Consistent lawn observations for Dave

Dave's ordinary prompt previously described the same plant with native zero-based coordinates in lawn.planted and player-facing one-based positions in story.actualDefenses. Enemy sightings already used player-facing positions. The prompt also asserted that fog hid enemies solely from the fog-stage flag, including after Blover had exposed all nine planting columns in every row.

Lawn rows, living plants and the roof favor now use player-facing positions in the prompt. Native request data, save data and reply commands remain unchanged. Active legacy episodes receive a separate commandTargets mapping only when they advertise targeted directions. The model copies its rowIndex and columnIndex into the unchanged native command. Terminal episodes omit that mapping. Two source 1-9 event descriptions now name the same fixed cell as the living-plant observations (row 3, column 1).

Dave and plant conversations share the current visibility description. Complete measured clearance describes the planting grid as unobscured; offscreen enemies remain unknown. Missing or partial measurements establish neither a fog wall nor a safe lane. The existing public fogClearCells protocol is sufficient; this does not enable the separate unreleased fog-density experiment.

Verification

• Coordinate and complete-clearance regressions fail on the previous behavior. A legacy fixture initially combined incompatible identity protocols; its corrected form proves both native direction kinds keep their original targets.
• All 523 NPC checks pass, including source choices, plant names, roof favor identity, account context, saves, provider behavior and command validation. Two older whole-snapshot expectations were updated to the new display row; native snapshots and source chronology are still compared unchanged.
• Hosted build, typecheck and open-source export verification pass.
• Six single-attempt real-provider responses use captured compiled snapshots, projected onto the unchanged public observation protocol. Qwen and Gemini each answer an English closest-enemy question, an English post-Blover plant question and a Chinese fog question. All four closest-enemy answers identify the actual observed row/column; both post-Blover answers recognize the clear planting grid and keep the light decision null. No account, notebook, native save or Yumina billing records are written. Six upstream generations were made, without rerolls.

Partial-fog prose remains below the broader writing-quality goal. Qwen still describes the bottom lane too broadly as unreadable; Gemini treats the minimum measured clear columns too much like a precise boundary. The Plantern's vague boast after Blover is also not evidence that it caused the clearance. These samples do not establish universal factual accuracy and are not a reason to release the separate fog-density candidate. No extra prompt-only retries are being made to turn this sample into a pass.

Receipts: game worktree output/lawn-observations/writing/; this worktree output/lawn-observations-{npc-v2,build,typecheck,oss}.log. This is a host-only change: no new game UI, pauses, permissions, native effects, difficulty, schemas or save formats. Account-selected models, persona and writing settings retain their existing paths.