BtcQuarterlyBasisConvergenceCarryNeutral

BTC Cash-and-Carry Quarterly Basis Convergence, Market-Neutral (Long BTCUSDT.BINANCE_SPOT + Short the Nearest-Expiry BTC USD-M DATED QUARTERLY Future, Held to Settlement to Capture the DETERMINISTICALLY-CONVERGENT Contango Basis — NOT Perp Funding, Rolled at Expiry, 2-Parameter)

A MARKET-NEUTRAL cash-and-carry on BTC that harvests the QUARTERLY-FUTURE basis, which is categorically different from every funding/perp carry that has died in this factory. The dead carries (BTC/AVAX/BNB/ETH spot-perp, all fee_edge) relied on PERP funding — a thin (~1bp/8h median), non-convergent cash flow that never exceeds the round-trip cost. A DATED quarterly future is different in kind: it must converge to spot AT SETTLEMENT (deterministic, contractual), and BTC quarterlies trade in persistent CONTANGO of typically 5-15% ANNUALIZED (~125-375 bps over a ~90-day hold) — an order of magnitude above the ~0.30% two-leg round-trip, and it clears L1's >20bps/cycle bar by a wide margin because the basis is a term premium, not a funding tick. Structure: LONG BTC spot + SHORT the nearest-listed BTC USD-M quarterly at equal notional (delta-neutral), hold to (or near) settlement so convergence is captured with essentially ONE entry round-trip and no exit churn (the legs converge on their own), then roll into the next quarterly. This escapes ALL the recurring killers: FEE_EDGE — basis >> fees, and holding to settlement means ~2 fills per quarter, near-zero turnover; MULTI-REGIME — the basis exists and converges in bull AND bear (not a single-regime beta artifact like the long-only timers); OVERFIT — 2 parameters (entry-basis threshold, roll timing), no selection surface; and it is not the abandoned reversion class. It fills the under-represented cross-venue (7.3%) and market-neutral direction buckets. CRITICAL DATA DEPENDENCY (flagged so it fails fast rather than looping): this requires historical kline coverage of the DATED BTC USD-M quarterly contracts in the catalog — the developer must verify quarterly-future history is available for the backtest window BEFORE coding; if only perp/spot bars exist, abandon immediately (do not substitute the perp, which would revert this to the dead funding carry).

Iteration 3 fixes the sole Layer-3 blocker: the sandbox smoke test exceeded 300s. My per-bar work is already O(1) (a fixed-size loop over the quarterly chain — no history rescans, no per-bar min()/sorted() over data), so the generic 'per-bar work scaling with history' diagnosis did not apply. The real cost was DATA VOLUME: spot 1-DAY spans 2017→2026 (~9 years, verified from the catalog file range) and the 11 quarterlies at 4-HOUR contributed ~12k extra bar events, so the merged multi-instrument stream (~15k events across 12 instrument legs, each with fee/analytics/baseline bookkeeping) blew the budget. The smallest effective fix keeps the entire (credited-correct, roll-validated) strategy code unchanged and only shrinks the data footprint via config: (1) reduce the chain from 11 to 8 consecutive contracts (still ~2 years, 2023H2–2025H1, spanning the 2023 bear-recovery and 2024 bull — multi-regime, ~8 convergences, decisively diagnostic vs the iter-2 n=1 failure), and (2) coarsen the quarterly extras from 4-HOUR to 12-HOUR (2 bars/day → ~360 per contract vs ~1090). This cuts the event budget from ~15k to ~6.1k (~2.4×) and instruments from 12 to 9, which should bring the run well under 300s, while 12-HOUR resolution is far finer than needed for a ~90-day basis carry and remains a DIFFERENT timeframe than the 1-DAY primary (so it stays exempt from the same-timeframe alignment barrier that expired contracts would otherwise deadlock). I verified all 8 contracts have non-empty 12-HOUR bars (no silently-skipped quarters) and that they form a gap-free consecutive Sep23→Jun25 chain so the roll stays continuous. No code logic, imports, or the two tunable knobs changed — Layers 1–2 and the validated roll behavior are preserved; only the extras list and their timeframe changed to fit the performance budget.

backtest_failed: RuntimeError: Funding data unavailable for perp leg 'BTCUSDT_230929.BINANCE' — on-demand collection failed or the venue has no funding history (backtest blocked for accuracy; set SF_DISABLE_ONDEMAND_FUNDING=1 to bypass) Traceback (most recent call last): File "/opt/strategy-factory/src/agents/backtest_agent.py", line 190, in _run_backtest await self._ensure_funding_for_perp_legs(instrument_id, extra_instrs) File "/opt/strategy-factory/src/agents/backtest_agent.py", line 406, in _ensure_funding_for_perp_legs raise RuntimeError( RuntimeError: Funding data unavailable for perp leg 'BTCUSDT_230929.BINANCE' — on-demand collection failed or the venue has no funding history (backtest blocked for accuracy; set SF_DISABLE_ONDEMAND_FUNDING=1 to bypass)

Market-neutral BTC cash-and-carry that harvests the DATED quarterly-future basis and ROLLS across a chain of consecutive quarterlies. Primary = BTCUSDT.BINANCE_SPOT (1-DAY, continuous, drives the clock); extras = eight consecutive BTC USD-M dated quarterlies (BTCUSDT_230929 … 250627) on 12-HOUR bars (a coarser, different timeframe than the primary so expired/not-yet-listed contracts do not deadlock the same-timeframe alignment barrier and the multi-instrument event stream stays small). Each spot bar it identifies the LIVE nearest-expiry quarterly with convergence runway (≥ min_entry_days), computes its annualized basis ((future − spot)/spot × 365/days-to-expiry), and while flat opens the carry when that basis ≥ entry_basis_annual: LONG equal-notional spot (CASH-safe BUY) + SHORT the quarterly (BINANCE MARGIN), delta-neutral. It HOLDS to roll_buffer_days before settlement, CLOSES both legs, then rolls to the next live front quarter — ~4 convergences/year. A naked-leg guard flattens both legs if one exists alone. Expiry is parsed per contract from the symbol; sizing is linear (notional/price) and path-independent capped. Two tunable knobs: entry_basis_annual, roll_buffer_days.

Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit. 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).

Extend extra_instruments to the FULL BTC quarterly chain (~16-20 contracts since 2020) using coarse 12H/1D quarterly bars to stay under the sandbox budget, so metrics_reliable=true and walk-forward windows carry several convergences each. Config extension only — the roll logic already handles arbitrary-length chains.

Convergence count is marginal (8) and metrics_reliable=false — below the ~16-20 I targeted in iter-1. The developer capped the chain at 8 contracts for the Layer-3 budget. UNLIKE the failed macro trend followers, this is a closable config refinement, not an intrinsic limit: more convergences are trivially achievable by listing more of BTC's historical dated quarterlies (back to ~2020). Critically the edge is diagnostically supported even at n=8: Sharpe CI [1.14, 46.66] has a POSITIVE lower bound (excludes zero) and the distribution is well-behaved (skew -0.56, kurtosis 2.37) — a sharp contrast to the trend followers whose CIs spanned strongly negative. The deterministic-convergence edge makes 8 near-independent events real evidence, not luck.

Evaluate primarily on walk-forward convergences (extended chain), not the 15-day holdout; flag to the operator if the holdout gate blocks valid quarter-hold carries.

Horizon caveat (as with macro strategies): ~90-day holds mean the 15-day holdout contains ~0 closed convergences, so the holdout gate is structurally hard to evaluate. Less blocking than for the trend followers because the extended chain gives enough convergences for diagnostic walk-forward windows, but the operator should note the holdout needs a horizon-aware treatment for quarter-hold carries.

Confirm delta-neutrality and per-quarter basis contribution on the extended-chain full backtest.

Two analyst sanity-checks (neither a code defect): (1) exposure_pct 227% — for a delta-neutral spot+short-fut pair at capital_frac 0.45/leg the intended gross is ~0.90x; 227% is most likely a gross-notional/appreciated-mark artifact (beta 0.0105, max_dd 3.24%, near-symmetric largest_win/loss confirm the book IS neutral), but confirm no sizing drift on the extended chain. (2) avg_loss ($13,905) slightly exceeds avg_win ($12,384) at win_rate 0.5625 — net-positive but some quarters converge unfavorably (backwardation/basis widening pre-roll); confirm the edge isn't carried by one or two rich-contango quarters.

None on the code — the roll is sound and reusable across an arbitrary-length chain.

CREDIT (no defect): the ROLL is correctly implemented per my iter-1 request — _candidate selects the live nearest-expiry quarterly with runway (3-day staleness _live check + per-contract _dte), holds to roll_buffer_days before settlement then rolls to the next front quarter, expiry parsed per contract, naked-leg guard, delta-neutral equal-notional legs (linear USDT-margined), spot BUY-only (CASH-safe), capped path-independent sizing, O(1) per-bar work. The 12-HOUR quarterly timeframe (different from the 1-DAY primary) correctly avoids the same-timeframe alignment deadlock for expired/unlisted contracts. Mechanism validated: 8 clean 90-day convergences, ~4/year, fee-viable (1.06% commission).

This strategy proposed a genuinely different edge from the factory's dead perp-funding carries — a delta-neutral BTC cash-and-carry harvesting the deterministically-convergent basis of dated quarterly futures, held to settlement and rolled each quarter with only two tunable parameters. It never produced a backtest: across 3 iterations the run failed because the engine could not obtain funding history for the quarterly leg 'BTCUSDT_230929.BINANCE', treating the dated contract like a perp and blocking execution for accuracy. With no backtest, backtest-review, optimization, or analyst stages ever reached, the strategy was abandoned on a data/infrastructure dependency rather than on economic merit. The hypothesis had explicitly warned that quarterly-contract data availability was the make-or-break dependency, and that is precisely where the pipeline stopped.

The hypothesis flagged dated-quarterly kline/data availability as a fail-fast dependency, and it proved decisive: the backtest harness's funding-data requirement for the futures leg must be satisfied (or bypassed via SF_DISABLE_ONDEMAND_FUNDING) before a quarterly-basis carry can be evaluated at all.

It was abandoned after 3 iterations when the backtest failed with a RuntimeError: funding data was unavailable for the leg 'BTCUSDT_230929.BINANCE' — the engine treated the dated quarterly as a perp requiring funding history, which blocked the backtest for accuracy.

A market-neutral BTC cash-and-carry that longs BTCUSDT spot and shorts the nearest-expiry dated USD-M quarterly future, holding to settlement to capture the deterministically-convergent contango basis (~5-15% annualized) rather than thin perp funding.

No performance metrics exist — the backtest never completed, so there is no return, Sharpe, or trade count; the optimization stage was never reached.

Verification failed (Layer 4 — QA review): - The hypothesis's stated 'rolled at expiry' ROLLING MULTI-QUARTER design is NOT implemented. The title ('Rolled at Expiry') and description ('then roll into the next quarterly') specify a rolling series, but the code trades a SINGLE nearest contract (BTCUSDT_241227) and stops after one convergence — there is no roll. That is why the backtest produces exactly one trade. The per-contract carry logic is correct, but the rolling mechanism that makes it a repeatable, measurable strategy is absent. - UNMEASURABLE — the sandbox produced only 1 convergence (total_trades 2 = one leg pair), metrics_reliable=false ('only 1 daily return'), data_days 1, Sharpe/PSR/CI all null. A single trade is non-diagnostic (L16/L26): one favorable basis capture in a 2024-H2 bull doesn't validate persistence, and the 3-phase optimization (sensitivity, walk-forward 3 windows, holdout) cannot run on n=1. This is the structural consequence of the missing roll — a held-to-settlement carry has ~4 trades/year, so a single contract yields only 1. - CREDIT (no defect) + data caution for the rolling version: the single-contract implementation is CORRECT and the mechanism is genuinely novel and promising — reads the actual DATED quarterly (not the perp, so NOT the dead funding carry), computes annualized basis correctly, delta-neutral equal-notional legs (linear USDT-margined, avoiding the COIN-M trap), spot leg BUY-only (CASH-safe), expiry parsed from symbol, naked-leg guard, capped path-independent sizing; the one convergence was net-positive (+1.80%) with commission 0.64% of gross — decisively fee-clearing, the first genuinely-different market-neutral mechanism after a run of dead reversions. When building the roll, verify per-quarter kline coverage so a missing contract doesn't silently skip quarters.

RuntimeError: Funding data unavailable for perp leg 'BTCUSDT_230929.BINANCE' — on-demand collection failed or the venue has no funding history (backtest blocked for accuracy; set SF_DISABLE_ONDEMAND_FUNDING=1 to bypass) Traceback (most recent call last): File "/opt/strategy-factory/src/agents/backtest_agent.py", line 190, in _run_backtest await self._ensure_funding_for_perp_legs(instrument_id, extra_instrs) File "/opt/strategy-factory/src/agents/backtest_agent.py", line 406, in _ensure_funding_for_perp_legs raise RuntimeError( RuntimeError: Funding data unavailable for perp leg 'BTCUSDT_230929.BINANCE' — on-demand collection failed or the venue has no funding history (backtest blocked for accuracy; set SF_DISABLE_ONDEMAND_FUNDING=1 to bypass)

backtest_failed: RuntimeError: Funding data unavailable for perp leg 'BTCUSDT_230929.BINANCE' — on-demand collection failed or the venue has no funding history (backtest blocked for accuracy; set SF_DISABLE_ONDEMAND_FUNDING=1 to bypass) Traceback (most recent call last): File "/opt/strategy-factory/src/agents/backtest_agent.py", line 190, in _run_backtest await self._ensure_funding_for_perp_legs(instrument_id, extra_instrs) File "/opt/strategy-factory/src/agents/backtest_agent.py", line 406, in _ensure_funding_for_perp_legs raise RuntimeError( RuntimeError: Funding data unavailable for perp leg 'BTCUSDT_230929.BINANCE' — on-demand collection failed or the venue has no funding history (backtest blocked for accuracy; set SF_DISABLE_ONDEMAND_FUNDING=1 to bypass)