PDA Working Prompts

These prompts are meant for actual use in ongoing work, not as full summaries of the method.

They try to preserve the result of the pda-lab optimization without forcing the user to recreate the story-1-interactive-pip pattern through dozens of steering moves.

The optimization work behind them lives in pda-lab.

Everyday Prompt

Use this when the task is ambiguous enough that premature implementation would likely produce the wrong result, but not so high-stakes that you want maximum strictness.

Apply Progressive De-Arbitrarization. Do not answer the presenting task directly if hidden decision-relevant ambiguity remains. First localize the governing ambiguity. Separate forced structure from contingent policy. Descend only on the live layer. If the task seems specificity-sensitive, preserve already-settled concrete directions. If the task seems multi-authority, keep distinct authorities and state transitions explicit. When residual ambiguity becomes explicit and enumerable, switch to one numbered question at a time with lettered options and your best default marked (*). Recompose only when the remaining ambiguity is forced, explicitly chosen, or irrelevant to execution. Then produce an implementation-ready spec or carry out the implementation.

High-Stakes Prompt

Use this when the cost of wrong framing is high and you want stronger resistance to smooth but shallow progress.

Apply Progressive De-Arbitrarization and optimize for elimination of concealed arbitrariness rather than speed. Do not jump to implementation while hidden decision-relevant policy remains. First localize the governing ambiguity. Separate forced structure from contingent policy. Use the minimum grounding needed for the task:

• if the task is closure-sufficient, work from the raw task and target closure
• if it is specificity-sensitive, preserve already-settled concrete directions
• if it is multi-authority, keep distinct authorities, state transitions, and failure semantics explicit Descend only on the live layer. Do not elaborate without refining. When residual ambiguity becomes explicit and enumerable, switch to one numbered question at a time with lettered options and your best default marked (*). Recompose only when the remaining ambiguity is forced, explicitly chosen, or irrelevant to execution. Then produce an implementation-ready spec, delegation-ready task statement, or implementation.

What These Capture

These working prompts are not the whole theory. They are a compact operationalization of what the optimization phase established:

• wrong-abstraction answering is the main early failure mode
• closed-formulation is the strongest common grounding artifact
• some tasks need preserved concrete closure more than others
• one-at-a-time optioning is still the cleanest move once ambiguity becomes enumerable

When To Use Which

• Use the everyday prompt when you mainly want to avoid repeating the long rescue pattern from the interactive PiP story.
• Use the high-stakes prompt when the task is more likely to be specificity-sensitive or multi-authority, or when a smooth but wrong answer would be expensive.