BC-0075 · CORE GUIDE

Editorial review 2026-09-26

Roblox Build Prompt Iteration Plan

Plan prompt iterations as decision gates rather than a fixed script of messages. Establish a small loop, inspect it, repair the observed blocker and expand only after the relevant checks pass; keep alternate next steps for unexpected results.

Objective

Build an adaptive iteration plan with observable gates and scope decisions.

Before you start

  • Choose a small initial loop with a clear outcome.
  • Define what can be inspected after the initial request.
  • Keep optional systems separate from the minimum coherent experience.

Steps

  1. Write the initial interaction and its first acceptance check.
  2. Prepare distinct next actions for success, a reproducible defect and unresolved behavior.
  3. Inspect before selecting the next prompt.
  4. Add regression checks for behavior already working.
  5. Defer expansions that depend on unverified systems.
  6. Finish with a record of implemented behavior, remaining defects and deliberately excluded scope.

Verification

  • The next stage depends on an observed result rather than the planned calendar.
  • Each stage has a concrete output and acceptance check.
  • Working behavior is retested after relevant changes.
  • Deferred features are not presented as implemented.
  • The final record distinguishes a coherent smaller version from an unfinished larger promise.

Limitations

  • An iteration plan does not guarantee a fixed number of messages or a predictable generation outcome.
  • Some design changes require revisiting an earlier gate.
  • The sample sequence is editorial guidance and does not claim an actual game passed these stages.

Working notes

A rigid sequence assumes each generation behaves as intended. If the initial pickup fails, a planned shop prompt is no longer the sensible next step. Write the next decision in terms of observed state: continue when the core action works, repair when a specific transition fails, or narrow the design when the intended behavior cannot be inspected.

Give each stage a useful output. The first stage produces an inspectable interaction; the next resolves its most important discrepancy; a later stage improves clarity without changing the established rule. A stage described only as 'make it better' has no completion criterion and can consume an open-ended series of revisions.

Attach preservation checks as the game develops. Once retry works, later art or layout changes should include a retry regression check. The plan should accumulate known responsibilities, not assume that passing a check earlier protects it from future edits.

Keep scope cuts available. If a planned advanced system depends on unverified behavior, defer it and complete the smaller experience. Record what was removed and why so the listing and later prompts do not continue to promise it. Stopping at a coherent smaller loop is different from declaring an incomplete larger design finished.

Examples

  • Decision plan: create a parcel-to-depot interaction; if collection or completion fails, repair that transition; if both work, inspect failure and retry; once the loop is coherent, improve the goal feedback; only then consider an optional route whose behavior can be tested separately.
  • Branch after observation: the scenery is appealing but the depot does not recognize delivery. The next instruction addresses carrying-to-completion state, not more scenery. If the corrected delivery breaks retry, the next gate remains loop reliability rather than expansion.
  • Scope cut note: persistent upgrades were proposed but are not part of the inspected prototype. Retain an optional within-session route choice instead, and remove saved-upgrade promises from the working brief and presentation until that separate system is verified.