The first problem
Early tests showed that near-answer hints moved players forward but removed the satisfaction of discovering a rule. Repeating the objective was equally weak for someone who had already searched the room. One player had missed a key while another held the same key but could not identify its destination, so a single fixed sentence could not resolve both situations.
The problem became more visible across five connected stages. A system that told a player to find a key after the electrical box was already open immediately lost credibility. We therefore stopped treating hints as a list of prose and rebuilt them as rules that read the current game state.
Dividing progress into states
Each stage is divided into investigation, clue discovery, item collection, item use, puzzle input, and exit states. A strong hint checks inventory, opened devices, successful inputs, and visited spaces, then suggests only the nearest incomplete action. Consumed keys and tools have separate completion flags so they are never requested again.
Wording preserves the final act of reasoning. Instead of printing a code, the hint may ask the player to read shapes as stroke counts. For the magic-square puzzle, the system first explains that a grid beginning at zero totals twelve and one beginning at one totals fifteen, then introduces the shape rule in a later hint.
What failed and what changed
Normal and strong hints differ in the information they provide, not in sentence length. A normal hint restates the current objective. A strong hint uses the current progress state to narrow the next location to investigate or the purpose of an item already held. Strong-hint use appears beside time and deaths in the result record. It documents the help used during the run; requesting a hint does not block stage completion or the ending.
We also found a bug where closing an already opened panel returned it to a locked message. Visibility and lock state had shared one value. Separating those states stopped used keys from disappearing into a loop and prevented completed interactions from requesting the same action again.
Review criteria and result
Every hint is called from the opening state through the final action in test mode. We verify that it does not reveal an unseen clue, repeat completed work, or provide different information in Korean and English. Even at the last step, the system explains the rule and its application before giving a complete solution.
Clear screens now show time, strong hints, and deaths in the same order for every stage. Hints act as a safety net that restarts observation rather than a button that plays the game. New puzzles will continue to begin with a state table and likely failure points before any hint sentence is written.
Related records
Hint writing depends on puzzle rules and interface feedback. The puzzle devlog explains clue placement and difficulty review, while the play guide documents the difference between normal and strong hints.