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

## Decision (JSON)

{"direction":"short","leverage":750,"take_profit_bp":15,"take_profit_price":81134.62,"bust_distance_bp":8.53,"bust_price_estimate":81325.82,"expected_net_return_on_margin_pct":88.0,"confidence":0.4,"regime":"trending_down","key_levels":{"support":80772,"resistance":81803},"entry_mid":81256.5,"data_time_utc":"2026-10-16T18:12:51Z","reasoning":"The higher timeframes are bearish: price is down about 117 bp over 6h and 105 bp over 24h, it sits below the 5m EMA21 (81306), the 5m VWAP (81603) and the 15m EMA21 (81593), and the last hour's 1m candles had median downside excursion of 4.5 bp against 2.6 bp upside. The book within 10 bp leans to the ask (59 BTC vs 39 BTC bid). In a 30-bar 1m backtest over the last hour, shorts won slightly more often than longs at every target. The minimum leverage of 750x gives the widest bust distance (8.5 bp) against a 1m realized volatility of about 5 bp per minute, and a 15 bp target keeps about 78% of the raw gain after the haircut, for about 88% return on margin.","risks":"A single 1m bounce of 8.5 bp, about 1.6 standard deviations of a 1m return, liquidates the trade. The last hour's 103 bp range (80772-81803) and the low-side flush followed by a rebound show that a short squeeze toward 81326 is very possible."}

## Evidence (facts: fetched from https://api.hyperliquid.xyz/info at 2026-10-16T18:12:51Z)

| Item | Value | Source query |
|---|---|---|
| Best bid / ask / mid | 81256 / 81257 / 81256.5 | l2Book nSigFigs 5 |
| markPx / oraclePx / premium | 81250 / 81276 / -2.3 bp | metaAndAssetCtxs |
| Funding (hourly) | 0.00125% (longs pay, small) | metaAndAssetCtxs, fundingHistory |
| Open interest / 24h notional | 41,700 BTC / $2.69B | metaAndAssetCtxs |
| prevDayPx | 82117 (price now about 105 bp below) | metaAndAssetCtxs |
| 1h (1m) range | 80772 - 81803; net +19.6 bp over the hour | candleSnapshot 1m |
| 6h (5m) change | -116.8 bp; VWAP 81602.5; EMA9/21 81256/81306 | candleSnapshot 5m |
| 24h (15m) change | -104.8 bp; high 82898; VWAP 81907; EMA21 81593 | candleSnapshot 15m |
| 1m realized vol (stdev of returns) | 10.9 bp over 60m, 5.3 bp over the last 15m (falling) | 1m candles |
| 1m median excursion | up 2.6 bp / down 4.5 bp | 1m candles |
| Book within 10 bp | bids 39.3 BTC, asks 59.1 BTC | l2Book |
| recentTrades | only 10 prints, mostly buys (0.25 vs 0.008 BTC), too few to read anything into | recentTrades |

### Barrier backtest (inference from the last hour of 1m candles)
Each 1m close is treated as an entry and tested over the next 30 bars to see whether it reached the bust or the TP first, using the bar's high and low. If both fall inside the same bar, the bar is counted as a bust, so the result is conservative.

| Leverage (bust) | TP 8 | TP 15 | TP 20 | TP 30 |
|---|---|---|---|---|
| 750 (8.5 bp) long | 25/56 | 21/53 | 17/49 | 16/48 |
| 750 (8.5 bp) short | 33/56 | 22/53 | 18/49 | 13/47 |
| 1000 (5.2 bp) short | 32/59 | 21/57 | 17/55 | 12/53 |

Rough expected value per unit of margin for the short at 750x and 15 bp: 0.42 x 0.88 - 0.58 x 1.0 ≈ -0.21. Every combination tested is negative EV on this sample, and the differences between them are within noise. I picked 750x and 15 bp as a middle ground between hit rate and payoff. Without the "big gains" requirement, an 8 bp target would have been the most attractive (59% hit rate but only about 41% return on margin).

## Facts, inferences, uncertainty
- **Facts:** the values in the table above, taken from live Hyperliquid API responses. Haircut and bust numbers are computed with the brief's formulas: keep = 0.782, net return = 15 bp x 750 x 0.782 ≈ 88%.
- **Inferences:** the downtrend continuing over the next 30 minutes; the book's ask lean reflecting supply rather than spoofing.
- **Uncertainty:** the backtest covers about 56 overlapping windows from one hour, so it is not statistically meaningful. At 750-1000x the bust distance is inside normal 1m noise, so the outcome is close to a coin flip. Confidence is 0.40.
- **Unanswered:** I couldn't confirm the macro calendar or news for 2026-10-16. One web search returned only forecast pages and no current news, so no scheduled release was ruled in or out. There was no liquidation-heatmap data and no cross-exchange order-flow data.