Skip to content

Repository files navigation

quant-proj

quant-proj is the single active source repository for this personal quantitative research workspace. It combines three tracked responsibilities:

  • the deterministic research runtime and formal research primitives;
  • data staging, publication, qualification, and audit tools for the CN and Free-US warehouses;
  • source definitions for the Windows/WSL scheduled production jobs.

Mutable databases, raw downloads, caches, backups, deployed task state, credentials, and provider payloads live outside Git under explicitly configured data roots. A tracked registrar describes intended scheduler state; it does not prove that the currently deployed task or launcher bytes match a Git commit.

Legacy repositories and completed experiments are historical evidence, not runtime dependencies. The complete pre-convergence tree is preserved at the GitHub Release/tag archive/pre-governance-convergence-20260721.

Local setup

python -m venv .venv
.venv/bin/pip install -e '.[dev]'
.venv/bin/pytest
.venv/bin/ruff check .

The default settings are built into the package, so quant info also works from a non-editable wheel without a copied config/ directory. The default project root is the source checkout when present and otherwise the installed package directory; the current working directory is never inferred as a project or data root. Bind mutable data explicitly when needed:

export QUANT_DATA_ROOT=/absolute/path/outside/the/repository
quant info

Select an external TOML file with quant --config PATH ... or QUANT_CONFIG.

Market execution is fail-closed at the experimental mechanism boundary: security-scoped inputs must match their source subjects and immutable market, currency, exchange, and timezone profile; event timing and accounting identities are checked; and settlement/tax rules are selected from the trade date. The formal research path consumes canonical DecisionPacket bytes and does not dynamically load strategy code. Formal consumption also requires a clean Git worktree whose tracked quant_system bytes equal the imported runtime, and it matches the manifest's research-snapshot and universe artifacts to the packet and runtime objects. Its typed execution contract is checked before mutation and fixes the initial portfolio, sizing-price identities, complete execution inputs, any execution-calendar revision, costs, slippage, capacity, position limit, stage plan and anchor cutoff. RunManifest qualification emits only path-free identities for an external claim process. Generic captures bind caller-supplied bytes and metadata only; they do not qualify a provider or authorize a candidate. This is deliberately a small contract rather than a data-governance subsystem.

Broker execution is intentionally outside this research-only project. Production market-data retrieval and publication are active responsibilities of the tracked tools/, scripts/, and ops/ surfaces; their mutable outputs and live owner state remain outside Git.

Free U.S. research warehouse

Optional support utilities under tools/ maintain domain-owned DuckDB shards outside Git. Writers use unpublished build files; research readers use only the historical baseline plus catalog-selected immutable generations. This permits continuous ingestion while a research run binds one reproducible published snapshot. Set QUANT_DATA_ROOT explicitly for Free-US utilities. CN utilities accept explicit data roots; tracked registrar defaults are machine-local deployment inputs and do not prove the roots used by installed tasks. See the source catalog and warehouse operations.

Repository shape

The active branch contains the runtime, generic research primitives, data production and audit tools, tracked scheduler definitions, their tests, and the minimal architecture. Some active paths still require explicit host data roots or deployment reconciliation; Git review must not be used as evidence that those machine-local surfaces have converged. Manager roadmaps, task packets, mechanism registries, terminal experiment reports, and one-off runners remain outside the active tree.

About

Lightweight research-only quantitative system with one Python package, one CLI, deterministic backtesting, and fail-closed market/data semantics.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages