[Feature] UK equities (LSE .L) market support in market data + backtest routing

[cgycorey]

Feature request: UK market support

The loader layer and MCP get_market_data support US, HK, India, Korea, Canada, Vietnam, and crypto — but not UK equities. LSE (.L) symbols currently resolve to _unresolved.

Current failure (the motivation)

// get_market_data({codes: ["VOD.L"], ...}) with source=auto
{ "_unresolved": ["VOD.L"] }

Root cause: _SOURCE_PATTERNS has no .L rule, so UK symbols fall through to the tushare default. _detect_market also defaults them to a_share. The symbol is then routed down the China fallback chain (tencent → mootdx → eastmoney → baostock → akshare → tushare → local), which cannot serve LSE — several seconds of network attempts later it surfaces _unresolved, indistinguishable from a genuinely dead ticker.

Forcing source=yahoo works today:

VOD.L  → OK 15 bars (close 116.60, 2026-08-04)
SHEL.L → OK 15 bars
HSBA.L → OK 15 bars

Proposal (working fix, locally verified)

Treat UK as a first-class market with the same parity as Canada/US:

1. src/market_data.py — _SOURCE_PATTERNS: add ^[A-Z0-9&.\-]+\.L$ → yahoo (mirrors Canada .TO/.V); yahoo_loader _is_supported gate: add .L/.IL

2. backtest/engines/_market_hooks.py — _MARKET_PATTERNS: add .L → uk_equity; _MARKET_CURRENCY: uk_equity: GBP (currently returns UNKNOWN:uk_equity)

3. backtest/loaders/registry.py — FALLBACK_CHAINS: add "uk_equity": ["yahoo", "yfinance", "local"] (mirrors ca_equity)

4. backtest/runner.py — _MARKET_TO_SOURCE: add uk_equity: yahoo (currently backtests silently fall to the tushare engine); composite dispatch → GlobalEquityEngine(market="uk")

5. Price units — Yahoo serves LSE names in pence (GBp): VOD.L close is ~117p, HSBA.L ~1500p; without normalization the engine prices a £1.17 share at £117 (100× notional error on positions, and on SDRT). Normalize OHLC ÷100 at the loader boundary (Yahoo chart-meta currency == "GBp"; yfinance fallback suffix rule), declare currency_conversion in provenance, bump loader cache version so pre-normalization parquet is never served

6. SDRT cost model — 0.5% Stamp Duty Reserve Tax on purchases, penny rounding with exact-½p rounds-up (HMRC FA86/S99(13), STSM014030)

   Exemptions scope (deliberate): we model the statutory Main Market rate only. AIM shares, UCITS ETFs, gilts and new issues are not auto-detected — the engine has no symbol-aware asset-class classification (per-symbol tax databases drift; a wrong exemption is worse than the flat rate). Caller's control surface: uk_stamp_tax: 0.0 exempts a whole run (exempt-only baskets). The known boundary: a mixed Main-Market + exempt basket in one run pays statutory on everything. Possible follow-up (not built): per-symbol exempt config (uk_stamp_tax_exempt: [...]), which requires widening calc_commission to carry the symbol; AIM auto-detection via Yahoo exchange metadata is feasible, UCITS/new-issue auto-detection is not recommended.

7. Journal/shadow consumers — _infer_market_from_symbol: .L → "uk" (journal trades currently land in "other" and shadow analysis silently degrades to NaN price features); shadow _MARKET_KEY_MAP/labels: uk → uk_equity, "UK equity"

8. Financial statements — `get_financial_statements`: `.L` → Yahoo quoteSummary (income/balance/cashflow annual histories + indicators); `get_fundamentals` stays US-only by contract (SEC XBRL PIT panels — no honest UK source for filed-date PIT), fails loud for UK

9. Tests: source/market/submarket/currency; yahoo + yfinance GBp scaling; market-data provenance; engine economics incl. HMRC rounding cases; journal inference; registry chain contract

Question to maintainers

Is UK support in scope for the project? Happy to adjust scope per your call. Remaining candidates for follow-ups: London market-open spec in MARKET_SPECS (missing market fails loud today, never silent) and symbol-search UK suffix hints.

[cgycorey]

Open scope question — SDRT exemptions:

The implementation applies the statutory 0.5% Main Market rate to every UK buy, with exemptions (AIM shares, UCITS ETFs, gilts, new issues) deliberately not auto-detected — per-symbol tax classification would hard-code a drifting database, and a wrong exemption is worse than a flat rate. Caller's only knob today is uk_stamp_tax: 0.0 (whole-run).

Which direction do you prefer?

1. Caller-declared (current) — no per-symbol logic; exempt-only baskets set uk_stamp_tax: 0. Mixed Main-Market + exempt baskets overpay on exempt names.
2. Per-symbol exempt config — uk_stamp_tax_exempt: ["SYMBOL.L", ...]; requires widening calc_commission to carry the symbol (touches base + all engine overrides).
3. AIM auto-detection — classify via Yahoo exchange metadata (fullExchangeName), caller-declared for UCITS/new issues (auto-detecting those isn't reliable).