fn_registry/dev/issues/0088e-trading-strategy-capability-group.md

id, title, status, type, domain, scope, priority, depends, blocks, related, created, updated, tags

id	title	status	type	domain	scope	priority	depends	blocks	related	created	updated	tags
0088e	Trading: capability group `strategy` + Strategy contract	pendiente	feature	trading meta	multi-app	alta				2026-05-17	2026-05-17

0088e — Trading: capability group strategy + Strategy contract

Status: pendiente Created: 2026-05-14 Type: feature Parent: 0088 Depends: 0088a Blocks: 0088g, 0088h

Problema

No hay contrato de estrategia. Sin contrato cada estrategia se cablea distinto y backtester/live_runner no pueden tratarlas uniformemente. El contrato debe ser puro: entrada = market state + portfolio state + params; salida = lista de OrderIntent (no efectos).

Piezas

1. Tipos del registry en python/types/finance/:
   • MarketSnapshot (OHLCV ventana + last_price + ts).
   • PortfolioSnapshot (balance + positions).
   • StrategyParams (dict tipado por estrategia, validado via JSON Schema).
   • OrderIntent (lado, qty, instrument, kind=market/limit, price_limit, ttl, reason).
2. Contrato Strategy (Python protocol):

   def decide(market: MarketSnapshot, portfolio: PortfolioSnapshot, params: StrategyParams) -> list[OrderIntent]
   

   Funcion pura. Sin I/O, sin random sin semilla, sin tiempo system.
3. Dos estrategias de referencia (puras):
   • strategy_sma_cross_py_finance (SMA rapida/lenta cross → entry/exit).
   • strategy_rsi_meanrev_py_finance (RSI < 30 → long, > 70 → short con caps).
4. Funcion validate_strategy_output_py_finance(intents, portfolio, params) que aplica sanity (qty>0, instrumento valido, no oversize) antes de pasar a risk.
5. Tag strategy. Pagina madre docs/capabilities/strategy.md con ejemplo canonico end-to-end (snapshot fake → decide → list[intent]).
6. Tests deterministicos: dado snapshot fijado, decide retorna lo esperado.

Aceptacion

• mcp__registry__fn_search query="" tag="strategy" devuelve interface + >= 2 estrategias.
• Tests pasan en go test/pytest.
• Estrategias son puras: purity: pure en frontmatter; sin imports impuros.

No-objetivos

• Estrategias optimizadas, backtest report (sale a 0088g), parametrizacion via UI (analysis/strategy_lab futuro).