New Feature: Open Backtest Format Support
Finterion is the first platform to support the Open Backtest Format, an open, portable and verifiable standard for algorithmic trading backtest results.
Marc van Duyn
•
2026-02-02
Finterion now officially supports the Open Backtest Format (OBF) in its research environment. Backtests for quantitative trading strategies on the platform can now be organized by studies, universes and backtest windows, instead of being locked away as a single opaque report.
The Open Backtest Format (OBF) is an open, vendor-neutral specification for representing backtest results. Every algorithm on the platform now publishes its backtests in this format, and we believe every platform in this space will eventually do the same.
Why this format?
Backtesting is the single most important piece of evidence an investor has before allocating capital to an algorithm. And yet the industry standard for sharing that evidence is... a screenshot, a PDF, or a proprietary dashboard that only exists inside one platform's walled garden.
This creates three concrete problems:
- No portability. A developer who wants to publish the same strategy on multiple platforms, or move to a new one, has to redo their backtests from scratch because there's no common format to carry the results over.
- No independent comparison. Investors comparing two algorithms from two different platforms are comparing apples to oranges, because each platform computes and presents metrics slightly differently, with no shared source of truth underneath.
- No lineage. Backtest reports rarely carry any lineage across runs, across studies, or across the different universes and timeframes an algorithm has actually been tested against — so there's no way to tell whether this month's numbers came from the same code, or the same conditions, as last month's.
Underneath all three is the same root cause: backtest results have never had an open, portable data format. They've always been locked to the platform that produced them.
What problems does it solve?
The Open Backtest Format defines a single, structured schema for everything a backtest report needs to be independently verifiable:
- Strategy metadata — the algorithm identifier, version, and a hash of the exact code that was run, so the report is tied to a specific, unambiguous version of the strategy.
- Algorithm fingerprint — a deterministic hash of the algorithm's actual code, not just a version label, so it's provable whether the code behind two backtests is truly identical, or silently changed between them.
- Study-first structure — backtests aren't a flat trade log; they're organized as one or more studies, each a single research question scoped to one universe, one sample type (in-sample, out-of-sample, walk-forward), and one or more time windows. More on that hierarchy below.
- Execution configuration — the dataset, exchange, symbol universe, timeframe, and fee/slippage model a study's runs were executed under, recorded explicitly instead of buried in a platform's internal config.
- Metrics per run — CAGR, Sharpe ratio, Sortino ratio, Calmar ratio, maximum drawdown, win rate and recovery factor, calculated from the ledger using open, documented formulas.
- Study lineage — every window and run inside a study links back to it, so a strategy's track record within that study can be replayed window by window instead of read off a single aggregated headline number.
- Cross-backtest lineage — every run carries its algorithm fingerprint and study id forward, so separate backtest exports for the same algorithm chain together into one continuous, comparable track record instead of sitting as unrelated, siloed reports.
- Integrity checksum — so a report can be checked for tampering after the fact, and two platforms running the same strategy against the same data should arrive at byte-identical results.
Because the format is open, none of this is locked to Finterion. A backtest report exported in OBF can be read, re-verified, or re-plotted by any tool that implements the spec — including, eventually, tools we didn't build.
The hierarchy
The reason OBF works as well for judging one algorithm as it does for comparing several is that it doesn't store backtests as a flat pile of numbers — it nests them in a fixed hierarchy:
- Study — one research question ("does this momentum strategy hold up across market regimes?"), scoped to exactly one universe and one sample type (in-sample, out-of-sample, or walk-forward).
- Universe — the market, quoted asset, and base-asset symbols the study is run against. Every algorithm and every window inside a study shares this one universe, so nothing about a comparison is an apples-to-oranges artifact of different underlying data.
- Windows — a study is sliced into one or more walk-forward time windows (e.g. eight quarterly windows spanning 2022–2024) instead of a single date range, so a strategy has to prove itself across multiple periods and regimes, not just the one window that happened to look good.
- Runs — every (algorithm × window) pair produces a run: one simulated execution with its own trade ledger, equity curve, and computed metrics.
- Metrics — computed per run from that run's own ledger, then pooled per algorithm across every window in the study, so you get both a per-window view and a "how did this algorithm do across the whole study" view.
Here's a study rendered straight from OBF data — one universe, eight walk-forward windows, three versions of the same strategy:
Studies
Because a study is split into multiple walk-forward windows over the same universe, an algorithm can't lean on one lucky backtest period to look good. Its per-window runs either hold up quarter after quarter and regime after regime, or they don't — and that's visible directly in the ledger, not smoothed away into a single headline return. That's what metrics like consistency and stability in the schema are for: they're computed across a strategy's windows within a study, not within a single run.
Here are the study's eight windows themselves — each one runs all three algorithms and records how many came out profitable, so a strategy that only wins in bull-trending quarters stands out immediately:
Universes
Backtest Windows
Runs
Metrics
Walk-forward windows — 3 runs each (one per algorithm)
Universes
Because every algorithm in a study runs against the identical universe and the identical set of windows, their computed metrics are directly comparable — not "algorithm A backtested 2023 on Binance" vs "algorithm B backtested 2021–2024 on Kraken." Ranking algorithms within a study is a fair comparison by construction, which is exactly what the two tables below do: both are pooled across all runs for the three algorithm versions in the study above, and both key off the same algorithmIdHex so you can line up a strategy's computed metrics against its raw trading activity.
Computed metrics — pooled across the study's windows
Algorithm | Total Return | CAGR | Sharpe | Sortino | Calmar | Max DD | Volatility | Net Gain % |
|---|---|---|---|---|---|---|---|---|
3c4d5e6f | +120.3% | +31.2% | 1.95 | 2.68 | 1.61 | 19.4% | 22.1% | — |
a1b2c3d4 | +84.2% | +23.1% | 1.74 | 2.31 | 1.42 | 16.3% | 18.7% | — |
7f8e9d0c | +41.6% | +12.8% | 1.12 | 1.48 | 0.86 | 14.9% | 14.2% | — |
Algorithm
3c4d5e6f
Total Return
+120.3%
CAGR
+31.2%
Sharpe
1.95
Sortino
2.68
Calmar
1.61
Max DD
19.4%
Volatility
22.1%
Net Gain %
—
Algorithm
a1b2c3d4
Total Return
+84.2%
CAGR
+23.1%
Sharpe
1.74
Sortino
2.31
Calmar
1.42
Max DD
16.3%
Volatility
18.7%
Net Gain %
—
Algorithm
7f8e9d0c
Total Return
+41.6%
CAGR
+12.8%
Sharpe
1.12
Sortino
1.48
Calmar
0.86
Max DD
14.9%
Volatility
14.2%
Net Gain %
—
Trading activity — the runs behind those metrics
Algorithm | Profit Factor | Win Rate | # Trades | Avg Trade | Avg Duration (d) | Max DD Duration (d) |
|---|---|---|---|---|---|---|
3c4d5e6f | 2.12 | 61.0% | 301 | +0.8% | 22.0 | 15 |
a1b2c3d4 | 1.87 | 58.0% | 214 | +0.9% | 36.0 | 19 |
7f8e9d0c | 1.34 | 52.0% | 128 | +0.6% | 54.0 | 27 |
Algorithm
3c4d5e6f
Profit Factor
2.12
Win Rate
61.0%
# Trades
301
Avg Trade
+0.8%
Avg Duration (d)
22.0
Max DD Duration (d)
15
Algorithm
a1b2c3d4
Profit Factor
1.87
Win Rate
58.0%
# Trades
214
Avg Trade
+0.9%
Avg Duration (d)
36.0
Max DD Duration (d)
19
Algorithm
7f8e9d0c
Profit Factor
1.34
Win Rate
52.0%
# Trades
128
Avg Trade
+0.6%
Avg Duration (d)
54.0
Max DD Duration (d)
27
Same three algorithms, plotted instead of tabulated — the equity curve time series OBF records for every run, reconstructed straight from the ledger:
Equity curves — same universe, same windows
Algo A
+84.2%
Algo B
+41.6%
Algo C
+120.3%
And the same best-in-class comparison as big-number tiles rather than table rows — the view Finterion's own Summary and Study pages lead with, ranking best / worst / average per metric across the study's algorithms:
Best-in-Class metrics — ranked across the study's algorithms
None of these numbers are separate claims to take on faith — the windows, the tables, the equity curves and the KPI tiles are all views over the exact same runs, which is why they always reconcile with each other. Strategy metadata and input assumptions pin down what was run, against what universe, and over which windows; the integrity checksum covers the whole bundle, which is what makes tampering detectable after the fact.
How does Finterion support it
Supporting the Open Backtest Format is a first step, not a finish line. We're working on public OBF exports for every published algorithm, and we'd like to see the format adopted more broadly across the industry — the more platforms that speak the same language for backtest results, the more the entire space benefits from real, independently verifiable transparency.
Ready to see it in action? Browse our algorithm store and check out a backtest report, or read more about how we score algorithm quality in our Performance Score guide.
Questions about the Open Backtest Format? Reach out at support@finterion.com
Related Documentation
Table of Contents
Why this format?
What problems does it solve?
The hierarchy
How does Finterion support it
About the Author
Marc van Duyn
Founder & CEOMarc is the Founder and CEO of Finterion. He is passionate about making algorithmic trading accessible to everyone.