Structure & Formatting

Structure Rules:

• Statements should not include the quality of the outcome. For example, never use adverbs like ‘accurately’, ‘effortlessly’, ‘quickly’’, efficiently’, ‘easily’ anywhere in a success statement. Do not use them, or words like them, at all
• Do not begin statements with the work ‘if’
• State the success statement in the affirmative
• Do not use ‘and’ or ‘or’ in the statements
• Do not put suggestions about ‘how’ or ‘where’ in the statement
• Do not begin or end a statement with an adverb. Pay special attention to this
• Do begin each statement with a verb
• Do not use connective words in a statement. Never use “and” to connect to things. Never use “or”. These would be better suited for separate statements
• Do not reference end users in the statement. Do not use words like “you” or “your”. Do not begin a statement with “You” or “Your”.
• When you need to included examples, instead of using "such as" or "for example" please append the statement with a comma, then "e.g.," and finish with a comma and "etc."
• For statements about what must be avoided, begin the statement with the word ‘Avoid’

Instruction: this is the original form of the instructions for formatting and structure and approach to the success statements. It didn’t always work, but I keep them here even though I have replaced some of these with examples

Formatting Rules (Survey Version):

• Do not output any content before the list of statements
• Do not output anything after the of statements
• Do not output a test-fit example
• Always make sure the statement is relevant to the current step, and not a preceding or subsequent step
• Statements should be in a logical sequence of precedence and dependence.
• Do not generate a statement that restates the job step

Formatting: Most of these are not covered by examples, and in some cases they related to overall structure of the output. I may have stated elsewhere that ChatGPT by its very chat nature is chatty. I’ve done whatever it takes to get it to simply output a list. Most of the time it works. But there are two scenarios to watch out for: 1. If the first run gets verbose or chatty, simply re-instruct it on the error and then have it regenerate, or 2. if you’re running a long chat session (e.g., reinputting steps to bash out a map) memory begins to hit it’s limit and it has difficulty referring all the way back to the top where the instructions are.