Volatility System · Trading Record

Trading record

Every observation below is appended once and never revised. Each one is a separate dated commit, and CI rejects any change to an already-published day, so a day that has been published cannot be quietly revised. That property, not the returns, is what will make this page worth reading.

Trading has not started. No capital has been deployed and no observation has been recorded, so this page currently reports that and nothing else. Engineering progress — what is built, what is not, and what each stage has to demonstrate next — is on the build-status page.

When results do arrive, three kinds will be reported separately rather than joined into one curve: backtest from historical simulation, forward simulation against live data with nothing at risk, and broker-executed reconciled against actual fills. They answer different questions and carry very different weight.

That separation is enforced by the data contract rather than left to prose: every observation declares which kind it is, and each series is charted and scored from its own kind alone. Backtests are not appended to this record at all — its guarantee is that a day was published before its outcome was known, which a historical simulation run over data already in hand cannot meet, so a backtest is published as its own diagnostic and read alongside rather than inside the record.

Build status Evaluation protocol Commit history Raw data

Loading the track record Every figure on this page is computed in your browser from the data file, which you can read directly.

JavaScript is required for the figures This page derives its statistics in the browser rather than serving precomputed numbers, so that the page cannot disagree with the data behind it. The underlying record is plain JSON and is readable without JavaScript: /data/track-record.json.

— days elapsed —

Inception — · latest observation — · — trading days recorded

Is this long enough to mean anything?

Cumulative return

Indexed to 0% at the start of the selected series, net of all modelled frictions. Forward-simulated and broker-executed observations are never joined into one curve. Hover for daily values, or open the table below.

Drawdown from peak

Distance below the running high-water mark. Plotted separately rather than on a second axis, so neither scale distorts the other.

Statistics

Cumulative —

Annualised — Withheld under 21 observations

Ann. volatility —

Sharpe (ann.) — Raw — read with MinTRL

PSR vs zero — P(true Sharpe > 0)

MinTRL — Returns needed at 95%

Max drawdown —

Current drawdown —

Skewness —

Kurtosis — Raw; normal = 3

Cost share — Of gross P&L

Total costs —

Table view — every observation

All published observations, newest first
Date	NAV	Return	Gross P&L	Costs	Positions	Mode	Kind

Integrity

What this record is, and what it is not

This will be a versioned, self-reported record. That is a real and unusual property, and it is also a bounded one. Three mechanisms back it, and all three are checkable by anyone:

Every observation is a separate dated commit. The git history is the change trail — you can see when any number here first appeared, and what it was before.
CI rejects edits to published history. A validator runs on every push and fails the build if an existing observation is modified, removed, or reordered. Appending is the only legal change.
Every statistic is computed in your browser from the raw data file, so this page cannot quietly disagree with the numbers behind it.

What that does not establish is that any trade happened. Git commit dates can be set to arbitrary values and history can be rewritten, so what the mechanisms above prove is change control over a file, not the existence of a fill. The append-only validator is also path-filtered to the data file itself; branch protection on main, not the workflow, is the durable guard.

The independent check is broker reconciliation, and it is the intended one: Interactive Brokers Flex reporting is already connected read-only, so executed trades can be pulled from the broker and reconciled against what this record claims. Until that reconciliation is published alongside the record, treat every number here as self-reported and dated, which is exactly as much as it is.

The reported Sharpe ratio is deliberately not the headline. Alongside it sits the minimum track record length — how long this record would have to run before that Sharpe could be told apart from luck at 95% confidence. While the record is short, the page says so plainly.

This page was last updated September 8, 2026. The observations themselves carry their own commit dates, which is the timestamp that matters.