A data cutoff is different from an analysis outcome window

A paper's data cutoff tells us where its record stops. A particular analysis can stop earlier, use a subset, or require a later outcome horizon. The continuous record freezes the research phase at August 15, 2026, while its headline cumulative realized P&L uses June 8 through July 26. We should carry the narrower window with that result.

The continuous record companion reports minus $217,000 realized P&L over that June-to-July window at a common 5.5 basis points per side, and minus $148,000 at zero fee. These rounded amounts belong to that analysis. Extending their labels to August 15 would silently claim an extra period whose outcome is absent from the quoted summary.

Different windows answer different questions

The companion describes June 8 to August 15 as 69 days at the data cutoff. Under inclusive calendar counting, June 8 through July 26 covers 49 dates, leaving twenty dates between July 27 and August 15. That derived contrast concerns labels and calendar coverage, not twenty additional days of known P&L. Timestamp endpoints and trading-day conventions should remain explicit in any new calculation.

A research artifact needs at least a collection cutoff, input eligibility interval and outcome interval. For a position opened before July 26 and closed afterward, the analysis must specify whether it enters a realized-only result, remains open, or is marked at the endpoint. A position opened after July 26 may be available in the frozen dataset while remaining outside this particular analysis.

An illustrative evaluation freezes data on September 30, accepts entries from September 1 through September 20 and scores each over the following 24 hours. Its last scored horizon can end on September 21. Calling it a September 30 performance study would misdescribe the included decisions even though the source file truly contains data through that date.

We would store those three intervals in every metric receipt, together with timezone, endpoint inclusion and treatment of unresolved positions. A reader could then determine whether two figures share the same population and whether a later release genuinely added outcomes or merely changed the collection boundary.

Keep the historical systems separated

Our controls paper companion records DX Terminal Pro from February 26 to March 18, 2026, a 21-day real-capital market with twelve tokens and a frozen production model family. Its tournament window differs from every live-fleet interval. A six-month research lineage is broader than either system's individual measurement window.

The later record's cutoff also separates pre-alpha research from current DXAP. The paper can motivate architectural questions, but its shorter realized-P&L window neither establishes a current result nor predicts what happened after it. The durable reporting habit is to put the interval next to the metric, so the reader sees which outcomes actually support the claim.