The EXPORT CSV button on the Performance tab is a convenience that quietly transfers a large amount of responsibility to you. What lands on your disk is not the account's performance. It is one view of the account's performance, taken under a specific set of filter selections, at a specific moment, with a period definition that is implied by a button rather than stated in the file. Everything that makes the figures interpretable is sitting on the screen you exported from, and none of it necessarily travels with the download.
That is not a criticism of the export. Every performance surface works this way. It is a warning about the step that follows, which is someone pasting those figures into a presentation for an allocator who was not there when the filters were set.
An export is a view, not a dataset
Look at the control strip in the capture below. Four selections stack above the numbers, and each one changes them. The reality toggle chooses ALL ACCOUNTS (REAL) or BLOCKCIRCLE ONLY. The source filter runs ALL, AUTOPILOT, MTE, PAPER, MANUAL and WALLET. The account selector offers All accounts (cumulative) alongside named accounts, here Alpaca and Secondary. The period runs 1D, 7D, 14D, 1M, 3M, 1Y, ALL and CUSTOM, and in this capture it is on 1Y.
Two of those are composite construction decisions in everything but name. Choosing which accounts are in the view is defining a composite. Choosing which sources are in it decides whether discretionary manual trading, automated flow and any simulated activity are being presented as one track record or as several. If a PAPER source is left in the selection, the export mixes simulated fills with real ones, and no footnote can rescue a number built that way. Recording the four selections is therefore not administrative tidiness. It is the composite definition, and a reviewer is entitled to it.
The practical discipline is small. Before you press export, capture the state. Screenshot the control strip, or write the four values and the export date into the filename, or both. It costs nothing and answers the first three questions anyone will ask about the file.
The period label is doing more work than it looks
A button marked 1Y is a trailing window ending on the day you pressed it. That is a perfectly reasonable default and a poor thing to put in a report, for three reasons.
It is not a calendar period, so it cannot be compared with a calendar year benchmark return without adjustment, and an allocator will attempt exactly that comparison. It is not stable, so the same button pressed three weeks later produces a different figure and neither is wrong. And nothing in the artefact tells a reader when the window opened.
For anything going outside the firm, use the CUSTOM option and pin explicit start and end dates that match your reporting calendar, then state them in the document in full. If a trailing window genuinely is the right presentation, label it as trailing and give the end date. A GIPS-style presentation expects returns shown period by period against a stated benchmark for each period, and a rolling window that shifts every time somebody refreshes the page cannot satisfy that.
Add the as-of timestamp separately from the period. Venue data backfills, trades get amended, transfers get reclassified, and a figure for a closed quarter can move weeks later. Without an as-of stamp, two reports that disagree look like an error rather than two vintages of the same truth.
Cash flows decide whether the return means anything
This is the disclosure that governs all the others. The Account Equity panel underneath the tiles describes itself as the real cumulative balance summed across your connected accounts, and in the capture it carries a net figure of positive 396.07% while the tiles above it show a profit factor of 0.08 and expectancy of negative 1.20%.
A balance series moves for two reasons. Trading, and money arriving or leaving. Any figure derived from a balance series is therefore a money-weighted quantity unless external flows have been stripped out, and money-weighted returns describe the investor's experience rather than the manager's decisions. A GIPS-style presentation wants time-weighted returns precisely so that a manager is not credited or penalised for the timing of contributions they did not control.
So before a single number from an export goes into a report, establish which you have. Do not assume, and do not guess at the calculation. Test it. Choose a period that contains a deposit or a withdrawal you can identify, note the reported figure, then choose a comparable period without one, and see how the series behaves around the flow date in the exported balance column. If the reported return steps at the moment cash moved, you have a money-weighted quantity and you will need to compute time-weighted returns yourself by chain-linking sub-period returns around each external flow. If it does not step, say so in the footnote and record the test you ran.
State gross or net of fees in the same breath, and be specific about which fees. Venue commissions and funding are usually inside a trading P&L. Fund level management and performance fees usually are not, and an allocator assumes net unless told otherwise.
Two families of statistics, and saying which you cited
The tiles in the capture show Sharpe 1.23 and Sortino 9.84 sitting next to a win rate of 25.29%, a profit factor of 0.08, expectancy of negative 1.20%, max drawdown of negative 91.31% and 87 total trades. If you export that view and lift figures from it into a document, you will at some point be asked how a Sharpe above one coexists with a profit factor of 0.08. You need an answer ready that is not a guess.
The general reasons two performance statistics on one page legitimately differ are worth keeping as standing language. They can be measured over different windows. They can be computed on different underlying return series, and a per-trade series and a periodic balance series are different objects with different observation counts, different weighting and different handling of intervals when no position was open. One may be annualised and the other expressed per trade or per period. They can weight the distribution differently. And sample length differs sharply, since a ratio built on daily balance points has far more observations than one built on 87 closed trades.
I am not going to assert which of these explains this specific pair, because that requires the exact construction and I do not have it, and there is no reason to presume any figure is being computed incorrectly. The way to resolve it for your own reporting is to hold everything constant and vary one control at a time, then observe which tiles respond to a change in period and which respond to a change in account selection. That tells you which family each number belongs to. Cite one family per claim, label it in the report, and never present a trade-level statistic and a balance-level statistic in the same row of a table as though they described the same thing.
The footnote block, and what a share link is not
The minimum footnote block for anything leaving the firm reads roughly as follows. Composite definition, meaning which accounts and which sources are included and which are excluded and why. Period, stated as explicit start and end dates with an as-of stamp. Return basis, meaning time-weighted or money-weighted, with the treatment of external cash flows described. Fee basis, gross or net, with fee types enumerated. Valuation source and timing, since crypto marks depend on which venue and which snapshot. Benchmark, named per period, with the reason it is appropriate. Sample size, which here is 87 trades and should be disclosed rather than discovered. And the statement that past performance does not indicate future results, which is boilerplate and is also true.
The SHARE button next to the export deserves a separate line of thought. Whatever a share link does technically, treat it as a live surface rather than a document. A recipient opening it later may see figures that have moved because data backfilled, because the period is trailing, or because the underlying account kept trading. That is fine for internal collaboration and unacceptable as the artefact of record. For anything an investor relies on, freeze it. Export the file, produce a dated PDF, archive both alongside the view state, and send that. The link then sits next to the record as a convenience rather than standing in for it.