Minimum data requirements belong in the mandate contract

Minimum-data eligibility specifies the observations a strategy requires before it may propose an entry. We would compile that requirement as a predicate over the state snapshot. A model’s ability to write a plausible thesis does not establish that the required evidence was present.

An illustrative strategy requires twenty completed five-minute bars, a quote no older than three seconds and a reconciled position record. Snapshot A has all three. Snapshot B has nineteen completed bars plus a provisional current bar. Snapshot C has twenty completed bars but a seven-second-old quote. Only A qualifies under this proposed contract.

Count the observations the rule names

The provisional bar in B does not satisfy a requirement for completed bars. The compiler should name the relevant timeframe, venue and instrument, with a completeness flag and availability timestamp for each row. Twenty duplicate observations of the same bar supply one completed bar, rather than twenty independent observations.

We would specify the strategy’s behavior under partial eligibility. It can record an observation, request missing data or defer an entry. Managing an existing position may use a separate emergency data contract; reusing the entry requirement for every exit can produce an unintended inability to reduce exposure during a feed outage.

The proposed fixture includes the exact boundary: nineteen versus twenty bars, a quote exactly three seconds old and one just beyond that age. It also includes a complete history for the wrong instrument. Valid counts with incorrect identity should fail the predicate before the model proposes an entry.

The decision record should list which requirement failed, rather than reporting only “insufficient data.” That makes recovery specific: wait for a completed bar, refresh a quote or reconcile account state. A successful refresh should replace the relevant observation while preserving the lineage of the earlier failed snapshot.

We would measure false eligibility and unnecessary deferral separately. The first admits actions without the required information. The second suppresses eligible opportunities. Passing the fixture demonstrates interpretation of a data contract; evaluating whether the chosen requirements improve trading needs a different experiment.

This is a useful boundary for a builder because it converts a vague demand for “enough history” into an inspectable dependency set.

Our mandate compiler provides the wider instruction-to-action framework for this data eligibility rule. The published controls research supplies its historical background. DXAP publicly describes the corresponding separation between model proposals and external policy checks.