Interval
Work since the previous summary
The work began with a screenshot-led refinement pass covering the equipment room, gem balance, globe, reef maps, launch transition, sticker details, and development diagnostics. Grid Runner then exposed a mismatch between the intended visual balance and the production generator. The user’s challenge to the 52% vertical result sharpened the requirement, and a final prompt made the horizontal, vertical and combined-diagonal split explicit in generation, scoring, tests, documentation, and fallback behaviour.
Direction of travel
How the work evolved.
Each step connects the user’s prompt to the practical change it produced.
Refined
Simplified the journey from room to reef
Prompt: Use the approved equipment-room view, strengthen the app-wide gem display, remove unnecessary globe and reef-map chrome, improve label and route clarity, and replace generic navigation icons with miniature scene-based controls.
Outcome: The approved room artwork became the live environment, gem balance gained a larger high-contrast capsule, the globe now foregrounds overall completion, and the reef maps use scored placement to keep labels close to pins while protecting geographic dots and reducing crossing routes. The return controls now visually represent the room computer and the globe.
Polished
Reworked launch and sticker information
Prompt: Keep the launch icon still, reveal the app with the familiar grid-change bubble effect at a larger scale, and only show local population information when it is reliable while adding an interesting fact.
Outcome: Launch now uses a full-screen bubble transition with doubled starting bubble size instead of animating icon parts. Creature panels omit unsupported population claims and present curated interesting facts for collected stickers.
Measured
Made grid quality visible in Grid Runner
Prompt: Run an independent catalogue audit, report horizontal, vertical, diagonal and reversed placements, and score how well each grid follows the defined rules.
Outcome: Grid Runner gained orientation and reversal totals, a transparent 0–100 adherence score, and diagnostics for intersections, shared cells, edge placement, isolated vertical words and word-length limits. The first 720-grid audit found no hard placement failures, but revealed that vertical words still made up about 52% in both profiles.
Corrected
Turned balance from a preference into a rule
Prompt: Treat the excessive vertical result as a defect, make the split explicitly close to one third horizontal, one third vertical and one third diagonal, then recheck the complete catalogue.
Outcome: Each level now receives exact integer targets for the three orientation groups, with counts differing by no more than one. Grid Runner treats any deviation as a hard-rule failure and shows actual against required counts. Across all 720 catalogue grids, both profiles finished at 33.3% horizontal, 30.9% vertical and 35.8% diagonal, with every fallback grid also compliant.
Delivered
Consolidated the verified experience
Prompt: Bring the completed navigation, launch, sticker, audit, and orientation work together in the shared product state.
Outcome: The complete verified set was synchronized cleanly. Private delivery details are intentionally excluded from this record.
What improved
Improvements made.
Evidence-led changes grouped around the product, implementation, design and working process.
A clearer app-wide gem balance
The gem symbol is twice its previous visual size and sits with the balance in a shared high-contrast capsule, giving the same readable treatment across screens.
A quieter, progress-led globe
Instructional labels, redundant status keys, and extra headings were removed. Overall completion now summarises progress across reefs, sites, and play types at the bottom of the globe view.
More legible reef maps
A scored layout replaces the first-available placement heuristic. It keeps labels associated with their pins, leaves location dots visible, heavily penalises route crossings, and retains horizontal exploration for edge locations rather than compressing the geography.
Navigation controls that explain their destination
The globe return control uses a miniature computer-and-desk scene, while reef maps use a small textured globe with a back arrow. The controls communicate where the player is returning without additional text.
A more coherent launch transition
The app icon remains stable while a full-screen bubble field, based on the existing grid-change effect, clears to reveal the equipment room. Reduced implementation duplication keeps the transition aligned with the game’s established visual language.
More trustworthy creature details
Population information is conditional on having reliable local evidence. Existing curated species and site material now supplies an interesting-fact row instead of presenting uncertainty as a population figure.
Exact three-way grid orientation
Levels containing 6–12 words now have explicit horizontal, vertical and combined-diagonal targets. The resulting catalogue-wide mix is close to one third per group, and the requirement applies equally to normal and fallback generation.
Pain points
Friction, constraints and open edges.
Resolved problems are separated from constraints and work that remains open.
Artwork approval needed to be unambiguous
An earlier instruction had kept the proposed room image out of the live app until approval. The replacement paused until the user confirmed it explicitly, after which the matching canvas allowed the existing camera positions and interaction areas to remain unchanged.
Simulator testing initially stalled
No simulator was available at first, and a later cloned test app failed to launch cleanly. Once simulator access was available, focused and serial reruns separated infrastructure noise from real regressions and enabled a complete visual route check.
Launch and map checks exposed real regressions
The first test run found incorrect launch-bubble expectations and one map-overlap size. Both were corrected, while older session and generation checks were updated to validate the selected target and seeded word sets instead of assuming a fixed order.
A high score hid a strong vertical bias
Generation treated four-axis balance as a low-priority preference, while Grid Runner reported three orientation groups. That classification mismatch allowed roughly 52% vertical placement to coexist with strong adherence scores until the user challenged the result.
The exhaustive catalogue audit is too slow for routine tests
The full two-profile audit took about 14 minutes. It was used as final evidence, then removed from the normal regression suite while fast orientation tests and Grid Runner diagnostics were retained.
Outputs & proof
What was delivered and verified.
The interface changes were compiled, exercised with focused unit coverage, and taken through the main room-to-dive path in the simulator. Grid generation received both a fast multi-seed regression and an exhaustive all-site audit, with the final explicit orientation rule passing everywhere.
Verification record
- The app and test bundle compiled successfully after the navigation, launch, sticker, and Grid Runner changes.
- Focused checks passed for the launch transition, reef-map collision layout at supported sizes, protected geographic dots, globe-label separation, completion percentage, sticker facts, and affected game generation.
- The simulator route passed from the approved equipment room through the globe, reef map, and dive selection, including the corrected miniature-globe return control.
- A focused generator run covered 200 grids from ten seeds across both play profiles; every level matched its exact orientation target.
- The full Grid Runner audit covered 36 sites and 720 grids across both profiles, with no hard-rule failures.
- Final catalogue placement measured 33.3% horizontal, 30.9% vertical and 35.8% diagonal in both profiles; all tested fallback grids followed the same rule.
Key decisions
Promote the approved equipment-room image without changing the existing camera or interaction geometry because its canvas dimensions match the live asset.
Use one shared gem capsule across the app instead of applying screen-specific contrast fixes.
Score reef-map label positions as a complete layout, heavily penalising route crossings and obscured geographic dots rather than accepting the first collision-free candidate.
Reuse the established grid-change bubble renderer for the full-screen launch reveal while leaving the app icon static.
Hide local population information when the underlying entry does not provide an accurate figure, and use curated interesting facts as the dependable alternative.
Define orientation as three groups: horizontal, vertical, and both diagonal directions combined.
Allow integer orientation counts to differ by at most one; allocate a remainder to diagonal first and horizontal second, never vertical.
Treat an orientation mismatch as a hard failure, so a strong quality score cannot conceal a directional imbalance.
Keep the exhaustive catalogue audit as release evidence rather than a routine check, with smaller multi-seed tests guarding the rule continuously.
Visual checkpoints
Generated and edited images.
No images were generated or edited during this interval. Three screenshots served only as source evidence and are therefore excluded. The already-created equipment-room proposal was approved and promoted without altering the artwork.
No images were generated in this interval.
Next moves & open questions
Current position.
The approved equipment room, app-wide gem capsule, streamlined globe and reef maps, bubble launch transition, conditional sticker details, enhanced Grid Runner, and explicit orientation rule are delivered and synchronized. The interface route has been checked in the simulator, and every grid in the full catalogue audit meets the new balance rule.