fix(threed): close v1.9 asset and runtime contracts - #56
Closed
wh000wh000 wants to merge 67 commits into
Closed
Conversation
Normalize copied Unicode spacing and mixed punctuation that rendered as garbled text. Add a repository-wide UTF-8 and mojibake regression check so the issue cannot silently return.
Add the frozen classifier/GDD contract for a minimal three.js archetype while preserving every existing Phaser route. Includes regression fixtures for explicit 3D, 2D platformer, and 2D top-down prompts.
Provide the core3d skeleton, threed_basic gameplay layer, pinned npm lockfile, and builder-facing manuals. The template keeps DOM lifecycle screens and the three-key ESC fallback while limiting assets to the existing image pipeline.
feat(3d): add threed_basic v1 production path
Keep the player on authored floor patches and outside explicit static obstacle circles without introducing a physics dependency. Substepped resolution prevents high-delta tunnelling and the self-check covers road, obstacle, and slide behavior.
Teach the builder to preserve the pure XZ resolver, explicit collision radii, and v1 no-physics boundary so generated games consume the new capability instead of bypassing it.
feat(templates): add threed_basic manual collision
The three.js renderer sizes from its container and never consumes the inherited 2D reference dimensions. Removing the unused leaves gives both standalone core3d and threed_basic a zero-dead-config baseline.
Generated v1.1 code duplicated fog, pickup, and count values. Make the builder keep one canonical path per runtime value, remove superseded aliases, and record the leaf-to-consumer map in GDD completion notes.
The universal GDD reminder contradicted threed_basic zero-dead-config rules by forbidding deletion of inherited fields. Keep the Phaser reminder unchanged while requiring consumed, single-source leaves for 3D.
fix(templates): enforce consumed 3D config
A generated v1.3 game shipped teleport and time mutation hooks solely to make verification easy. Freeze the existing probe surface and require completion evidence through real browser input so acceptance measures gameplay instead of a debug backdoor.
…bridge docs(templates): keep 3D smoke bridge read-only
A real v1.4 game showed repeated Escape events can toggle pause multiple times and end screens can resume gameplay. Ignore keyboard repeats and lock the runtime after completion or failure.
Real-browser review found the core3d shell requests an undeclared favicon. Ship a tiny standalone SVG and reference it relatively, keeping binary data out of editable HTML.
fix(templates): stabilize 3D browser shell
The star-lamp maze GDD weakened an explicit sequential objective into unordered collection to match the scaffold. Add a threed_basic-only fidelity rule so private GameScene state may implement the requested mechanic without widening the public API or changing Phaser prompts.
Document that ordered objectives remain ordered and should use minimal private GameScene state instead of being reinterpreted as the scaffold's unordered pickup loop.
…tics fix(gdd): preserve ordered 3D objectives
Prevent generated acceptance criteria from strengthening declared win conditions or relying on instrumented artifacts, so production evidence reflects the shipped game.
Keep real media out of generated source and session context, and require complete playthrough evidence from the final uninstrumented artifact.
…tegrity fix(gdd): keep 3D verification faithful
Prevent generation sessions from reading screenshots as Base64; automated checks use smoke and visible DOM while human review opens the reported file path.
docs(debug): keep screenshot evidence path-only
Align W movement with the three.js camera forward vector and make in-page restart dispose input listeners plus scene-owned GPU resources. The source-level self-check locks both contracts without adding runtime test hooks.
Phase 6 changes must correct stale values and consumers in the GDD body instead of leaving contradictory appendices. Document the restart resource-ownership contract exposed by the template fix.
The core shell still used the opposite Z sign for the camera-right vector. Lock both components in the runtime contract self-check so yawed movement cannot regress independently.
fix(templates): close threed v1 scaffold truth gaps
v1.9 production still named runtime state and texture behavior that did not exist. Require unstable implementation details to be copied from the final consumer or removed, so the GDD cannot invent a plausible second truth.
docs(threed): anchor final claims to source
Generated GDDs retained stale renderConfig instructions and claimed a camera-far consumer that did not exist. Require final values and a named runtime leaf so production reviews can reject those false claims.
docs(threed): require final render config truth
Mathematical overlap let generated routes pass while leaving player-radius corridors and finish triggers on collision boundaries. Define measurable clearance and target-center rules, and pin final config descriptions to their real consumers.
docs(threed): require stable route clearance
The 3D GDD producer hard-coded a single-route collectible plan, so generated games could only reskin movement and pickup. Require verbs, pressure, feedback, escalation, win/lose states, and runtime consumers while leaving Phaser guidance untouched.
The module manual defined threed_basic as collect-everything on one route, which prevented the builder from planning distinct play. Define the minimal core-loop contract and allow only semantic SceneMap arrays and path-referenced image keys with real consumers.
A mandatory lose state would contradict peaceful game requests. Keep the gameplay contract strict while allowing an explicitly declared no-fail tension and progression response.
…ract fix(gdd): require real 3D core gameplay loops
Repeated generated games mislabeled camera far as config-backed and proved only one side of a claimed route choice. Put those checks in the shared GDD producer so later games must ground camera ownership and branch coverage in final code.
The first non-collection game claimed a committed fast/safe choice while only exercising one route. Require intended and forbidden adjacency checks plus final play coverage for every branch.
fix(threed): ground route decisions in final gameplay
A prose-only core loop can combine individually plausible timings and resource values into an unwinnable game. Require winning and losing traces with derived deadlines, state deltas, and authored-event closure before a 3D design is called playable.
Make the builder derive route timing and state deltas from final authored data, then reconcile event totals and terminal states after tuning. This keeps build and smoke from standing in for actual playability.
…dger fix(gdd): require executable 3D gameplay budgets
Prevent inferred or substitute play traces from being reported as exact-strategy acceptance evidence. Preserve the existing no-op behavior for 2D archetypes.
Keep feasibility evidence honest across arbitrary 3D mechanics by separating derived traces from exact observed playthroughs and carrying state through escalation thresholds.
Document the correct package-relative focused-test invocation so future loops do not mistake workspace broadcast failures for product regressions.
…nance fix(gdd): preserve exact playthrough provenance
fix(docs): remove corrupted text characters
Consolidate the accepted cross-repository requirements, minimal architecture, dependency choices, rejected alternatives, ten-step delivery plan, and production acceptance gates so the next integration work starts from one versioned source of truth.
docs(pixelforge): add integration execution pack
Generated 3D games can make SceneMap array lengths and config count leaves co-own the same value while using the config only in an equality assertion. State the single-truth rule at GDD production so counts remain authored data unless gameplay consumes a separate threshold.
Capture the repeated failure to keep apply_patch operations file-scoped so future work checks for multiple Update File headers before applying.
fix(core): prevent mirrored authored count config
Create the DOM HUD before GameScene emits its constructor-time progress event. This prevents generated games from overwriting a real initial total with the UIScene default and locks the ordering with the existing runtime contract self-check.
Document the repeated multi-file apply_patch mistake so future goal turns enforce the existing one-file edit rule mechanically.
Make the ordering guard mutation-honest by requiring both UIScene.show and GameScene construction to exist in the expected order; deleting either anchor can no longer pass through indexOf(-1).
…ogress fix(core3d): preserve initial HUD progress
Author
|
Superseded by #57, which rebases the complete threed_basic v1 milestone onto current origin/main and excludes unrelated fork-only changes. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Cumulative upstream proposal for the production-driven threed_basic v1 through the v1.9 closure wave. It supersedes #55, keeps the Phaser 2D route unchanged, and now includes the real-product HUD initialization defect found in sample R.
Frozen-layer files versus upstream main
Other cumulative files explicitly present in this head
These are not hidden inside the 3D list: they are already-reviewed text cleanup and PixelForge integration preparation carried by fork main.
Latest review focus
Verify that UIScene.show precedes GameScene construction, the mutation-honest self-check requires both anchors, initial progress is no longer overwritten, image normalization remains at the common save seam, GDD rules stay mechanism-neutral, and no Phaser file or runtime path changes.