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

## Decision

```json
{"direction":"short","leverage":800,"take_profit_bp":18,"take_profit_price":81293.91,"bust_distance_bp":7.7,"bust_price_estimate":81503.21,"expected_net_return_on_margin_pct":113.5,"confidence":0.35,"regime":"volatile_choppy","key_levels":{"support":81278,"resistance":81729},"entry_mid":81440.5,"data_time_utc":"2026-10-16T19:33:41Z","reasoning":"BTC is down 125 bp from the previous day's price and 43 bp below the 24h VWAP. The 18:40-18:55 dump to 80890 bounced to 81729 at 19:27 on a volume spike and has since faded 36 bp on mostly lower 1m highs; the 1m EMA9 slopes down and 5-minute ROC is -17 bp. The perp trades 4.5 bp below oracle, and ask walls of 9.5 and 14.5 BTC sit at 81449/81454, between entry and the bust price. 800x keeps the bust (7.7 bp) above those walls and above the median 1m up-excursion (2.5 bp), and an 18 bp target sits just ahead of the 81278 swing low while keeping about 78% of the raw gain after the haircut.","risks":"One more bounce leg like the 19:25-19:27 squeeze (+22 bp in two minutes) would liquidate the position in under a minute. A short-covering spike back toward oracle (81478) and the 81503 bust would also do it, since the perp discount means shorts are crowded."}
```

Data fetched from `https://api.hyperliquid.xyz/info` at 2026-10-16 19:33:41 UTC (Friday, about 27 minutes before the US equity close). Fetch and metric scripts are in `tools/` (`fetch.py`, `analyze.py`, `tail.py`, `decide.py`).

## Facts (measured from the Hyperliquid API at fetch time)

| Item | Value | Source query |
|---|---|---|
| Best bid / ask / mid | 81440 / 81441 / **81440.5** | `l2Book` nSigFigs 5 |
| markPx / oraclePx / premium | 81440.8 / 81478.0 / −0.000454 (−4.5 bp) | `metaAndAssetCtxs` |
| Funding (hourly) | 0.0000125, unchanged over the last 6 h apart from one 0.0000099 print | `metaAndAssetCtxs`, `fundingHistory` |
| Open interest / 24h notional volume | 41,461 BTC / $2.88B | `metaAndAssetCtxs` |
| prevDayPx | 82473 (mid is −125 bp from it) | `metaAndAssetCtxs` |
| 24h high / low (15m candles) | 82898 / 80772; mid sits at 31% of the range | `candleSnapshot` 15m |
| 24h VWAP / 1h VWAP | 81791 (mid −43 bp) / 81271 (mid +21 bp) | 15m / 1m candles |
| 1m realized vol (std of 1m log returns) | 6.5 bp (15 m), 6.5 bp (30 m), 9.5 bp (60 m): **contracting** | 1m candles |
| Median 1m up / down excursion from the open | 2.5 / 2.0 bp; median 1m range 7.6 bp | 1m candles |
| Median 5m up / down excursion | 7.5 / 11.1 bp | 5m candles |
| 1m: EMA9 slope (3 bars), 5-bar ROC | −3.8 bp, −17.2 bp | 1m candles |
| 5m: EMA9 slope, 5-bar ROC | +15.4 bp, +37.7 bp (the bounce) | 5m candles |
| 15m: change over 24 h | −125 bp | 15m candles |
| Recent price path | dump 81451→80890 (18:40–18:55); bounce to 81729 at 19:27 (1m volume 249 and 207 BTC); mostly lower 1m highs since then, last close 81440 | 1m/5m candles |
| Book within about 20 bp of mid | bids 50.1 BTC, asks 67.6 BTC (asks heavier); largest walls: bid 12.3 @81420, ask 14.5 @81454 and 9.5 @81449 | `l2Book` |
| recentTrades | only 10 prints over 6.8 s; slightly buy-heavy (+0.018 BTC), too small a sample to use | `recentTrades` |

## Backtest on the last hour of 1m candles (fact, with a stated method)

For each 1m open in the last 61 minutes, the backtest walked forward up to 30 bars and recorded whether the take-profit or the bust distance was touched first. When a single bar touched both, it counted as a loss. Selected rows (W/L = wins/losses out of 61):

| Lev | Bust bp | TP bp | Net RoM if TP | Long W/L | Short W/L |
|---|---|---|---|---|---|
| 750 | 8.5 | 15 | 87% | 26/34 | 23/35 |
| 800 | 7.7 | 15 | 93% | 25/36 | 21/37 |
| 800 | 7.7 | 20 | 128% | 23/38 | 15/41 |
| 900 | 6.3 | 20 | 143% | 20/41 | 13/43 |
| 1000 | 5.2 | 20 | 159% | 18/43 | 10/46 |

That hour included a 22 bp upside squeeze, so the long side scored better. Almost every cell has negative EV. That is expected: with no directional edge, the chance of hitting the TP before the bust is roughly bust/(bust+TP), which is 30% for 7.7/18. This tool's payoff structure only pays off when the directional call is right.

## Inferences (my judgement, not measured)

- **Direction: short.** The higher timeframe is bearish: −125 bp on the day, below the 24h VWAP, lower part of the 24h range. The 19:25–19:27 bounce was a volume-spike squeeze that was immediately rejected, and price has bled 36 bp in six minutes since. The perp discount to oracle (−4.5 bp) shows active selling on Hyperliquid itself. The asks are heavier near mid, with walls just above entry.
- **Leverage: 800x.** The 7.7 bp bust puts the liquidation price (81503) above the 81449/81454 ask walls and about 3× the median 1m up-excursion. At 900–1000x the bust (5.2–6.3 bp) is inside one ordinary 1m range (median 7.6 bp), so it is likely to be hit by noise alone.
- **TP: 18 bp (81293.9).** This sits just ahead of the 19:18 swing low (81278) and about halfway between entry and the 61.8% retracement of the bounce (about 81211). It needs roughly 3 minutes of the current 1m volatility in one direction, which is plausible within 30 minutes if the fade continues. At 18 bp the trader keeps about 78% of the raw gain, giving +113.5% on margin.
- **Regime: volatile_choppy.** The 5m and 1m trends disagree (the 5m is still up from the bounce, the 1m is down), and the last hour saw ±20–60 bp swings, even though 1m volatility is now contracting.

## Uncertainty

- Confidence is **0.35**. Under a driftless model, the TP-before-bust probability for 7.7/18 is about 30%, so the trade needs a real downside drift to have positive EV. The evidence for that drift is moderate, not strong.
- The −4.5 bp premium cuts both ways: crowded shorts can be squeezed back toward oracle (81478), which is only 2.9 bp from the 7.7 bp bust.
- The 1m backtest covers only one hour, and it favored longs.

## Unanswered / not checked

- **News and scheduled macro events were not checked**; no web search was run within the time budget. Friday 19:30–20:00 UTC normally has no scheduled US data releases, but I did not verify that for this date.
- External liquidation heatmaps and other exchanges' order flow (Binance, CME) were not pulled.
- Open-interest change could not be computed from a single snapshot.
- The book snapshot (nSigFigs 5, $1 buckets) only covers about ±20 bp of mid, so liquidity beyond the bust and TP prices is unknown.