Skip to content

PRODUCT_SCOPE.md

Latest commit

 

History

History
103 lines (79 loc) · 4.82 KB

PRODUCT_SCOPE.md

File metadata and controls

103 lines (79 loc) · 4.82 KB
Raw

Product Scope

Thesis

Build a product-shaped runtime for card-based Taiwan stock intraday strategy operations.

The runtime should let researchers and agents author cards, organize decks, validate manifests, and run replay/live-sim workflows through a stable CLI, while keeping live execution authority and risk governance under explicit operator control.

Product promise

  1. One shared auth/session and market-data connection model rather than one login/feed stack per strategy card.
  2. Many strategy cards per engine/deck, each with explicit identity, symbol pool, parameters, and auditability.
  3. Intent-first design: cards express intent and policy hints; the runtime governs whether those intents become broker actions.
  4. Low-latency by design for day trading, especially around stop and forced-exit paths.
  5. Replay-first productization so contracts harden before live control expands.
  6. Adapter-swappable I/O so the runtime is not permanently fused to one broker or market-data API.
  7. Recordable and replayable market data, with both replay sim and live sim as first-class execution modes.

Primary users

1. Operator

Owns live permissions, session posture, risk limits, runtime health, active account selection, and deployment posture.

2. Researcher / strategist

Creates and iterates cards, variants, thresholds, symbol pools, and deck composition.

3. Agent-assisted workflow

Helps draft cards, validate configs, prepare replay jobs, and assist configuration setup. Agent assistance does not imply unrestricted live trading authority.

v0.1 scope

  • Taiwan stock intraday only
  • Shared auth/session module for marketdata + trading/account surfaces where supported
  • Shared market-data hub with subscription planning
  • Single runtime instance / single deck at a time
  • Multiple cards and card variants inside the deck
  • Card-level symbol pools contributed into connection management when enabled
  • Card-level entry, exit, capital control, and required feature declarations
  • Separate feature/synthesizer pipeline for time-series and indicators
  • CLI-managed card/deck/global config and auth profiles
  • Replay runner and live-sim path as first-class features
  • Read-only browser Mission Control observability over replay/live-sim artifacts
  • Audit-friendly event, intent, risk, and execution receipts
  • Replaceable MarketDataAdapter and BrokerAdapter interfaces
  • Day-trading guardrails: emergency stop-loss, forced exit, final-auction flatten controls

Non-goals for v0.1

  • Futures/options support
  • Smart order routing across multiple brokers
  • Fully dynamic plugin loading without operator controls
  • Autonomous live trading authority for arbitrary agents
  • Portfolio management across multiple independent engines
  • Claiming production-grade reliability before replay/live-sim contracts are proven
  • Hiding broker/session complexity behind fake abstractions

Product boundaries

Included

  • Contracts
  • CLI surface
  • Core runtime decomposition
  • Auth/session model
  • Day-trading risk model
  • Adapter interfaces
  • Migration plan from the current engine

Deliberately deferred

  • Deep strategy library
  • Broker-connected or operator-authoritative GUI/control plane
  • Hosted multi-tenant service
  • Real credential storage implementation in this repo
  • Multi-market abstraction beyond Taiwan cash

Safety and authority stance

The product is intentionally agent-assisted, operator-governed.

That includes a practical safety boundary for some broker ecosystems:

  • support login mode with account + API key + cert + cert password
  • allow operators to provision API keys without trade permission when appropriate
  • let agents help with config/replay/validation inside that boundary
  • require stronger operator control before expanding into live trading authority

Success criteria for the seed repo

  • A new contributor can understand the product shape in under 15 minutes.
  • The repo makes the difference between card authoring, feature synthesis, risk governance, and execution obvious.
  • The CLI vocabulary is stable enough to guide future implementation.
  • Migration from the current sdk_manager_async.py + strategy_async.py world is phased, not hand-wavy.
  • The docs explain why latency matters for day-trading stops and forced exits.
  • Nothing in the repo over-claims live readiness.

Risks to manage early

  • Over-fitting the product model to one current strategy implementation
  • Letting cards bypass risk and execution governance
  • Recomputing indicators independently inside many cards and losing replay/live parity
  • Conflating replay behavior, live-sim behavior, and live behavior
  • Locking adapter contracts too tightly to one broker SDK
  • Failing to filter mixed order lifecycle events by active account and user_def
  • Underestimating latency-sensitive exit behavior in short-side day trading