Gold trading basics · Lesson 94 of 100

Gold Rule Changes: Separate Emergency Repairs from Strategy Redesign

Use a change register to preserve old evidence, evaluate proposed revisions and prevent hindsight repairs from contaminating results.

Version The Change: instructional diagram for lesson 94 — CYGURU GOLD · Original lesson schematic. Examples are illustrative.

Every change needs a reason and an evidence boundary

After a painful trade, moving a parameter can feel like learning. Sometimes the change repairs a documented implementation fault; sometimes it only removes an uncomfortable historical result. These have different implications. Classify changes as operational safety repair, specification correction or economic hypothesis. A safety problem can justify immediate suspension, while a proposed improvement in profitability requires a new evaluation. Preserve the original observations and outcomes. The objective is to improve the process without retroactively claiming that the improved version made decisions before it existed.

Record the smallest meaningful change

A change register should state the previous rule, proposed rule, motivation, affected data, decision owner and activation time. Suppose version A used a twenty-bar threshold and a researcher later tests thirty bars because two losing entries look less attractive under the longer window. Version B is a new hypothesis. Report the number of alternatives examined and the examples that motivated the search. Avoid describing the thirty-bar version as more robust simply because it improves the same data that selected it. A small parameter edit can still be a substantial change to the opportunity population.

Keep return segments and units honest

Imagine version A records twelve comparable completed trades averaging +0.10R, while a later version B records eight averaging +0.20R. Totals are +1.20R and +1.60R. A combined descriptive mean is 2.80/20=+0.14R if R is defined consistently. That combined statistic is not evidence that the frozen B rule earned +0.14R over twenty trades, nor does it make the samples comparable in market conditions or costs. If each version uses a different R denominator or reporting policy, even the combined R average may need a different reconciliation.

Version-segment reporting
Segment	Trades	Mean R	Total R
Version A	12	0.10	1.20
Version B	8	0.20	1.60
Combined description	20	0.14	2.80

Assumes a comparable R definition; combined history is not twenty observations from frozen version B.

Use an emergency path that does not erase history

If the code submits duplicate orders, stop new submissions and reconcile existing exposure before considering a corrected release. Label the affected interval and retain the duplicate records. A correction can be necessary without establishing that the previous economic results remain directly comparable. Re-run specification tests, document the root cause and decide whether previously reported statistics require correction. Do not merely delete the fault from the backtest and continue presenting the result as the same uninterrupted process. A trustworthy revision explains both what was repaired and what earlier conclusions changed.

Schedule research outside the next-order decision

Separate the operator's execution role from the research role, even when one person performs both. During a frozen paper segment, the operator follows or suspends the specified process. During the review window, the researcher proposes a bounded change and plans evaluation. This separation reduces the temptation to make every new chart an exception. It does not prohibit urgent risk action. It requires urgent action to be labeled as such rather than disguised as evidence that a different strategy had always been in place.

Practice task: write two change records

Create one duplicate-order safety repair and one thirty-bar research hypothesis. Specify which can trigger immediate suspension, which needs held-out evaluation and how each affects the published record. Reproduce the +1.20R, +1.60R and +0.14R calculations, then explain what the combined mean cannot establish. Add a rollback reference and an activation timestamp. The finished register should let another reader distinguish operational correction, exploratory tuning and prospective evaluation without relying on the author's memory of why a parameter was changed.

Gold Rule Changes: Separate Emergency Repairs from Strategy Redesign

Every change needs a reason and an evidence boundary

Record the smallest meaningful change

Keep return segments and units honest

Use an emergency path that does not erase history

Schedule research outside the next-order decision

Practice task: write two change records

Can you apply the idea?

Make this lesson part of your process.

Gold Trading Mastery

Continue the beginner course

Lesson sources and further study

Core course references

Connected reading