Correct-Order SMA Persistence: LTC and LINK Hyperliquid Long/Short

Hypothesis

Test the submitted moving-average ordering mechanism independently on LTC and LINK perpetuals using 4-HOUR bars. Preserve the five SMA periods, five-bar confirmation, rising ADX filter and ordering-break exit. Replace the source's fixed forex lot size with capital-relative volatility sizing. This is a source-derived replication hypothesis, not an established crypto edge. Maintain incremental indicators and bounded buffers rather than rescanning historical data.

Preserves the user's specific mechanism while moving away from saturated BTC futures and long-only experiments. Hyperliquid represents 7.7% of experiments against a 20% target; long-only represents 83.1% against a 55% ceiling. Two independent long/short sleeves improve venue, instrument and direction breadth without inventing a cross-venue mechanism. Corpus concentration was empty, so exact symbol shares cannot be established. Generic trend failures warrant low priority and a fixed-parameter experiment.

negative_expectancy (code: no_edge). This is the first measured report. Fee-floor comparison: avg_trade_return_pct is -1.20% against the +0.15% Hyperliquid floor. PF is 0.60, Sharpe -0.78 and total return -6.6% over 77 closed trades in 773 days. Funding and impact are both folded in, and nothing is open at the end. Commission is only 4.6% of gross, so the gross signal itself has no edge. It loses in all three volatility regimes and in all three calendar years. The shorts lose $6.7k (LTC -$3.8k, LINK -$2.9k), while LTC longs make +$0.46k on 14 trades and LINK longs lose $0.31k. The strict five-SMA ordering with 5-bar confirmation and rising ADX enters late and whipsaws on alt perps. There is no tunable region to optimize toward, and the sample is too thin anyway. QA's edge concern is confirmed by the full available history. This verdict covers the fixed 10/20/50/100/200 SMA-ordering mechanism on LTC/LINK Hyperliquid 4H bars only.

Implementation

Two independent Hyperliquid long/short sleeves using 4-hour SMA10/20/50/100/200 ordering confirmed for five bars, ADX above 20 and rising, ordering-break exits, entry-ATR close-based stops and 180-bar holding limits. Each entry risks 0.5% of equity before gaps and costs, with 25% per-sleeve and 50% combined entry exposure caps.

Recovered the preceding actual source from the task inbox because the referenced outbox artifact is unavailable. Preserved the hypothesis mechanism and fixed parameter values, restored the described primary incremental fast path, bounded the ADX seed buffer, removed silent parameter clamps and removed current-ATR substitution for a missing entry anchor. No backtests or protected market data were accessed. The supplied frozen plan must be enforced by the current pipeline, including the exclusive 2026-04-18 final-exam boundary and counted sign-off reads; strategy code cannot prevent upstream data reads. The exact requested outbox file could not be written because this session permits only filesystem reads; this JSON is the complete bound submission.

Verification Results

If the source required a fresh confirmation after a stop-out, reset bull_streak/bear_streak (or set a re-arm flag) when a stop or time exit fires.

When a sleeve is closed by the ATR stop or the 180-bar time stop while the SMA ordering still holds, its streak counter stays >= 5. It can therefore re-enter on the very next bar with no new five-bar confirmation. Exits after an ordering break correctly reset the streak to 0. Whether to re-enter after a stop is not stated in the hypothesis.

Optional: record the entry ATR in on_position_opened instead.

position_size() stores _ltc_entry_atr before the base template checks min_notional. If the order is then dropped, the stored ATR is stale. It is overwritten on the next successful sizing, so the effect is minimal.

None required. A resting stop_market would keep loss to the stated risk inside the bar if that is wanted later.

Stops are checked only at the 4H bar close against the entry-time ATR, so gaps or moves inside a bar can exceed the 0.5% equity risk. The hypothesis describes close-based stops, so this matches the hypothesis.

Make sure the full backtest folds in Hyperliquid funding history for LTC and LINK.

Funding is not modeled for these Hyperliquid perp positions (funding_folded=false, 0 funding events), even though holding periods run from days up to 30 days.

Sandbox (773 days, both sleeves): 77 closed trades, avg -1.06% of notional per trade, PF 0.64, Sharpe -0.70, losing in all three vol regimes. Fees are not the problem: Hyperliquid round-trip is about 0.09% against an average 3xATR(4H) stop distance of several percent. The losses look like whipsaw from the strict five-SMA ordering on alt perps. At roughly 35-40 trades a year over two sleeves, full Hyperliquid history (from about 2023) gives only about 100-120 trades, which is borderline for statistical power. Funding is not folded (funding_events_available=0) even though positions are held for multiple days on perps, so the backtest leaves out a real carry cost. The analyst should look at the per-sleeve split and how much of the P&L comes from the close-based stop versus the ordering-break exit.

Backtest Review

Sharpe

-0.78

Total return

-6.58%

Max drawdown

6.58%

Trades

77

Win rate

35.1%

Profit factor

0.60

Accounting is valid. Funding and impact are folded in, there are no open positions at the end and the inputs are clean.

Both sleeves trade as the hypothesis describes: they take long and short entries from the SMA ordering, with volatility-scaled size (avg_position_pct 6.8%).

Fee-floor comparison: avg_trade_return_pct is -1.20%, against the +0.15% Hyperliquid floor. PF is 0.60, Sharpe -0.78 and win rate 35%.

Fees are not the cause. Commission is only 4.6% of gross; the gross signal itself loses.

It loses in every volatility regime and in every calendar year (2024, 2025, 2026).

Per-sleeve split: LTC long +$459 (14 trades), LTC short -$3,832 (21), LINK long -$314 (18), LINK short -$2,897 (24). Nearly all of the loss comes from the bearish-ordering shorts, which whipsaw on both alts.

Win rate 0.35 with PF 0.60 is a failed payoff structure for a trend design. The 77-trade sample is too thin for optimization to be meaningful.

Iteration History

Verification failed (Layer 1.5 — hypothesis/config consistency) [class=hypothesis_mismatch]: - optimization_plan fixes leverage at 1.0

sandbox_timeout: Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit on a ~20000-bar window (2400 days of LTCUSD.HYPERLIQUID-4-HOUR-LAST-EXTERNAL). This almost always means per-bar work that scales with history — e.g. rescanning the full funding/supplementary series, or rebuilding a list and calling min()/sorted() inside calculate_signal()/on_bar() on every bar. Precompute sorted timestamp arrays ONCE in __init__ and use bisect, or cache lookups keyed by timestamp, so per-bar cost is O(log n) not O(n).

Verification failed (Layer 1 — static analysis): - Syntax error: invalid character '—' (U+2014) (<unknown>, line 1) (line 1)

Outcome Summary

With a strict five-SMA ordering, 5-bar confirmation and rising ADX, entries on alt perps come late and whipsaw, especially on the short side; a mechanism with no gross edge cannot be rescued by tuning or fee work.

The analyst's backtest review returned 'abandon' (negative_expectancy, code no_edge) before optimization: commission was only 4.6% of gross, so the gross signal itself lost money, and it lost in all three volatility regimes and all three calendar years. Optimization, analyst, risk and promotion stages were not reached.

A source-derived replication of a moving-average ordering trend system on LTC and LINK Hyperliquid perpetuals (4-hour bars), going long/short when five SMAs (10/20/50/100/200) stay correctly ordered for five bars with a rising ADX, exiting on an ordering break, with capital-relative volatility sizing.

Over 773 days and 77 closed trades it returned -6.58% with Sharpe -0.78, profit factor 0.60, win rate 35% and avg_trade_return_pct of -1.20% against the +0.15% Hyperliquid floor. Max drawdown was 6.58%; shorts lost about $6.7k while LTC longs made +$459.

Analysis

The correct order (version 1.0.01)