Status: Draft — v0.1.0-draft. The specification is not yet stable and has not yet been adopted outside its reference implementation. Field names, envelope layout, and MUST-level requirements MAY change until v1.0.0.

The Open Backtest Format (OBTF) is a versioned, engine-agnostic, studies-first on-disk format for storing the full evidence of a quantitative trading strategy's backtest results — metrics, trades, orders, portfolio snapshots, universes, cost assumptions, and Monte-Carlo significance tests — in a single self-contained file.

One .obtf file per algorithm. Portable across environments, engines (vector / event-driven / hybrid), regimes, and toolchains.

Every quantitative research team ends up inventing the same thing: a directory or pickle or ad-hoc JSON dump that holds "the results of this backtest." These artefacts are not portable between engines, not comparable across teams, and not readable a year later without the exact code that produced them.

OBTF is a boring, versioned, spec'd container for those results, so that:

• Analysis tools (dashboards, notebooks, ranking pipelines, LLM agents) can consume outputs from any engine without per-engine adapters.
• Research teams can archive and share bundles across environments and framework versions.
• Engine authors (event-driven, vector, walk-forward, cross-sectional) can standardise on a common on-disk contract without giving up their internal execution model.

• OBTF is not a live-trading log format.
• OBTF is not a tick, bar, or market-data store — it holds the outputs of a backtest, not its inputs. Data-source references are permitted; raw market data is not.
• OBTF is not a strategy source-code exchange format.
• OBTF is not a query API or a data lake — it is a single-file envelope.
• OBTF does not standardise how metrics are computed. Sharpe ratio, drawdown, and friends are producer-defined; the format specifies only how their values are stored.

The specification lives under spec/. Read it in order:

• 0-overview.md — goals, non-goals, terminology, RFC-2119 language
• 1-envelope.md — magic bytes, framing, compression
• 2-data-model.md — Bundle → Study → Run → Metrics
• 3-metrics.md — per-run and summary metric catalog
• 4-blobs.md — Parquet blob extraction & the @blob reference protocol
• 5-universe-windows.md — universes, date ranges, sample_type vocabulary
• 6-execution-config.md — cost / slippage / commission / fill model shape
• 7-reader-contract.md — MUST/SHOULD for readers
• 8-writer-contract.md — MUST/SHOULD for writers
• 9-versioning.md — semver policy for the format itself
• 10-extensions.md — vendor extension namespace (x-*)

The reference Python implementation currently lives inside coding-kitties/investing-algorithm-framework. It will be extracted into a standalone quantos-bundle-py package as OBTF stabilises. Conformance test vectors will live under conformance/ in this repo (TBD).

The spec is in early draft. Small edits, typo fixes, and clarifying questions are welcome as pull requests or issues. Substantive changes should go through the change-proposal process — see GOVERNANCE.md.

OBTF is currently single-vendor-authored (QuantOS / Quant-Commons) and will transition to a multi-vendor governance structure as external implementations emerge. See GOVERNANCE.md.

Apache License, Version 2.0. See LICENSE.