Common client boundary
Keep market-data providers, the local PaperBroker, and any eventual broker connection behind explicit adapter boundaries.
LOADING...
v0.6.0Paper firstHuman approved
ForgeCommand subsystem
Stock Trading Automation
A paper-first market-analysis and decision-support project that keeps human approval between generated signals and consequential actions.
The work is evolving as a Trading module inside ForgeCommand, bringing market context, risk controls, proposals, and reviewable decision records into one operational interface.
PythonPyQt6SQLiteTwelve Data
ForgeCommand v0.6.0 · paper and learning cockpit
Read-only Twelve Data market context, local practice accounting, deterministic order controls, and company research share one operational surface.
The engineering question
How can market analysis be useful without allowing uncertainty or automation to bypass control?
Trading systems combine noisy data, changing conditions, execution assumptions, and strong emotional incentives. A plausible-looking signal is not evidence that a strategy works, and a generated recommendation should never be treated as authority.
This project treats analysis, risk policy, approval, execution, and audit as separate responsibilities. The working rule is persistent paper evidence first: observe, measure, and understand the system before considering any real-money boundary.
Architecture change
From isolated script to governed subsystem
The original standalone plan is now part of ForgeCommand, providing a wider control surface for status, configuration, approvals, audit history, and AI-assisted explanation.
// v0.6.0 decision boundary
Cached market data + persistent paper state
↓ analysis + explainable signal
Trade intent
↓ deterministic risk policy + human confirmation
Local PaperBroker fill + correlated audit record
Current direction
Build the boundaries before expanding the automation
Explainable signals
Record the inputs, indicator state, and reasoning behind each proposal so a result can be reviewed later.
Approval before action
Separate generated analysis from the deterministic controls that authorise, reject, or suspend an action.
Measurable performance
Track P&L, drawdown, hit rate, expectancy, slippage, latency, and missed fills instead of relying on anecdotes.
ForgeCommand v0.6.0 · 22 August 2026
From paper preview to a persistent learning cockpit
The current milestone joins market observation, practice accounting, company research, and deterministic order handling without granting an AI—or the interface itself—live brokerage authority.
Persistent accounting
SQLite now records practice accounts, cash, intents, risk decisions, orders, fills, positions, and realised or unrealised P&L using decimal-safe values.
Quota-aware market board
Twelve Data ingestion uses a durable cache, freshness indicators, request-budget awareness, and a previous-day share board instead of invented active prices.
Company Learning Hub
Cached market facts, current paper holdings, persistent research notes, and prepared AI questions now share a dedicated learning workspace.
Safer local operation
Secrets can be imported outside the repository; missing AI or market services remain visible failure states rather than silent fallbacks.
v0.6.0 removes legacy mock accounts, positions, orders, and generated prices from the active workflow. The practice ledger remains simulated, no live broker is connected, and nothing on this page represents demonstrated investment performance.
Human confirmation boundary
A proposal cannot silently become a paper fill
The order preview names the instrument, direction, quantity, reference price, estimated notional, account environment, and exact ledger affected. The operator can still decline it. Existing holdings and filled practice orders remain visible behind the confirmation, keeping the decision in portfolio context.
Learning before automation
Research stays attached to the market and the paper portfolio
The company workspace combines cached previous-day facts, whether the company is currently held, durable operator notes, and structured prompts such as a balanced bull-and-bear case. Preparing a question does not submit a trade or grant the AI execution authority.
Risk and controls
Safety conditions before scope
- ▸Paper or practice funds while strategies and controls are evaluated.
- ▸Per-strategy loss limits, daily loss limits, and drawdown halts.
- ▸One-click suspension and automatic circuit breakers on policy breach.
- ▸Audit records linking signals, proposals, approvals, simulated fills, and outcomes.
- ▸Pinned code and data versions for repeatable backtests.
Technical foundation
Current module stack
- ▸Python and PyQt6 for the Windows-native ForgeCommand interface.
- ▸Provider-neutral domain models and broker-adapter boundaries for explicit paper and live environments.
- ▸Quota-aware Twelve Data ingestion and PaperBroker persistence in separate tables within a local SQLite database.
- ▸Deterministic risk checks, preview-and-confirm orders, conservative fills, restart-safe holdings, and correlated audit events.
- ▸A company research workspace with optional AI context that remains separate from execution authority.
Validation
Evidence the project still needs to produce
Expectancy
Net outcome per trade and per day, including the full win/loss distribution.
Drawdown
Maximum drawdown measured against the risk envelope and suspension thresholds.
Execution realism
Fees, slippage, latency, partial fills, and missed fills in simulation.
Regime stability
Walk-forward performance across changing volatility and market conditions.
Roadmap
Progressive proof, not a rush to execution
Data foundation
Provider-neutral domain, market cache, and decimal-safe paper ledger established in v0.6.0.
Paper operations
Build sustained practice history, reports, alerts, and failure handling from the persistent ledger.
Strategy baselines
Compare simple momentum and mean-reversion references with repeatable metrics.
Control validation
Exercise limits, approvals, suspension behaviour, and audit reconstruction.
Evidence review
Consider a tightly bounded live-cap experiment only after sustained paper evidence.
Current status: v0.6.0 paper and learning foundation
The persistent PaperBroker, cached market board, deterministic previews, P&L projection, and company learning workspace are implemented. Sustained paper-operation evidence, repeatable strategy results, wider control validation, and any case for live execution remain future gates.