A gold target has three decimals. Can the chosen contract trade there?

Preserve the raw signal before translating it

A signal calculated from a spot-price feed may contain three decimal places. A futures test may accept only a particular price increment. Moving the raw level onto that grid changes the question being tested, even when the adjustment looks tiny on a chart. The researcher needs to preserve both the original level and the declared translation rule.

This is a research-method article, not a result case study. It reports no CyGuru trade, hit rate or profit. Its purpose is to prevent a data-format decision from quietly becoming a favorable outcome. The relevant comparison is between a model’s reference threshold and the valid price grid of the exact instrument used in the experiment.

Gold products do not share one universal tick

CME’s Gold futures fact card specifies a minimum price fluctuation of $0.10 per troy ounce for the standard GC contract. Its current 1-Ounce Gold FAQ specifies $0.25 per troy ounce for 1OZ. These primary product references were checked on 10 November. The examples show why a generic field called “gold tick size” is insufficient.

They do not determine the quotation precision or order rules of OANDA:XAUUSD, another broker’s CFD, or a physical dealer. Record the venue, symbol, contract and applicable specification. Where the instrument in the original signal differs from the replay instrument, the analysis must label that translation rather than present the replay as identical evidence.

Source notes: [1] CME Group — Gold Futures and Options fact card · [2] CME Group — 1-Ounce Gold futures FAQ

A hypothetical target sits between two prices

Consider a hypothetical long target at 4,120.26 on a grid spaced at 0.10. The adjacent grid levels are 4,120.20 and 4,120.30. If a hypothetical bar high is 4,120.20, the raw target has not been reached. A target rounded down to 4,120.20 has been touched, while a target rounded up to 4,120.30 has not. None of these invented values describes a historical CyGuru signal.

The example does not choose the correct trading rule. It exposes why the rule must be chosen before inspecting the outcome. “Round to nearest,” “round outward” and “reject off-grid levels” are different model policies. A consistent policy also needs a stated treatment for halfway cases. Selecting whichever rounded value turns a miss into a hit destroys the comparison.

Entry, stop and target translations interact

Changing only the target is not enough when the entry and stop also require translation. Recompute the reference distances from the actual translated fields and retain the raw distances alongside them. A displayed reward-to-risk ratio should identify which set it uses. Otherwise the headline ratio can describe one geometry while the outcome table evaluates another.

I would store raw entry, stop and targets; the tick specification and version; each translated field; the direction of the adjustment; and the difference in ticks. Then check that the translated ordering remains meaningful for the declared direction. If an adjustment collapses two distinct levels onto the same price, the process needs an explicit invalid or degenerate classification rather than a silently accepted row.

Implement comparisons without display-driven arithmetic

A chart label is for reading, not necessarily for computation. Preserve the machine values used by the source, and perform comparisons with a documented decimal or integer-tick representation. A two-decimal label should not overwrite a more precise source value merely because it looks cleaner. Conversely, extra displayed digits should not be mistaken for additional market precision.

The audit should include values just below a threshold, exactly at it, and just above it, plus both directions and the chosen halfway rule. These are synthetic verification cases and must remain clearly separated from production signal records. Their purpose is to prove that the implementation follows its policy, not to demonstrate that a trading model works.

Report sensitivity without rewriting history

If multiple reasonable grid conventions materially change a nominal result, disclose that sensitivity. Keep the main convention fixed and label alternatives as separate analyses. Do not revise the original event levels to match a later preferred convention. A stable evidence ledger should let a reader trace every reported outcome back to the exact input and rule set.

Even a correct threshold classification remains a nominal price observation. It does not establish that a resting order filled, that the relevant bid or ask was available, or that commissions, spread and slippage were covered. The Research Lab can test those questions only when matching evidence exists. A valid tick is one necessary input to a reproducible experiment, not a certificate of executable profit.