Live evaluation · LLM commodity signals

Live results across liquid commodities.

Copilot LLM signals are scored, sized and evaluated in a defined paper book covering liquid commodity contracts across agriculture, energy and metals. The path is compared with the Bloomberg Commodity Index (BCOM) and the S&P 500. This is a live evaluation of those signals — not a marketed product return.

Last mark
Universe 19 contracts
Window
Frequency Weekly marked
Construction Equal-vol long/short
Total return
Signal book, period to date
Vs BCOM
Excess over BCOM price
Ann. volatility
Daily returns, annualised
Max drawdown
Peak to trough, NAV path
Hit rate
Weeks with positive P&L
Sharpe (rf 0)
Daily, not GIPS

How the live book is built

Signals are produced from Permutable’s commodity intelligence layer, then passed through Copilot models that rank direction for each contract.

01 / Universe

Liquid contracts

Agriculture, energy and metals names overlapping the Bloomberg Commodity Index universe — scored on the same point-in-time cut used in the live paper book.

02 / Signal

LLM direction and persistence

The model converts driver indices, event flow and source evidence into a signed signal. Tone is separated from directional pressure so media volume alone cannot flip the book.

03 / Risk

Risk overlay

Each contract is scaled to similar risk. Gross exposure is levered, net exposure floats with the cross-sectional signal, and sector mix is allowed to vary with the book.

04 / Mark

Daily equity, weekly chart

The book is marked daily (NAV starts at 100). The chart shows Friday marks plus the first and last session. Fills, fees and slippage are not fully modelled.

Signal book versus BCOM and S&P 500 · weekly

Signal book NAV versus the Bloomberg Commodity Index (BCOM) and the S&P 500, all based at 100 on the first date. The book is a levered long/short paper path; BCOM and the S&P 500 are long-only price indices, shown as public references rather than like-for-like product benchmarks. The book line is daily mark-to-market cumulative P&L, added rather than compounded. Click a series to hide it.

Signal book versus the Bloomberg Commodity Index and S&P 500
Weekly marks

Individual ticker performance

Standalone performance of each contract, ordered most liquid first. Each line is daily mark-to-market cumulative P&L on a 100 base: while a position is open the curve moves every session with the roll-adjusted price, measured against the entry price and unweighted, so it is independent of book sizing. Win rate is the share of closed trades with positive P&L.

Asset performance breakdown

Every contract in the live book, ordered most liquid first. Standalone P&L is the unweighted position-level result, in points of entry notional, independent of sizing. Book contrib is the weighted P&L this contract added to NAV — those sum to the book total. Avg |wt| is the mean absolute weight over the window.

Click a column to sort
Liq. rank Ticker Contract Sector Avg |wt| Trades Win rate Avg trade Standalone P&L Book contrib

FAQ

Are the signals backward-generated?
No. They are live, point-in-time signals, produced as the information arrived, and have been running in this form since 2024.
Why are you sharing this?
Permutable is open in showcasing the properties of its dataset and technology. We operate as a technology company, not as an asset manager.
Do you trade this book?
No. This is a paper evaluation of live signals. Permutable does not trade on its data and this path is not a marketed product return.
How are prices constructed?
P&L is marked on Permutable’s own rolling of the front-month contract — a continuous front series, not fills on every expiry a live book would have held. Those prices are subject to capacity.
Does this include transaction costs?
No. The evaluation does not include fees, roll costs or execution slippage. It is a signal-path mark, not a traded P&L.
Why do some contracts start later than others?
The live universe was phased in. Names appear when that contract first received a live signal in this book, not because history was filled in afterwards.
Is this the Bloomberg Commodity Index?
No. BCOM is shown as a public price-index reference. The book is a long/short evaluation over overlapping liquid commodity contracts, not an official BCOM product or replication.
Is this investment advice?
No. The page is published to show how the dataset and models behave in a defined evaluation, not as a recommendation to buy, sell or allocate capital.
What is the in-sample / out-of-sample split?
Headline figures are for the full live window. The vertical line on the chart is a 1 January 2026 reporting cut: KPIs also show the path before that date (in-sample) and after it (out-of-sample) side by side. Signals themselves were live throughout — the split is a reporting convention, not a change in how they were produced.
Is the Sharpe ratio GIPS-compliant?
No. It is a simple daily Sharpe with a zero risk-free rate on the evaluation path, shown for comparison only.
What is Copilot?
Copilot is Permutable’s own signal layer: language models that turn commodity intelligence into a signed direction for each contract. It is not Microsoft or GitHub Copilot.
What does “subject to capacity” mean?
Front-month futures are not infinitely deep. Marks on this page assume those contracts can be used as the pricing series; they do not claim a given notional could be traded without moving the market.
How often is this updated?
The book is marked daily. The chart shows weekly Friday marks plus the first and last session. The date in the header is the last mark on this page.

Discuss the dataset or API. Email enquiries@permutable.ai or book a demo.