QuantForge

01 — The Research Problem

The Research Problem

A chart of a backtest is not evidence unless the run can be replayed, the timing can be audited, and the ledger does not drift.

• A signal

  against Information it was not allowed to see

• One interesting run

  against A run someone else can rebuild

• Floating-point cash

  against A ledger that stays on its quantum

• A summary metric

  against The event that produced it

02 — Architecture

Architecture

The terminal never talks to an exchange. It asks a same-origin proxy for an allow-listed route, and FastAPI runs the engine.

1. Next.js terminal
2. ↓→
   Allow-listed API proxy
3. ↓→
   FastAPI
4. ↓→
   Backtest engine
5. ↓→
   Analytics

• GET/api/healthService status
• GET/api/configBounds, scenarios, strategies
• POST/api/backtestOne run, with its event trace
• POST/api/compareStrategies across cost presets
• GET/api/research-blockEmpirical-study gate

03 — Causal Event Loop

Causal Event Loop

Each timestamp is handled in a fixed phase order. A signal cannot trade on the bar that produced it.

1. 01

   Bar open

   Only the open is revealed.

2. 02

   Fill due

   Fill, or expire, at this open.

3. 03

   Bar complete

   The prior bar's OHLCV is released.

4. 04

   Signal

   The strategy decision becomes available.

5. 05

   Submit

   Place, reject, or cancel one order.

6. 06

   Valuation

   Mark equity at the observed price.

• A signal is available one hour after its decision bar opens.
• A submitted order is due one bar later, and fills at that bar's open.
• If that open is missing, the order expires instead of sliding forward.
• Strategies read only observations up to the decision bar.

04 — Deterministic Scenarios

Deterministic Scenarios

The terminal's prices are generated. The same seed rebuilds the same bars, and the run's identity does not depend on the wall clock.

The seed is an integer offset into a fixed wave table. The generator does not draw random numbers.

• Scenario

  Generator, seed, length, name, and clock.

• Run

  Assumptions, signals, and the full bar series.

• Order / fill

  Each order and fill has its own identity.

• Experiment

  Binds the run identities for the terminal session.

Identities are SHA-256 hashes of canonical JSON.

05 — Decimal Ledger

Decimal Ledger

Cash, quantity, fees, and prices are Decimal values. Risk ratios are the exception: volatility, Sharpe, and Sortino are computed in binary floating point and can be left undefined when the hourly grid is incomplete.

• Accounting context uses 50 digits of precision.
• Cash and quantity sit on a 0.00000001 quantum.
• Buy quantity rounds down. Buy debit and fees round up.
• Sell proceeds round down after the fee is rounded up.
• Cash and quantity cannot go negative.
• Marked equity equals cash plus quantity times the mark price.
• The ledger reconciles with fills at the end of the run.

The book is long or flat, and all-in. It does not short and it does not use leverage. The default starting cash is 10,000.

These are metrics the engine can compute. This page does not report a result for any of them.

06 — Research Terminal

Research Terminal

Six workspaces sit on the same run. The screenshots are from the deployed terminal, on a synthetic uptrend scenario. The numbers in them belong to that generated path.

• Market

  Candles, volume, and fill markers.

• Lab

  Strategy, scenario, seed, and cost preset.

• Performance

  Equity against cash and buy-and-hold, plus drawdown.

• Replay

  Step through the engine's own events.

• Compare

  Both strategies across cost presets. Nothing is ranked.

• System

  Architecture, causal timing, and the research gate.