":{"support":81570,"resistance":81765},"entry_mid":81678.5,"data_time_utc":"2026-10-16T20:55:27.087Z","reasoning":"BTC is recovering from the 24h low of 80772: 15m closes form higher lows, price is 41 bp above the 6h VWAP and 15-18 bp above the 5m EMA21 and 15m EMAs, the 1h move is +16.8 bp, and resting bids slightly outweigh asks near the mid. 1m volatility is low and steady (about 3.5-4.3 bp per minute), so a 750x bust distance of 8.5 bp sits beyond the 90th-percentile 1m adverse wick (4.2 bp). A 12 bp target keeps about 71% of the raw gain after the haircut. In a replay of the last hour of 1m bars, it had the best expected value of all the leverage and target pairs tested (56% TP-first vs 30% bust-first within 30 min).","risks":"A move back to the 81570-81600 pullback zone (the last 3 minutes already slipped from 81759 to 81678, and recent tape was sell-heavy) would liquidate before the 81765 high is retaken. On 5m bars over 6h the same setup busted more often than it hit, because 5m candles routinely span 20+ bp."}

## Evidence (facts: fetched from https://api.hyperliquid.xyz/info at 2026-10-16T20:55:27Z)

All raw responses are saved in `data/raw.json` and all computed metrics in `data/metrics.json`. They were produced by `scripts/fetch_and_analyze.py`.

| Metric | Value | Source |
|---|---|---|
| Best bid / ask / mid | 81678 / 81679 / **81678.5** | l2Book nSigFigs 5 |
| markPx / oraclePx / premium | 81679 / 81708 / -3.55 bp | metaAndAssetCtxs |
| Funding (hourly), last 6 h | +0.00125% (≈1e-5 to 1.25e-5) | metaAndAssetCtxs, fundingHistory |
| Open interest / 24h notional volume | 41,232 BTC / $2.87B | metaAndAssetCtxs |
| prevDayPx | 82100 (mid is 51 bp below) | metaAndAssetCtxs |
| 24h high / low (15m) | 82898 / 80772; mid at 43% of the range | candles 15m |
| 6h high / low (5m) | 82072 / 80772 | candles 5m |
| 1h high / low (1m) | 81765 / 81480; 1h change +16.8 bp | candles 1m |
| Distance to 24h VWAP / 6h VWAP | -9.9 bp / +41.2 bp | computed |
| 1m EMA9 / EMA21 distance | -1.7 bp / +0.7 bp; EMA9 slope +0.13 bp per bar | computed |
| 5m EMA9 / EMA21 distance | +4.0 / +14.8 bp; EMA9 slope +1.4 bp per bar | computed |
| 15m EMA9 / EMA21 distance | +17.6 / +18.2 bp; EMA9 slope +5.1 bp per bar; last 5 bars +45 bp | computed |
| 1m realized vol (stdev of close-to-close) last 15 / 30 / 60 min | 4.2 / 3.5 / 4.3 bp | computed |
| 1m adverse wick (open−low), median / p90 | 1.0 / 4.2 bp | computed |
| 5m adverse wick (open−low), median / p90 | 6.6 / 19.8 bp | computed |
| Book within ±20 levels (about ±2.5 bp at 5 sig figs) | bids 57.6 BTC vs asks 48.5 BTC (imbalance +0.09) | l2Book |
| recentTrades (last 10 prints, about 2 s) | 0.001 BTC bought vs 0.606 BTC sold | recentTrades |

### Path replay (inference; in-sample and small)

For every 1m bar in the last hour I took that bar's open as the entry. I then walked forward up to 30 bars and recorded whether the TP or the bust was touched first. If a single bar touched both, I counted it as a bust (the conservative choice). I repeated this on the 6h of 5m bars with a 6-bar horizon. EV is p(TP)×net return on margin − p(bust)×100%.

| Leverage | TP | Long 1m: TP / bust / open of 61 | EV 1m | EV 5m |
|---|---|---|---|---|
| 750 | 12 | 34 / 18 / 9 | **+8.6%** | -34.5% |
| 800 | 12 | 33 / 20 / 8 | +6.7% | -36.7% |
| 800 | 10 | 38 / 19 / 4 | +5.7% | -39.8% |
| 1000 | 12 | 30 / 27 / 4 | +0.6% | -37.0% |
| 750 | 15 | 18 / 18 / 25 | -3.6% | -35.0% |

None of the short combinations reached the top 20 by 1m EV. The full grid is in `data/metrics.json`.

## Reasoning chain (inference)

1. **Direction: long.** The 15m and 5m structure is a rebound from the 24h low: higher lows, rising EMAs, and price above the 6h VWAP. The 1m chart is flat to slightly positive. The book is mildly bid-heavy. The perp trades 3.5 bp below the oracle while funding is still positive, which leaves room for the basis to converge upward. Against this, the 24h trend is still down (-82 bp), the mid is 10 bp below the 24h VWAP, and the latest few seconds of tape were sell-dominated. That is why confidence is only moderate.
2. **Leverage 750.** At 1000x the bust distance is 5.2 bp. The 1m p90 adverse wick is 4.2 bp and 1m vol is about 4 bp per minute, so a 5.2 bp bust is roughly a 1.3-sigma one-minute move. In the replay, raising leverage cut EV monotonically because busts grew faster than gains. 750x gives 8.5 bp of room.
3. **Target 12 bp.** 12 bp keeps about 71% of the gain after the haircut, versus 61% at 5 bp. It sits at the 1h high (81765) plus about 1 bp. It is the 1m replay optimum: 10 bp hits more often but pays less, and 15 bp often does not hit within 30 min.

## Uncertainty and unanswered questions

- The replay covers only 61 one-minute starts that overlap one another, all from a rising hour. It is not an out-of-sample estimate. The 5m replay is negative for every combination, which means wider 5m swings would bust this trade. Treat the 56% hit rate as optimistic.
- I did not check news or scheduled macro releases (a time limit). 20:55 UTC on a Friday is after the US cash close, so I *assume* no scheduled release falls inside the 30-minute window. I have not verified this.
- I did not use liquidation heatmaps or other venues' order flow.
- The l2Book at 5 significant figures shows only about ±2.5 bp of depth, so I could not measure resting liquidity out to the TP or bust prices.
- Prices move after the fetch time. The entry, TP and bust prices assume a fill at 81678.5.