SpotMajorsBreadthScaledLongCashBasketWeekly

Crypto-Majors BREADTH-Scaled Long/Cash Basket, Long-Only (BINANCE_SPOT, 10 Equal-Weight Deep Majors, Weekly: Invested Fraction = the SHARE of Names Above Their 100-Day SMA — Scale INTO Stablecoin as Market Breadth Deteriorates — a Cross-Sectional Signal No Deployed Basket Uses, UNLEVERED Spot So Liquidation Is Structurally Impossible, 2-Parameter)

A LONG-ONLY, diversified, SELF-DE-RISKING basket on BINANCE_SPOT that is engineered against the two killers that dominated this session's deaths: (1) TOTAL LIQUIDATIONS (AVAX sweep, SOL/AVAX pairs, BTC grid all hit -100% in a bear/vol event because the developer stacked near-full LEVERAGED gross exposure) and (2) the >50% single-regime DRAWDOWN that hard-abandoned every long-only leveraged momentum basket. Both are structurally eliminated here: it trades on the SPOT CASH account (no leverage, no margin) so liquidation is mathematically impossible, and its exposure AUTOMATICALLY scales toward stablecoin as the market weakens. The signal is MARKET BREADTH — the fraction of a fixed 10-name deep-history majors universe (BTC, ETH, SOL, BNB, XRP, ADA, AVAX, LINK, DOGE, LTC) trading above their own 100-day SMA. That fraction IS the invested percentage: broad uptrend (8/10 above MA) → ~80% invested, equal-weighted across the qualifying names; deteriorating breadth (2/10) → ~20% invested, 80% in USDT; bear (0/10) → fully in cash. Breadth is a cross-sectional aggregate signal NO deployed basket uses (the deployed baskets use per-name relative-strength momentum and per-name negative funding — L24's requirement for a genuinely new signal input, from the dispersion family L24 names). It is NOT a rank/rotation basket (equal-weight ALL qualifying names, no top-K sort → not the L12/L24 duplicate), NOT concentrated (max ~10% per name → no single-name blowup), NOT carry/options/liquidation-feed/single-name-reversion/trend-timing-of-one-asset. Because breadth deteriorates AHEAD of the price index in a topping process, the book de-risks early. Pure OHLCV, data-safe; 2 core parameters (SMA lookback, weekly evaluation cadence).

Iteration 2 makes the two minimal fixes the reviewer required, starting from previous_code and leaving all previously-passing signal/structure logic intact. (1) REBALANCE-GATE FREEZE: the old rebalance computed `above` in a loop that `return`ed and aborted the ENTIRE weekly rebalance whenever any single one of the 10 legs lacked 100 bars or stopped updating — which froze the book at 2021-10 and never re-entered the 2023-2024 bull. It now skips any name without an evaluable SMA and computes breadth + allocates over the AVAILABLE names only (guarding `if not above: return` so a fully-empty bar just waits without consuming the week); `held`, the sell loop, and the buy list all iterate over the evaluable set. (2) ACCOUNTING ARTIFACT: I confirmed via base_template that get_account_equity() returns the USDT stable balance only (not net-liq), so _total_value() = cash + coins marked at last close is a correct net-worth figure and is not double-counting. The impossible +262% single-day / +268% monthly figures were driven by the freeze itself — a year of unrealised gains booked on a handful of dump-bars produces lumpy realised-PnL daily-return spikes. Fixing the freeze so the book rebalances weekly across the whole 2017-2026 window realises gains regularly and keeps each bar's realised PnL bounded by the ~10%-per-name cap. As additional hardening for the 'settled-USDT miscalculation' concern, deferred buys now size off the FRESHLY-settled portfolio value at execution time (recomputing total and target_per on the buy bar) instead of a stale rebalance-time target, and each buy is still capped at available settled USDT so total deployment can never exceed ~100% of net worth. Daily bars match the macro (weeks-to-months) breadth horizon; 100-day SMA warmup is a small fraction of the multi-year window so the gate fires early and often.

Iteration 2 did not resolve either flagged defect, and the result is unchanged from iteration 1. The freeze persists: entry_diag.bars_seen is still exactly 379 and all trading is still confined to 2020-10→2021-10 with zero activity 2022-2026 (28 trades total). The developer's gate-skip fix could not help because the binding root cause is not the rebalance gate — it is a BINANCE_SPOT data-coverage wall: the primary BTCUSDT.BINANCE_SPOT feeds only ~379 bars into _process_primary, so the strategy has no data to trade past 2021 and stays frozen. The accounting artifact also remains unfixed and worse: a +337.8% single-day return (2021-06-21) and +214%/+337% months are physically impossible for an unlevered spot basket capped at ~10%/name (≤100% invested), making PF 33.5, calmar 686, avg_trade_return_pct 396% and the +2056% 2021 all uninterpretable garbage; max_drawdown is 60.3% (>50%) even in the corrupted window. Two targeted iterations have produced no change, so there is no path to a clean, evaluable backtest within the review loop — the strategy cannot be assessed on the available spot data and the PnL artifact is unresolved. Note for the research lead / data engineer: the breadth + unlevered-spot concept is genuinely sound, but it requires (a) full BINANCE_SPOT daily history for all 10 majors backfilled into the catalog (the current ~379-bar primary coverage is the real blocker) and (b) a corrected rebalance/settlement PnL path; until the data is available it cannot be evaluated. Abandoning rather than iterating a third time on an unchanged, non-functional result.

Long-only, unlevered BINANCE_SPOT (CASH) breadth-scaled basket over 10 deep majors (BTC/ETH/SOL/BNB/XRP/ADA/AVAX/LINK/DOGE/LTC). Weekly, each name trading above its own 100-day SMA gets an equal ~1/10 slice of the portfolio; the balance sits in USDT. Invested fraction therefore equals market breadth (share of names above SMA) — the book scales into stablecoin as breadth deteriorates and back into coins as it recovers. Rebalance sells names that fell below their SMA on the rebalance bar and buys newly-qualifying names the next bar off settled USDT; stable holds are kept to limit churn/fees. No leverage, no shorts — liquidation is structurally impossible.

Evaluate the FULL-history run (now unfrozen): confirm (a) the +262%/+268% artifact is gone, (b) the book re-enters bulls (2023-2024) net of whipsaw, and (c) breadth de-risking holds DD well below 50% through 2022. Abandon if it only whipsaws without bull up-capture.

The sandbox is essentially unchanged from iter-1 (-11.75%, Sharpe -1.84, max_drawdown 12.0%, win_rate 0.0, PF 0.0, 14 trades) because the recent ~1yr window has all 10 legs at full 100-day history, so the freeze fix doesn't alter it — the fix matters only for the FULL 2017-2026 backtest. The win 0.0 / PF 0.0 remain the SMA-whipsaw signature in an adverse down/choppy window (re-verified: buy-above/sell-below polarity correct, winners held to trend-break, sizing correct, book correctly de-risked to cash at a contained 12% DD), NOT a code defect. Edge/regime outcome for full history, not a recodable fault.

Optionally include held-but-unevaluable names in a defensive close; not blocking given the deep-history universe.

Minor stranded-leg edge case: `held` is built only from evaluable (`above`) names and the sell loop iterates `above`, so a currently-HELD name that loses its evaluable SMA this week is neither sold nor tracked — it stays open until evaluable again. Negligible for the 10 liquid majors with continuous history, but a latent asymmetry (a stranded holding can't be de-risked while its data is missing).

The breadth signal + unlevered-spot concept remains sound and novel; the developer did attempt the requested gate-skip fix

No leverage, so liquidation is structurally impossible (the one design goal that holds)

STILL non-functional after iteration 2: bars_seen unchanged at 379, all trading confined to 2020-10→2021-10, zero activity 2022-2026 (28 trades total) — the book is still frozen

Root cause is a BINANCE_SPOT data-coverage wall: the primary feeds only ~379 bars, so the strategy has no data to trade past 2021 — the gate-skip fix could not address this

Accounting artifact UNFIXED and worse: +337.8% single day (2021-06-21), +214%/+337% months — impossible for an unlevered ≤100%-invested spot basket; PF 33.5, avg_trade_return_pct 396%, calmar 686 are all garbage

max_drawdown 60.3% (>50%) in the corrupted 2021-only window; metrics uninterpretable, single-regime

Two targeted iterations have produced no change — no path to a clean backtest within the review loop

The backtest is non-functional and its metrics are uninterpretable — do NOT optimize it. Two specific fixes are required before this can be evaluated: (1) REBALANCE-GATE FREEZE: in _process_primary the loop `for iid in self._names: s = self._sma(iid); if s is None ...: return` aborts the ENTIRE weekly rebalance whenever ANY single one of the 10 spot names lacks 100 bars or has a data gap. All trades/returns stop at 2021-10 and there is zero activity through 2026 (bars_seen only 379 vs bars_processed 3264, 23 trades total) — the book froze and never re-entered the 2023-2024 bull. Fix: skip names with missing/insufficient data and compute breadth + allocate over the AVAILABLE names, never abort the whole rebalance. Also verify BINANCE_SPOT catalog coverage for all 10 names spans the full window (SOL/AVAX/LINK/DOGE spot may have limited history, and the 379 primary-bar count suggests BTC spot data itself may be incomplete — trigger on-demand collection if so). (2) ACCOUNTING ARTIFACT: the +262% single-day (2021-06-21) and +268% monthly returns are impossible for an unlevered spot basket capped at ~10%/name (≤100% invested). Investigate the sell-then-deferred-buy settlement path and _total_value(): likely double-counting PnL across the close/reopen or a settled-USDT miscalculation (same family as the +707% month in the SPOT dual-momentum sibling). After both fixes, re-run over the full 2017-2026 window; only a clean run that actually trades all regimes can show whether breadth-scaling delivers the promised early de-risking and a controlled (<30%) drawdown. If, once it trades the full window, the edge is marginal or the breadth signal still eats a >50% drawdown, abandon.

This strategy was explicitly engineered against the session's two dominant killers — liquidation and >50% drawdown — by trading unlevered spot and letting market breadth automatically scale the book toward stablecoin as the market weakens, a genuinely novel cross-sectional signal. The no-leverage design goal held, but the backtest was non-functional: a BINANCE_SPOT data-coverage wall fed only ~379 bars, freezing all trading into a single 2020-2021 window with nothing thereafter, while a rebalance/settlement accounting artifact produced impossible returns (a +337.8% day for an unlevered basket) that made every metric uninterpretable. Two targeted iterations changed nothing, so the analyst abandoned it — noting the concept is worth revisiting only once full spot history is backfilled for all 10 majors and the PnL path is fixed, since until then it simply cannot be assessed.

The breadth-scaled unlevered-spot concept is sound and its one design goal (no liquidation) held, but it cannot be evaluated without full BINANCE_SPOT daily history for all 10 majors backfilled into the catalog and a corrected rebalance/settlement PnL path — a data-coverage wall and an accounting artifact, not the signal, blocked assessment.

It was abandoned at the pre-optimization backtest-review gate after two iterations left both defects unresolved: the binding root cause was a BINANCE_SPOT data-coverage wall (the primary feed supplied only ~379 bars, so the strategy had no data to trade past 2021 and stayed frozen), and the rebalance/settlement PnL path still produced physically impossible returns — leaving no path to a clean, evaluable backtest within the review loop.

A long-only, unlevered spot basket on BINANCE_SPOT over 10 equal-weight deep majors whose invested fraction equals market breadth — the share of names trading above their own 100-day SMA — so exposure scales automatically into stablecoin as breadth deteriorates, using a cross-sectional signal no deployed basket uses and no leverage so liquidation is structurally impossible.

The backtest could not be validly assessed: trading was frozen to a single 2020-10→2021-10 window with zero activity 2022-2026 (28 trades total), and the metrics were corrupted by an accounting artifact — impossible +337.8% single-day and +214%/+337% monthly returns for an unlevered ≤100%-invested spot book — producing garbage figures (profit factor 33.5, calmar 686, avg_trade_return_pct 396%, +1971% total) and a 60.3% max drawdown in that corrupted window.