fix(regime): decouple ratio gate from proxy history

[brndnmtthws]

Add a regression test for an uninvested non-anchor symbol in the ratio basket, fetch aligned closes explicitly for the ratio gate, and widen create_limit_order kwargs typing so the ty pre-commit hook passes.

[Copilot]

Pull request overview

This PR adjusts regime-aware rebalancing to avoid coupling the ratio-gate computation to the proxy-series history, adds a regression test for an “uninvested rest symbol” scenario, and fixes a typing issue in order creation so static checking passes.

Changes:

• Update regime proxy-series helper to return only dates/series, and fetch aligned closes explicitly for ratio-gate evaluation.
• Add an async regression test covering a ratio basket that includes an uninvested non-anchor symbol.
• Widen create_limit_order kwargs typing to satisfy ty/pre-commit checks.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
thetagang/trading_operations.py	Widens kwargs typing in create_limit_order to satisfy static typing.
thetagang/strategies/regime_engine.py	Refactors proxy-series return shape and adds a separate aligned-close fetch for ratio gating.
thetagang/portfolio_manager.py	Updates proxy-series wrapper return type to match the regime engine change.
tests/test_regime_rebalance.py	Adds a regression test scenario for ratio gate with an uninvested “rest” symbol.

Comments suppressed due to low confidence (1)

thetagang/strategies/regime_engine.py:96

• In _get_regime_proxy_series, proxy_symbols is derived from weights_override, but _get_regime_aligned_closes is still called with the full symbols list. This means the proxy-series calculation can fail (or do unnecessary IBKR history requests) due to a non-proxy symbol having missing/invalid history, even though that symbol isn't used in the weighted series. Consider fetching aligned closes for proxy_symbols here (and only requiring history/validation for the symbols actually used in the proxy series).

        proxy_symbols = list(weights_override.keys()) if weights_override else symbols
        sorted_dates, aligned_closes = await self._get_regime_aligned_closes(
            symbols,
            lookback_days,
            cooldown_days,

[Copilot]

aligned_closes for the ratio gate is fetched via a second _get_regime_aligned_closes() call, but the returned dates are discarded and the code indexes aligned_closes[*][idx] using idx from dates returned by _get_regime_proxy_series(). Since these are separate IBKR history requests, there’s no guarantee the two date sets/lengths match, which can lead to IndexError or misaligned ratio calculations. Use the dates returned with aligned_closes (or assert they match dates and fail deterministically) before indexing.

[Copilot]

Enabling ratio_gate now triggers an additional full historical-data fetch for all symbols (on top of the proxy-series history fetch). This increases runtime and the number of IBKR historical requests, which can hit pacing limits. Consider fetching aligned closes once and reusing them for both the proxy-series and ratio-gate computations (or otherwise sharing the fetched histories).

Suggested change

	_, aligned_closes = await self._get_regime_aligned_closes(
	symbols,
	regime_rebalance.lookback_days,
	regime_rebalance.cooldown_days,
	)
	if "aligned_closes" not in locals():
	dates, aligned_closes = await self._get_regime_aligned_closes(
	symbols,
	regime_rebalance.lookback_days,
	regime_rebalance.cooldown_days,
	)

[Copilot]

This regression test only asserts isinstance(orders, list), which will pass even if the ratio-gate logic still regresses (e.g., incorrect orders emitted, or the ratio gate being bypassed). Since this PR changes regime/ratio gating behavior, strengthen the assertions to validate the expected order list (symbols + share deltas) for this scenario, so it actually detects future trading-behavior regressions.