Interval
Work since the previous summary
The work moved through three linked quality passes. First, reef-map callouts and the treasure reveal were rebuilt around hard visual rules. Second, a screenshot exposed incorrect globe geography, leading to a review of every reef anchor and independent label positioning. Finally, the treasure animation and grid variety were refined while a separate codebase review identified performance and stability risks that remain candidates for future work.
Direction of travel
How the work evolved.
Each step connects the user’s prompt to the practical change it produced.
Constrained
Made reef-map callouts measurable
Prompt: Keep pins within 200 pixels of their geographic dot, centre names beneath the dot within 50 pixels, and prevent pins, titles, level graphics, and neighbouring callouts from overlapping.
Outcome: The map layout now targets a maximum 198-pixel pin distance and a 48-pixel title distance on a high-density phone display. Titles centre on their own dot, and collision checks include the complete pin, play-mode graphics, label, and geographic point rather than testing each element in isolation.
Rebuilt
Turned the treasure reveal into a seabed finale
Prompt: Place the chest on the seabed from the beginning, animate only the lid, remove doubloons, and use production-ready gems with varied colours and orientations.
Outcome: The chest is now settled from the first frame and only its lid opens. Five faceted gem treatments—cyan, green, red-purple, amber, and aquamarine—use deterministic scale, mirroring, foreshortening, and rotation, with no doubloon path or falling chest animation.
Corrected
Placed every globe marker over its reef
Prompt: Move the Caribbean padlock to the Caribbean, verify every reef icon against its real location, separate colliding name plates, and centre all label text inside its box.
Outcome: All seven artwork anchors were reviewed and corrected. Geographic icons remain fixed over their reef while label plates can move independently; the Caribbean and Mesoamerican names use opposing offsets, and both one-line and multiline labels are centred horizontally and vertically.
Refined
Finished the gem landing and resurface placement
Prompt: Animate every gem down to the seabed near the chest and place the Resurface control 150 pixels above it.
Outcome: Each gem now follows an individual arc and settles near the base of the chest. A simulator capture exposed an overlap with the open lid, so placement was recalculated against the actual chest bounds and rechecked with a genuine 150-pixel clearance.
Varied
Reduced recognition of repeated grid patterns
Prompt: Randomise the target-word display order, avoid repeatedly using the same rows, columns, and diagonal starts across a site campaign, retain the one-third orientation rule, and remove the old edge-start restriction.
Outcome: Target capsules now use a stable seeded shuffle, and generation carries placement-line history across levels so otherwise valid layouts prefer less-used rows, columns, and diagonals. Edge starts and endpoints no longer affect generation or scoring, while complete straight-edge words remain discouraged and the exact orientation quota remains enforced.
Reviewed
Audited the product for performance and stability
Prompt: Run an independent whole-codebase review focused on performance and stability while the gameplay changes are implemented.
Outcome: The review found three high-priority risks involving concurrent campaign starts, cancellation and prioritisation of grid preparation, and silent progression resets from corrupt or unsupported saves. It also identified several medium-priority allocation, main-thread, image-memory, and fallback concerns.
Delivered
Consolidated the verified work
Prompt: Bring the completed map, treasure, grid-generation, launch, and room changes together in the shared product state.
Outcome: The complete verified change set was synchronized cleanly. Private delivery details are intentionally omitted from this record.
What improved
Improvements made.
Evidence-led changes grouped around the product, implementation, design and working process.
Pins stay close to the geography they describe
Hard distance limits replace loosely weighted positioning, keeping each pin within 198 pixels of its source point while preserving the point itself as visible map evidence.
Location names are easier to associate
Names sit centrally below their geographic point and within 48 pixels. The layout tests the label together with its marker and play graphics, reducing ambiguous callouts in crowded regions.
Crowded maps have room to breathe
When existing regional canvases could not satisfy the hard distance and collision rules, the horizontally navigable map area was widened instead of weakening the constraints or moving labels away from the correct geography.
Globe markers now respect real locations
The Caribbean marker moved out of the mid-Atlantic, every reef anchor received the same review, and icon placement was separated from label placement so geographic accuracy no longer competes with name-plate legibility.
Treasure feels physically grounded
The chest is present on the seabed before the reveal, only the lid animates, and the reward gems finish around its base rather than floating or behaving like falling coins.
Gem rewards have clearer visual variety
Five colour families retain faceted highlights while varied rotation, reflection, size, and perspective prevent the reward set from reading as repeated blue copies.
Target lists no longer reveal placement order
A deterministic shuffle separates the displayed target sequence from grid construction order, preserving repeatability for testing without exposing a visual pattern to players.
Campaign grids reuse fewer placement lines
Generation tracks prior rows, columns, and diagonal starts within the campaign and favours underused lines among otherwise valid candidates, reducing easy pattern recognition between successive levels.
Grid scoring reflects the current rules
The obsolete edge-start preference was removed from generation and Grid Runner scoring. The separate quality rule discouraging entire words along an outer edge remains intact.
Pain points
Friction, constraints and open edges.
Resolved problems are separated from constraints and work that remains open.
The original map canvases were too narrow
Several regions could not place five complete callout groups within the new 200-pixel, 50-pixel, and non-overlap limits. Widening the navigable panorama preserved both geographic truth and the requested constraints.
The first chest replacement did not land
After an earlier change attempt, the old entrance, treasure tumble, and doubloon behaviour was still active. The finale component was replaced directly and covered with assertions for settled placement, lid-only motion, gem variety, and safe positioning.
Gem colours initially retained too much cyan
The red, purple, and amber variants were visually pulled toward the source gem’s blue tone. Stronger colour treatment and additional mirroring and foreshortening made the five reward families distinct while retaining the original facets.
The first globe capture was centred on the wrong view
An initial verification image placed the Atlantic labels near the globe edge because the test camera was centred too far into the Pacific. The view was recalibrated before visual approval so clipped evidence was not mistaken for a product defect.
The resurface control still overlapped the open lid
Automated checks passed before the simulator view showed that the control was not truly clear of the chest. Its position was recalculated from rendered chest geometry and visually confirmed at the requested clearance.
Three expensive generation checks remain unstable
A complete run passed 172 of 175 checks. Three pre-existing word-pool generation cases produced 69 or 71 unique targets instead of 81; the map, treasure, and interface checks all passed, but the generation shortfall remains unresolved.
Three stability risks need dedicated fixes
Rapid interaction can begin concurrent campaigns, background grid preparation lacks prompt cancellation and prioritisation, and corrupt or unsupported saved progress may reset silently. The review identified these risks but did not change their behaviour in this interval.
Outputs & proof
What was delivered and verified.
The new geometry and treasure rules were exercised in focused tests and in live simulator flows. The explicit orientation split was rechecked in both profiles, while a complete suite exposed three unrelated generation shortfalls and the independent review documented the principal remaining stability risks.
Verification record
- Every authored regional map passed maximum pin distance, maximum title distance, non-overlap, and unobscured-dot checks.
- Focused globe tests passed for all seven reviewed anchors, preserved icon centres, separated Caribbean and Mesoamerican plates, and horizontal and vertical text centring.
- A live Atlantic globe check showed the Caribbean and Mesoamerican icons over their intended reef areas with readable, separated labels.
- Treasure checks passed for a settled chest, lid-only reveal, no doubloons, five-colour cycling, varied rotation, and safe gem placement.
- A live phone simulator confirmed the map geometry, seabed chest, varied gem appearance, individual gem landing arcs, and 150-pixel resurface clearance.
- Grid Runner reconfirmed the explicit split in both profiles: 27 of 81 horizontal placements, 25 of 81 vertical placements, and 29 of 81 diagonal placements.
- Focused gameplay, Grid Runner, and treasure tests passed after the final changes.
- The complete suite passed 172 of 175 checks; the three remaining failures are the documented word-pool generation shortfalls.
Key decisions
Measure callout distance from the geographic dot itself, not from a displaced marker or label.
Treat the pin, play-mode graphics, label, and geographic dot as one collision problem.
Widen horizontally navigable regional maps when density makes the requested geometry impossible, rather than weaken the rules.
Keep globe icons anchored to reviewed reef coordinates while allowing name plates to move independently for legibility.
Reuse and strengthen the established faceted gem art instead of introducing a conflicting visual style.
Place the chest from the first frame, animate only the lid, and let each gem settle naturally near its base.
Use a seeded target-list shuffle so presentation varies while tests and saved campaigns remain reproducible.
Carry placement-line history across levels and use it as a preference among valid layouts, not as a rule that can prevent generation.
Remove edge starts and endpoints from scoring while continuing to discourage complete straight-edge words.
Keep the independent performance and stability findings as explicit follow-up work rather than expanding this visual and gameplay pass into unrelated architectural changes.
Visual checkpoints
Generated and edited images.
One simulator capture was created to verify the corrected treasure finale. Its temporary source was no longer available when this record was rendered, so the visual checkpoint is represented by a privacy-safe placeholder. User-supplied screenshots were source evidence and are intentionally excluded.
Image omitted because it was unavailable or not safe to store.
Next moves & open questions
Current position.
Reef maps now satisfy explicit proximity and collision rules, globe markers align with all seven reviewed reef locations, and the treasure finale presents a grounded chest with varied gems that settle around its base. Target display order and cross-level placement variety are improved without weakening the three-way orientation rule. Three high-priority stability risks and three expensive generation-test failures remain documented for focused follow-up.