QST

Archived agent-ready research prototype for typed quantitative strategy records.

QST turns strategy intent into deterministic Graph Kernel Records (GKR): typed .gkr.yaml files that can be validated, canonicalized, hashed, reviewed, and handed off to agents or downstream systems.

It is not a trading bot, broker adapter, exchange adapter, backtester, optimizer runtime, or production execution engine.

Status

This repository is closed as an archived agent-ready research prototype.

The final tree includes:

• typed qst-ir/0.4 strategy records
• deterministic qst-canonical/0.4 canonical JSON
• graph, parameter, and instance hashes
• token surface governance and conformance tests
• Coverage Frontier v0.3 evidence and publication gate
• active agent prompt pack and handoff docs
• a Qlib partial workflow adapter proof

Install

pip install -e ".[dev]"

Quick Check

python -m qst.cli vocabulary --check
python -m qst.cli validate examples/strategies/kdj_cross_basic.gkr.yaml
python -m qst.cli hash examples/strategies/kdj_cross_basic.gkr.yaml
python -m qst.cli canonicalize examples/strategies/kdj_cross_basic.gkr.yaml --output .local_audit/kdj.canonical.json

Qlib Adapter Proof

QST includes a partial Qlib workflow YAML importer:

python -m qst.cli adapter qlib import examples/adapters/qlib/workflow_config_lightgbm_alpha158.yaml --output .local_audit/qlib_lightgbm_alpha158.gkr.yaml --coverage .local_audit/qlib_lightgbm_alpha158.coverage.json
python -m qst.cli validate .local_audit/qlib_lightgbm_alpha158.gkr.yaml
python -m qst.cli hash .local_audit/qlib_lightgbm_alpha158.gkr.yaml
python -m qst.cli canonicalize .local_audit/qlib_lightgbm_alpha158.gkr.yaml --output .local_audit/qlib_lightgbm_alpha158.canonical.json

The adapter is record-layer evidence only. It does not import Qlib, run qrun, train models, run inference, execute backtests, connect to brokers or exchanges, or claim lossless Qlib conversion.

What It Provides

• Record layer: stable .gkr.yaml strategy records with typed nodes, token refs, capabilities, and port signatures.
• Identity layer: canonical JSON plus deterministic graph, parameter, and instance hashes.
• Token layer: accepted, experimental, reserved, and custom-token surfaces with maturity and execution-support metadata.
• Evidence layer: public demos, reference diagnostics, hash sentinels, coverage matrix, coverage report, and dogfood cases.
• Agent layer: active prompt pack, takeover prompt, playbook, usage guide, and secondary development guides.

Boundaries

QST deliberately does not provide:

• live trading
• broker or exchange integration
• full backtesting
• Qlib runtime replacement
• qrun execution
• model training or inference execution
• lossless Qlib conversion
• arbitrary Python strategy parsing
• production portfolio optimization
• profitability claims

Custom-token execution is explicit and approval-bound. Verification does not import or execute user code; execution requires integrity verification, local approval, an execution grant, and output validation.

Handoff

Start here:

• Final Handoff
• Final Scope
• Final Acceptance
• Agent Takeover Prompt
• Agent Playbook
• Usage Guide

Key Paths

qst/                         Python package and CLI
examples/strategies/         12 public GKR strategy examples
examples/adapters/qlib/      Qlib partial workflow adapter examples
examples/custom_token/       Custom-token reference example
tests/reference/             Deterministic reference fixtures and traces
tests/adapters/qlib/         Qlib adapter proof tests
docs/agent/                  Agent handoff and prompt guidance
docs/adapters/               Adapter boundary and Qlib adapter guide
docs/reports/                Coverage Frontier and acceptance reports

Documentation

• Architecture
• Security
• Reference
• Token Family Registry
• Token Coverage
• Coverage Report
• Qlib Adapter Boundary
• Agent Guidance

Project Identity

• Python import package: qst
• CLI: qst
• Distribution name: quant-strategy-tokenizer
• Editable strategy source: .gkr.yaml
• Packaged record suffix: .gkr
• IR: qst-ir/0.4
• Canonical schema: qst-canonical/0.4