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

## Decision

{"direction":"long","leverage":500,"take_profit_bp":12,"take_profit_price":81855.609,"bust_distance_bp":15.2,"bust_price_estimate":81633.2286,"expected_net_return_on_margin_pct":44.84,"confidence":0.4,"regime":"ranging","key_levels":{"support":81619,"resistance":81944},"entry_mid":81757.5,"data_time_utc":"2026-10-14T22:04:07.700Z","reasoning":"BTC is down about 210 bp on the day and sits in the bottom 10% of its 24h range (81497-84103), but over the last 6h it has ranged between 81497 and 82072 with a slight upward drift (+13.6 bp). The latest 1m candle swept to 81685 and closed back at 81758, and the perp trades about 3.2 bp below the oracle, which leans toward a mean-reversion bounce. In a 60-minute 1m first-touch backtest, longs beat shorts at every leverage and target tested. Realized volatility is about 3.7 bp per minute (roughly 20 bp per 30 min), so any bust distance under 10 bp tends to get hit by noise. That is why I chose the minimum 500x (15.2 bp bust, below the 22:03 sweep low). The 12 bp target keeps about 75% of the raw gain after the haircut and sits just under the 81850-81940 congestion.","risks":"The most likely way this gets liquidated is the 24h downtrend resuming. A break of the 81685 sweep low and the 81619 1h low would run through the 81633 bust, helped by heavier asks within 10 bp of mid (110 BTC of asks vs 47 BTC of bids). On this data the expected value is negative under conservative assumptions; see the Uncertainty section."}

## Evidence (facts: values returned by the Hyperliquid info API, fetched 2026-10-14T22:04:07Z)

All data came from `POST https://api.hyperliquid.xyz/info`. The script that fetched it and computed the metrics is `scripts/analyze.mjs` in this repository.

| Item | Value | Source query |
|---|---|---|
| Best bid / ask, mid | 81757 / 81758, mid 81757.5 (spread 0.12 bp) | `l2Book` nSigFigs 5 (second fetch, taken last) |
| markPx / oraclePx / premium | 81757 / 81784 / −0.000318 (perp ≈3.2 bp below oracle) | `metaAndAssetCtxs` |
| Funding (hourly) | 0.0000125, unchanged across the last 6 hourly prints | `metaAndAssetCtxs`, `fundingHistory` |
| Open interest / 24h notional volume | 40,575 BTC / $2.94B | `metaAndAssetCtxs` |
| prevDayPx / 24h change | 83510 / −210 bp | `metaAndAssetCtxs` |
| 24h high / low (15m candles) | 84103 / 81497; mid sits at 10% of the range | `candleSnapshot` 15m |
| 6h high / low (5m candles) | 82072 / 81497; 6h ROC +13.6 bp | `candleSnapshot` 5m |
| 1h high / low (1m candles) | 81944 / 81619; 60m ROC −0.2 bp, 15m ROC −11.1 bp | `candleSnapshot` 1m |
| 24h VWAP | 82613 (mid is −104 bp below it) | 15m candles |
| EMAs (distance of mid from EMA21) | 1m −2.3 bp, 5m −3.9 bp, 15m −9.3 bp; EMA9 slopes negative on all three | computed |
| Realized vol (stdev of 1m log returns) | 3.6 / 4.1 / 3.7 bp over 15 / 30 / 60 min (flat, not expanding) | computed |
| 5m realized vol | 7.8 bp (last hour) vs 8.6 bp (6h) | computed |
| 1m excursion | median up 1.0 bp, median down 1.1 bp, mean range 3.9 bp, max 12.2 bp | computed |
| 5m excursion | median up 5.0 bp, median down 5.7 bp, mean range 12.8 bp | computed |
| Last 1m candles | 22:02 close 81700; 22:03 o 81701, low 81685, close 81758 (sweep and reclaim) | 1m candles |
| Book within ±10 bp | bids 47.3 BTC, asks 109.9 BTC (imbalance −0.40, ask-heavy) | `l2Book` |
| 1m volume | last 15 min avg 9.9 BTC/min vs prior 45 min 9.1 | 1m candles |

### Backtest (computed from the last 61 one-minute candles)

From every 1m open I walked forward up to 30 candles and recorded whether the take-profit or the bust level was touched first. When both were touched inside the same candle I counted it as a loss, which is conservative. EV is p·return − (1−p)·100%, using only paths that reached one level or the other.

| Lev | Bust bp | TP bp | Dir | win/loss/open | p(win) | Return if TP hits | EV % of margin |
|---|---|---|---|---|---|---|---|
| 500 | 15.2 | 12 | long | 29/18/13 | 0.62 | 44.8% | −10.6 |
| 550 | 13.4 | 12 | long | 29/19/12 | 0.60 | 49.3% | −9.8 |
| 600 | 11.9 | 12 | long | 28/21/11 | 0.57 | 53.8% | −12.1 |
| 700 | 9.5 | 12 | long | 27/25/8 | 0.52 | 62.8% | −15.5 |
| 500 | 15.2 | 12 | short | 25/24/11 | 0.51 | 44.8% | −26.1 |
| 600 | 11.9 | 12 | short | 21/28/11 | 0.43 | 53.8% | −34.1 |

The full grid covers leverage 500–1000, TP 5–30 bp and both directions; it is regenerated by the script into `metrics.json`. Longs scored better than shorts in every cell. Leverage above 700x lowered EV further, because the bust distance (≤ 7.7 bp) is about two minutes of typical noise.

### Haircut check
The brief's formula with move = 11.8 bp (12 bp minus the 0.2 bp deadband) gives scale × 0.98 = 74.7% of the raw 12 bp. That works out to 44.84% return on margin at 500x, consistent with the brief's table (≈72% at 10 bp, 77% at 15 bp).

## Inferences (my judgment, not measured)
- **Direction (long):** I read the setup as a range with a fresh sweep and reclaim of 81685, made inside a 6h range whose floor (81497) has held. The perp is at a discount to the oracle, which tends to close by the perp rising, and longs beat shorts in the backtest. Against this, the 24h trend, all EMA slopes and the ask-heavy book point down. I weighed this as only a mild long edge.
- **Leverage (500x):** with about 20 bp of 30-minute sigma, the survival odds drop quickly as the bust distance falls below 10 bp. The extra return from higher leverage (62.8% at 700x vs 44.8% at 500x) did not make up for the higher bust rate in the backtest.
- **Target (12 bp):** the backtest's EV peaked around 10–15 bp. 12 bp keeps about 75% of the gain after the haircut and sits below the 81850–81944 supply zone left by the 21:35–21:45 5m candles.
- **Regime "ranging":** 1h ROC is ≈0 and realized volatility is flat, but the 24h trend is down. "Ranging" is the closest label.

## Uncertainty and limits
- **Expected value is negative on this sample.** Every leverage/TP/direction cell in the 60-minute backtest has negative EV, because the 500x breakeven win rate at a 12 bp TP is about 69% and the measured rate was 62%. The brief requires a direction, so this is a forced best-of-available choice, not a positive-edge trade. Confidence of 0.4 reflects that.
- The backtest uses only 60 overlapping, autocorrelated windows from a single hour, so its statistical power is weak. A 1m candle cannot show the order in which high and low printed, hence the conservative tie rule.
- `recentTrades` returned only 10 prints covering 3.6 seconds, so CVD and average trade size are not meaningful and were not used.
- **Not checked:** news, the scheduled macro calendar, liquidation heatmaps, and other venues' order flow. I made no web lookups for these. 22:04 UTC is after US cash-market hours, which I infer is a lower-catalyst window, but I did not verify that against a calendar.
- Open-interest change could not be computed because only a single OI snapshot was taken.
- **Price drift:** the decision is based on the 22:04:07Z mid. If entry happens later at a different mid, TP and bust shift with it.

## Unanswered questions
- Will the 81685/81619 lows hold if the 24h downtrend resumes? This decides between take-profit and bust.
- Was there any news or liquidation cluster behind the −210 bp daily move? I did not research this.