stickman/docs/tech_debt_and_optimizations.md

# Technical Debt & Optimization Log

## Overview

This document tracks known technical debt, optimization opportunities, and minor improvements for the Stickman Studio project. Items are prioritized by impact and effort. Update this log as issues are identified, resolved, or deprioritized.

---

## Current Observations

| #   | Issue                                                                                                                                                                                                         | Priority    | Status | Notes                                                                                                                                             |
| --- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------- | ------ | ------------------------------------------------------------------------------------------------------------------------------------------------- |
| 1   | **Ragdoll Neck Angular Limits** — The ragdoll neck `PinJoint2D` remains unrestricted, so a violent physical tumble can still rotate the head freely.                                                                                                  | Medium      | Open | Phase 4.5 fixes the separate ANIMATED collision-flinch accumulation bug by bounding offsets from a captured base rotation; it does not change ragdoll neck limits. Consider a tuned angular band/stiffness when natural ragdoll head motion is revisited. |
| 2   | **Torso Capsule Origin** — The torso capsule's local origin should be at its midpoint for natural rotation. Currently derived from bone distance; verify alignment.                                           | Low         | Open   | Test by rotating the torso body in ragdoll mode — it should spin about its center, not its top. Adjust `position` offset if needed.               |
| 3   | **Collision Layers Separation** — Bodies and terrain share layer 1/mask 1. This may cause self‑collision issues (limbs clipping through each other) under high stress.                                        | Low         | Open   | Future enhancement: assign ragdoll limbs to layer 2, terrain to layer 1, and use masks to allow limb‑limb collision only where desired.           |
| 4   | **Performance (Ragdoll Pooling)** — Spawning 10 bodies + 9 joints procedurally is fine for a single rig. If the scene ever contains dozens of ragdolls, consider a pooling system to avoid allocation spikes. | Low         | Open   | Not needed now, but worth noting if scaling to large crowds.                                                                                      |
| 5   | **Animated Collision Proxy Approximation** — Stickman collision proxies use fixed-radius capsules for torso and limbs rather than exact mounted-shape polygons.                                                                                             | Low         | Open | Phase 4.5 synchronizes ten proxies to the rendered parts and grows the head circle to cover mounted geometry (minimum radius 90 px), but torso/limb capsule radii remain approximate and do not adapt to arbitrary custom widths. Consider per-shape collision generation only if visible misses justify the complexity. (2026-09-12) |
| 6   | **Recovery Animation Starting Pose** — The `stand_up` animation must work from any captured ragdoll pose. Currently uses a fixed start frame.                                                                 | High        | ✅ Resolved | Phase 11: `_start_recovery()` captures the 10 bodies' rig‑local pose, snap‑solves the skeleton via the 6 IK targets, then `_play_stand_up()` tweens the markers **directly** from the captured values to `STAND_POSE` (`STAND_UP_DURATION`, sine ease‑in‑out) — the baked `stand_up` animation is **not** played (a fixed first keyframe can never match an arbitrary rest pose; the earlier bridge‑into‑the‑animation approach caused a visible jump and was removed). Revision: the snap now derives the hip (`pos − dir·half`) and wrist/ankle (`pos + dir·half`) from the capsule ends and subtracts the Torso bone's `bone_angle` for the marker rotation, so recovery starts from the ragdoll's exact final pose (e.g. sitting stays sitting). (2026‑08‑27) |
| 7   | **Transition Visual Pop** — The crossfade between kinematic and ragdoll currently uses a simple `modulate.a` lerp. This may cause ghosting if the kinematic and ragdoll poses are misaligned.                 | Medium      | ✅ Resolved | Phase 11: the entry is now an **instant handoff** — the ragdoll is built from the **current solved bone positions** (`AnimationPlayer.stop(true)` keeps the pose), then `Body/*` is hidden and the IK stack disabled in the same call. The earlier opacity crossfade + pin‑softness ramp was **removed on director feedback** (it read as ghosting, since both poses are identical). Recovery snap‑solves the kinematic skeleton to the captured ragdoll pose before re‑showing `Body/*`, eliminating the pop on both directions. (2026‑08‑27) |
| 8   | **Rest Timeout UI** — The director can adjust `rest_timeout` via inspector, but there is no in‑world UI in the physics harness yet.                                                                           | Low         | ✅ Resolved | Phase 11: added a Rest `SpinBox` (0.1–10 s, step 0.1) to the harness UI that writes `_rig.rest_timeout` (runtime‑only), plus a "Recover Now" button → `_rig.request_recovery()`. (2026‑08‑27) |
| 9   | **Animation Generation DRY** — `create_walk.gd` is a standalone script. It should be merged into a unified `create_animations.gd` that also generates `stand_up` and idle animations.                         | Medium      | ✅ Resolved | Phase 11: `create_walk.gd` deleted; new `scripts/create_animations.gd` (`@tool extends EditorScript`) bakes `walk_left`/`walk_right` (same keyframes) and a one‑shot `stand_up` into the scene's `AnimationLibrary` (the baked `stand_up` is an authored reference only — runtime recovery tweens the IK targets directly). (2026‑08‑27) |
| 10  | **Rig Collision Proxy Re‑addition** — The proxy is re‑added on ragdoll exit, but may cause a brief visual pop if it appears while the kinematic rig is visible.                                               | Low         | Open   | Phase 11 still re‑adds the proxy as soon as `RECOVERING` begins (`state_changed` handler), while `Body/*` is already visible — the static box can pop in around the standing figure before the stand‑up completes. Consider delaying re‑addition until after recovery finishes (`ANIMATED`). (2026‑08‑27) |
| 11  | **Stage Freeze Abstraction** — Sandbox Stage EDIT‑mode freezing is type‑specific: `RigidBody2D.freeze_mode = FREEZE_MODE_KINEMATIC` for props, `StickmanRig.set_ragdoll(false)` for stickmen, nothing for `StaticBody2D` terrain. There is no unified "freeze" abstraction over the mixed physics population. | Low         | Open   | A future physics type (e.g. `Area2D`‑based sensors) will need another case in `scripts/sandbox_stage.gd` `_enter_edit_mode()` / `_enter_play_mode()`. Consider a duck‑typed `set_simulating(bool)` interface once more physical object kinds appear. (2026‑08‑27) |
| 12  | **Stage AABB Selection Precision** — `StageSelection.get_world_aabb` uses conservative world‑space AABBs (polygon point union / fixed rig rect), not point‑in‑polygon.                                          | Low         | Open   | Clicks in the bounding‑box corners of large or rotated terrain may select a block even outside its polygon, and overlapping blocks can mis‑select. Refine with `Geometry2D.is_point_in_polygon()` for `TerrainBlock`/`PropBlock` polygons (and circle distance for ball props) once selection precision matters. (2026‑08‑27) |
| 13  | **Slope-aware walking physics & dynamic obstacle avoidance (deferred)** — Phase 3a bakes a real navigation mesh (per-`TerrainBlock` polygon decomposition into a code-built `NavigationRegion2D`) and `walk_to` follows `NavigationAgent2D` paths, but the figure walks the path with an upright pose (no tilt to the slope, no physics sliding) and `avoidance_enabled = false`, so it can path through props and other stickmen. | Medium      | Open   | A stickman crossing a ramp/stair follows the sloped footprint but looks flat-footed, and does not avoid moving props or each other. Future: tilt/rotate the figure to the path slope and enable RVO avoidance (`avoidance_enabled`, avoidance layers, `velocity` handling) once props/other stickmen are registered as obstacles. (2026‑08‑29) |
| 14  | **Phase 4 event engine is O(rigs×props + areas×movables) per physics frame** — `sandbox_stage.gd` `_update_area_entry()` / `_update_stickman_prop_collision()` run all-pairs geometric tests every physics frame in PLAY. | Low         | Open   | Fine at sandbox scale, but degrades quadratically with dozens of dynamic objects. A cell-keyed spatial broadphase (e.g. `_grid_cells`) can cull candidate pairs before the visible-AABB/feet-point tests. Phase 4b implemented that index (see #16, now Resolved) but the event engine does **not yet query it** — #14 remains Open until `_update_area_entry()` / `_update_stickman_prop_collision()` use it. (2026‑08‑30 / note updated 2026‑09‑02) |
| 15  | **Walk arrival jitter (mode re-evaluated per frame + arrival-radius mismatch)** — `StickmanRig._update_walking()` re-computes `_walk_mode` from `is_target_reachable()` every physics frame, and finishes via `is_navigation_finished()` (`NAV_TARGET_DESIRED_DISTANCE` 12 px, feet-space) *or* the `ARRIVE_DISTANCE` 8 px root-space guard. Near the nav-mesh boundary (clicked waypoints often sit just off placed terrain) `is_target_reachable()` can flip, swapping `root_target` between the nav path point and the raw waypoint — two points with a small vertical offset → up/down jitter. | Medium      | ✅ Resolved | Phase 4b: `StickmanRig._update_walking()` now **latches `_walk_mode` once per walk** (probes up to `LATCH_PROBE_MAX_FRAMES` = 6 after the map syncs, latching `"nav"` when reachable or `"direct"` on the bound) so it can never flip between frames, unifies arrival on the **final target** at `ARRIVE_DISTANCE` 8 px (snap-on-arrive), steers straight at the final target when within `2 × ARRIVE_DISTANCE`, and re-asserts the standing markers one extra physics frame after stop (`_settle_walk_markers()`) to clear any residual body-bob. Exactly one `arrive` fires; `global_position` is unchanged after arrival. See `plans/PHASE_4b_SPEC.md` §3. (2026-09-02) |
| 16  | **No grid spatial dictionary** — terrain placement (`sandbox_stage.gd` `_place_at`) is single-click with no occupancy tracking; the 3-state "empty/same-type/blocked" drag-paint query and any spatial broadphase need a cell→nodes index. | Low         | ✅ Resolved | Phase 4b adds an **advisory** grid spatial dictionary `_grid_cells` (cell `Vector2i` @ `TERRAIN_GRID_SIZE` 16 → `Array[Node2D]`) plus `_rasterize_aabb_to_cells()` to index every world AABB's covered cells; it drives the terrain drag-painting 3-state ghost query and the director target-validity test (skipping cells whose nodes include a `TerrainBlock`). It is populated on place, rebuilt on move/rotate/delete, and is **never authoritative** (the `World` tree is). It is **not yet wired into the Phase 4 event engine** — #14 stays Open for that. See `plans/PHASE_4b_SPEC.md` §2.6. (2026-09-02) |
| 17  | **Thumbnail caching has no eviction / cap; render is deferred one-per-frame** — Phase 3b (`stickman_library.gd` / `prop_library.gd` + `thumbnails/*`) caches rig/prop thumbnails to `user://thumbnails/` keyed by stickman basename+mtime and prop `id_v<PROP_VERSION>`. | Low         | Open   | PNGs accumulate unboundedly on disk (only `clean_stale_stickmen` prunes superseded basenames; nothing caps total bytes), and in-headless the renderers return `null` → a placeholder is shown until a **manual F6 run** populates real captures. Future: a disk-size/LRU eviction policy, a version/cleanup sweep on stage open, and an explicit cache-warm pass (or skip-the-placeholder note) for headless/CI. (2026-09-03) |
| 18  | **Asset selection is session-only, not saved** — `StageSpawner.selected_stickman_path` / `selected_prop_id` reset on scene reload (Phase 3b decision, per spec). | Low         | Open   | Deliberate for Phase 3b (spec §2 decision 1: no disk save, do not extend `_save_settings`). A future persistence phase could persist the last-chosen stickman/prop to `user://sandbox_settings.json` for convenience. (2026-09-03) |
| 19  | **Phase 3c editor-popup host-visibility discipline is ad-hoc** — the new `QueuePanel` / `RulePanel` / `ActionEditor` / `RuleEditor` are `exclusive = true` `PopupPanel`s, so the stage must hide a host popup before entering a stage-click capture (or opening a nested editor) and re-show it on resolve/cancel. That logic is spread across `_open_action_editor`, `_on_action_editor_committed` / `_cancelled`, `_begin_walk_edit`, and `_on_rule_editor_trigger_target_requested`, with one-off `_restore_queue_panel` / `_restore_rule_panel` / `_restore_rule_build_panel` helpers. | Medium | Open | Works, but fragile: any future editor popup must remember to hide its host before every capture or the exclusive window swallows the stage click. A small popup-stack abstraction (push host → capture → pop) would make that impossible to forget. (2026-09-04) |
| 20  | **Rule Panel "Add Rule" has no source in waypoint-filtered panels** — `_on_rule_panel_add_requested` bails with a toast when `_rule_panel_source_id < 0`, which is exactly the state of the waypoint-filtered "Edit Trigger Rules" panel (opened from the waypoint context menu). | Low | Open | The trigger source is genuinely ambiguous there (a waypoint can be targeted by rules authored by several stickmen). Acceptable for now; a future pass could default the source to `_ctx_waypoint_rig` (the rig that owns the waypoint) or open the builder in "any stickman" (source −1) mode. (2026-09-04) |
| 21  | **Phase 3c drag-reorder uses nearest-row-center, not an insertion point** — `QueuePanel` / `RulePanel` reorder computes the drop target as the row whose vertical center is nearest the mouse, then moves the dragged item to that index. | Low | Open | Functional, but the drop position can read off-by-one near row boundaries (it snaps to a whole row rather than an edge between rows). A true insertion-point indicator (a line drawn between rows) would be clearer; not worth the complexity until reordering long queues/rules is common. (2026-09-04) |
| 22  | **`--check-only --script` needs a prior `--editor` rescan for new `class_name` scripts** — a freshly added `class_name` (e.g. `QueuePanel`) is not resolvable by `--check-only --script res://scripts/sandbox_stage.gd` until `--headless --editor --quit` regenerates `.godot/global_script_class_cache.cfg`. | Low | Open | Verification-workflow gotcha only (not a runtime bug): in-editor the cache is always current. CI/headless scripts that add new `class_name`s should run an `--editor` pass before per-script `--check-only`. (2026-09-04) |
| 23  | **Phase 3c confirmation dialog is popped over an exclusive panel** — Queue delete / Clear All (`_on_queue_panel_delete_requested` / `_on_queue_panel_clear_requested`) and Rule delete / Clear All (`_on_rule_panel_delete_requested` / `_on_rule_panel_clear_requested`) call the shared `ConfirmationDialog` (`_ask_confirm`) **without first hiding the `exclusive = true` `QueuePanel` / `RulePanel`**. | Low | Open | Distinct facet of #19 (host-visibility discipline): popping a second exclusive window over a visible exclusive panel logs a **non-fatal engine warning** and the confirmation may render **non-modal**. It still functions (the panel is re-shown after the callback) — cosmetic. A fix would hide the host panel around `_ask_confirm` and re-show it in the confirm callback (and on the dialog's cancel/hide). Deliberate non-debt note: plan §11.2's separate confirmation dialog for editing `ragdoll`/`recover` actions was implemented instead as a paramless `ActionEditor` pre-fill — recorded in `docs/phase_3c_editor_spec.md` §2 decision 5, not as a debt row. (2026-09-04) |
| 24  | **Sandbox trigger rules use session-only instance IDs** — `entered_area` source/target references and action targets are `Object.get_instance_id()` integers. | Low | Open | Correct for the current in-memory rule engine and shared by the builder, matcher, visualizer, cleanup, and action executor. Rules survive `EDIT ⇄ DIRECT ⇄ PLAY` but are not serialized; a future saved-stage format needs stable authored object IDs instead of runtime instance IDs. (2026‑09‑11) |
| 25  | **Contact direct-state timing fallback** — `PropBlock` may receive `body_entered` before a matching direct-state contact row is available. | Low | Open | `_on_body_entered()` uses the documented center-direction / prop-position fallback when the row is absent, so the signal remains safe but the normal and position are less precise on that edge. (2026-09-12) |
| 26  | **Impact speed is a coarse magnitude** — collision reactions currently scale from `prop.linear_velocity.length()`, not normal-relative velocity or an impulse computed from contact mass. | Low | Open | This keeps the signal contract simple and deterministic. Use the normal component and/or contact impulse if reaction strength needs to reflect glancing impacts accurately. (2026-09-12) |
| 27  | **Post-IK reaction timing is engine-order-sensitive** — bounded flinch offsets are applied at the end of `StickmanRig._physics_process()`, after the IK/head-pin work. | Low | Open | The current slot is verified and prevents cumulative rotation, but future changes to Skeleton2D modification ordering could overwrite or delay the visible reaction. Revisit only with a targeted engine-order regression. (2026-09-12) |
| 28  | **Airborne collision sweep cost scales per proxy** — Phase 6a tests the horizontal movement against each of the rig's ten synchronized collision proxies every airborne physics frame. | Low | Open | The per-proxy `PhysicsShapeQueryParameters2D.cast_motion()` checks reuse the existing proxy shapes and avoid a second body-size approximation; this is acceptable for the current sandbox scale. With many simultaneously falling rigs, the cost grows with `rigs × 10` casts per physics frame. Consider a combined query shape or broadphase candidate filter if airborne crowds become a target. (2026-09-17) |
| 29  | **Prop Editor multi-shape → single-shape `PropBlock` flattening** — Phase 9 (`.prop` v1.0) stores a full multi-shape `shapes[]` stack, but `PropFile.flatten_to_payload()` collapses it into the one payload `PropUtils.spawn_prop()` consumes (frontmost closed polygon → circle → first shape). Also: a circle whose vertices were hand-edited still flattens by an average-distance radius, and a non-NONE preset thumbnail ignores payload colors because `PropThumbnail` re-tints from the preset. | Low | Open | Deliberate contract until a multi-shape `PropBlock` composition (multiple `Polygon2D`/`Line2D` children + merged collision) exists; recorded per `specs/phase_9_prop_editor.md` §6. Custom `.prop` thumbnails are mtime-keyed (`ThumbnailCache.prop_file_key`/`prop_file_png`, `<basename>_<mtime>.png`), so re-edits invalidate the cached PNG; built-in template PNGs stay `id_v<PROP_VERSION>`-keyed. (2026-09-18) |
| 30  | **Prop Editor Round 2 interaction model** — vertex editing and shape transform are separated into two explicit canvas modes (`EditMode { TRANSFORM, POINTS }`, kid-facing "Move & Resize" / "Edit Points") driven by an always-visible toolbar toggle; "Create Point" inserts at the cursor's segment projection and flashes the new vertex (0.6 s) instead of inserting an invisible collinear midpoint. | Low | Open | Deliberate, user-approved design (`specs/phase_9_prop_editor_round2.md` §5): modes gate *input* only — all public math APIs stay mode-agnostic. "Create Point" stays outline-only per user decision; auto-begin-drag was rejected as modal/surprising. Revisit only if kids find the two modes confusing in practice. (2026-09-18) |
| 31  | **Prop Editor Round 3 chrome/autosave decisions** — exit-save skips the thumbnail render (fast, deterministic quit; the PNG is a cache nicety), autosave is debounced 1.0 s to `user://prop_editor_autosave.prop` via `PropFile.save_safe` (atomic tmp/bak swap), and all human-facing strings (mode labels/emoji, menu copy) live in the `prop_editor.gd` controller while the canvas stays string-free. | Low | Open | User-approved (`specs/phase_9_prop_editor_round3.md` §6). `save_safe` leaves a `.bak` window during the rename — acceptable on desktop, revisit if autosaves ever run on a filesystem without atomic rename. (2026-09-18) |
| 32  | **Prop Editor mode buttons drop keyboard focus** — `_make_mode_button` sets `focus_mode = FOCUS_NONE` because the Button `focus` stylebox draws after `pressed` and (with the themed `focus` = normal panel) masked the selected/accent state until the cursor moved away. | Low | Open | Round 4 fix (`specs/phase_9_prop_editor_round4.md`). Trade-off: the Arrange/Reshape toggle is mouse-only. Revisit if keyboard accessibility for the mode toggle becomes a requirement — a focus stylebox matching the pressed state (or a focus-visible-only border) would restore it. (2026-09-19) |
| 33  | **Prop File save-serialization zeroed every point (fixed, Round 5)** — `PropFile._serialize_shape` assigned an export-format `Array` of `{x,y}` dicts directly to a typed `PackedVector2Array`; Godot's implicit conversion cannot build `Vector2` from a dictionary and silently yielded `Vector2.ZERO` per element, so every `.prop` saved by Round 1-4 wrote all points as `(0,0)` (loading as invisible/degenerate geometry). | High | ✅ Resolved | Round 5: `_serialize_shape` now routes points through `_coerce_points`; regression assertions check point *values* through `to_file_dict` and the full round-trip, plus an end-to-end canvas→save→load test. The previously saved `props/eye.prop` and `props/box.prop` are unrecoverable (coordinates were zeroed at save time) and were re-authored. Lesson: any typed `Packed*Array` assignment from dictionary/JSON data needs a Variant-coercion helper. (2026-09-19) |
| 34  | **Prop Editor per-shape `reshaped` flag** — every shape carries `reshaped` (false by default; set by vertex insert/remove/drag and non-uniform circle scaling, cleared by `set_shape_radius`), which gates primitive-specific inspector fields (today: the circle Radius row, replaced by a muted "reshaped" note). | Low | Open | Round 6 (`specs/phase_9_prop_editor_round6.md`). Only the circle consumes the flag today; future primitive params (box W/H, plank length/thickness) should reuse it instead of adding per-type detection. An "un-reshape" affordance (restoring a clean primitive from its params) is a possible future feature. (2026-09-19) |
| 35  | **Prop Editor selection uses a two-tone halo** — after Round 6's fill-based contrast failed for the reported "shape in front of a white shape" case (the selection extends beyond the shape onto the background), every selection element (outline polyline, bbox, rotation dot, scale crosses, vertex handles) is drawn as a dark backing pass + bright core pass. | Low | Open | Round 7 (`specs/phase_9_prop_editor_round7.md`). The pair is visible on any background but doubles the selection draw calls; negligible at sandbox scale. If a custom selection color is ever wanted (e.g. per-user theme), it should replace the core pass while keeping the halo. (2026-09-19) |
| 36  | **Phase 9a anchor roles have no gameplay consumers yet** — `.prop` assets can author/query `grip`, `emitter`, `impact`, `mount`, and `custom`, but Phase 9a is Prop Editor-only. Secondary grips do not drive two-hand IK, and no role dispatches runtime effects, damage, mounting, or equip behavior. | Medium | Open | User-approved staging in `specs/phase_9a_prop_attach.md`: land a stable authoring/schema seam plus an isolated one-hand IK preview first. A separately approved future runtime phase must design custom-prop loading, equip/release ownership, two-hand IK, emitters, melee impacts, accessory mounting, throwing, and selector/stage persistence without assuming the editor preview is gameplay-authoritative. (2026-09-20) |
| 37  | **Prop Editor still has no document-wide undo/redo** — Phase 9a anchor edits use the existing dirty/autosave/recovery pipeline, but there is no `UndoRedo` history for either shapes or anchors. | Medium | Open | An anchor-only stack was explicitly rejected because Arrange/Reshape edits would remain non-undoable and the UI would be inconsistent. Future work should introduce one document-level command/history boundary covering shapes, anchors, material/name changes, and preset replacement together. (2026-09-20) |
| 38  | **Phase 9a preview does not establish runtime attachment parity** — the Prop Editor uses a private, non-physics, full-multi-shape rig preview with editor-only facing and z-order conventions. It intentionally proves anchor authoring and arm-IK evaluation, not Sandbox loading, collision, physics, mirroring, per-limb interleaving, or attachment lifecycle. | Medium | Open | A future runtime spec must independently validate the anchor transform against gameplay animation/physics and decide its visual representation, sockets or solved-wrist API, mirror strategy, collision/freeze behavior, per-limb draw order, release/ragdoll handling, and multi-shape fidelity. Do not treat preview acceptance tests as runtime tests. (2026-09-20) |
| 39  | **Phase 9b prop centering is geometric AABB centering, not physical center of mass** — `Center Prop` uses the union AABB of internal polygon/line/circle points and deliberately excludes outline expansion, anchors, material mass, and collision geometry. | Low | Open | This is deterministic and appropriate for the fixed authoring origin, but a future physics-pivot workflow may need a separately designed center-of-mass or collision-bounds command. Phase 9b keeps canonical `origin` fixed/serialized, hides it from editing, never auto-centers after edits, and uses an explicit-save warning heuristic of `max(1024 px, 2 × bounds diagonal)`. Document-wide undo remains tracked separately by #37. (2026-09-20) |
| 40  | **Phase 9c walking preview uses exclusive gait-or-manual ownership of the selected hand target** — the private AnimationPlayer owns the target while gait plays; beginning a cyan-target drag pauses the gait before pointer input takes ownership. | Low | Open | This deterministic policy prevents AnimationPlayer and manual IK from overwriting the same Marker2D in one frame. It means an author cannot add a live hand offset while the walk continues. If that becomes valuable, design a preview-only additive offset/blend layer rather than relying on process order. The gait remains an in-place, non-physics authoring preview and is not navigation/runtime parity (#38). (2026-09-21) |

---

## Suggested Future Improvements (Beyond Current Scope)

| Improvement                             | Description                                                                                                             | Status / Priority |
| --------------------------------------- | ----------------------------------------------------------------------------------------------------------------------- | ----------------- |
| **Soft Transition Blending**            | Add physical blending (joint stiffness ramp) to complement visual crossfade.                                            | ✅ Done, then removed — implemented with Phase 11, but the whole crossfade was removed on director feedback (ghosting); entry is now an instant handoff. |
| **Ragdoll Recovery Interruptibility**   | Allow the director to force a mode switch mid‑recovery (e.g., if they want the character to ragdoll again immediately). | ✅ Done   |
| **Multiple Rig Support**                | Ensure all state variables are instance‑specific (already true) and that the harness can manage multiple rigs.          | Future   |
| **Animation Blending (IK vs. Physics)** | Blend between the animation‑driven pose and the ragdoll pose during transition to prevent snapping.                     | Future   |
| **Save/Load for Ragdoll State**         | Save the current ragdoll pose to `.stk` (e.g., for storyboarding a fall).                                               | Future   |

---

## How to Use This Log

1. **New Issues:** Add a row with `#`, description, `Priority` (Low/Medium/High), `Status` (Open/In Progress/Resolved/Deprecated), and `Notes`.
2. **Resolving Issues:** Change `Status` to `Resolved` and add a note with the date and fix details.
3. **Deprioritizing:** If an issue is no longer relevant, mark it `Deprecated` with a brief reason.
4. **Quarterly Review:** Revisit this log every major milestone to reassess priorities.

---

## Change Log

| Date       | Change                                                         |
| ---------- | -------------------------------------------------------------- |
| 2026-08-26 | Initial creation — migrated observations from Phase 10 review. |
| 2026-08-27 | Phase 11 resolved #6 (recovery starting pose), #7 (transition visual pop), #8 (rest timeout UI), #9 (animation generation DRY); #10 (proxy re‑addition) remains open with updated scope. Later revision: stand‑up recovery switched from bridge‑into‑baked‑animation to a direct marker tween (captured pose → `STAND_POSE`), fixing a visible jump; baked `stand_up` kept as authored reference only. Second revision: ragdoll entry builds from the current solved bone positions (IK disabled only after the blend completes) and the recovery snap derives joint ends from capsule half‑heights with Torso `bone_angle` compensation, fixing the entry pose‑pop and the "recovery starts lying" bugs. Third revision: the entire crossfade/blend (`transition_duration`, `BlendDirection`, opacity fade, softness ramp) was **removed on director feedback** — entry is now an instant handoff (build at current pose → hide `Body/*` → disable IK in one call), since the ragdoll spawns at the identical pose and a fade only read as ghosting. |
| 2026-08-27 | Sandbox Stage Builder (Phase 2) added `scripts/sandbox_stage.gd` + `stage_spawner.gd` / `stage_selection.gd` / `stage_gizmos.gd` + `scenes/sandbox_stage.tscn`. Logged #11 (no unified freeze abstraction over the mixed `StaticBody2D` / `RigidBody2D` / `Node2D` population) and #12 (selection hit‑testing uses world‑space AABBs rather than point‑in‑polygon). |
| 2026-08-29 | Phase 3a (Core Director Functionality) spec written (`docs/phase_3a_spec.md`). Logged #13 (slope-aware walking physics + dynamic obstacle avoidance deferred — the nav mesh itself is built in 3a). Notable non-debt decisions recorded in the spec: Play mode now runs the director script instead of auto-ragdolling stickmen; `walk_to(target)` treats `target` as a feet/ground destination via `FOOT_OFFSET`, with the `NavigationAgent2D` child placed at the feet `(0,+385)` so it paths on the ground-level nav mesh. |
| 2026-08-29 | **`walk_to` stops-after-a-few-px bug fixed.** `_update_walking` (Phase 3a) now defers nav reads until `NavigationServer2D.map_get_iteration_id(...) != 0` (map-sync guard) and forces the path query via `get_next_path_position()` before any empty-path/finished check. **Follow-up (same day):** the initial "warn + finish in place" unreachable-target policy was itself reported as "stickman stands still with a waypoint" and was replaced by **hybrid nav/direct steering** — an on-mesh target follows the nav path (`_walk_mode = "nav"`), an off-mesh/unreachable target walks **straight to the clicked waypoint** (`_walk_mode = "direct"`, root target = waypoint + `FOOT_OFFSET`), with **no** `push_warning`; `_walk_path_grace` removed (map-sync guard + forced path query replace it); the debug trace now carries `mode=nav|direct`. Off-by-default `DEBUG_WALK` (`stickman_rig.gd`) / `DEBUG_STAGE` (`sandbox_stage.gd`) traces added. #13 remains **Open** (slope physics + RVO avoidance are still deferred; the fix only changes unreachable-target handling). Logged in `BUGS.md`; verified with a 44-assertion headless regression suite. |
| 2026-08-30 | Phase 4 (Triggers & Event System) implemented: `scripts/trigger_area.gd` (NEW placeable sensor), `sandbox_stage.gd` rule system (`_event_rules`, geometric event engine `_update_area_entry` / `_update_stickman_prop_collision`, rule-builder UI state machine, `_cleanup_rules_for_nodes`), `stickman_rig.gd` (`arrived` gains a `target` payload; new `enqueue_reactive`), `prop_block.gd` (`collided` physics signal), `stage_director_visuals.gd` rule visualization, `stage_spawner.gd` / `stage_selection.gd` `"area"` palette + duck-typed `get_area_rect`. Logged #14 (all-pairs event engine scales O(rigs×props + areas×movables); spatial hash suggested). |
| 2026-09-02 | Phase 4b (Polish) spec written (`plans/PHASE_4b_SPEC.md`). Logged #15 (walk-arrival jitter: `_walk_mode` re-evaluated per frame + arrival-radius mismatch) and #16 (no grid spatial dictionary; Phase 4b adds one for terrain drag-painting + event-engine broadphase). |
| 2026-09-02 | Phase 4b (Polish) implemented. Resolved **#15** (walk-mode latch once per walk, unified `ARRIVE_DISTANCE` arrival with snap-on-arrive, steer-to-final-when-close, one-frame marker re-assert) and **#16** (grid spatial dictionary `_grid_cells` @ `TERRAIN_GRID_SIZE` for the drag-painting 3-state query + director target-validity). #14 remains **Open** — the dictionary is not yet wired into the Phase 4 event engine (note updated). Added `res://sandbox_theme.json`, `scripts/stage_placement_overlay.gd`, the 3-segment mode switcher / status bar / badge / frame / cursors, and director tooltip/trajectory/reticle UX. |
| 2026-09-03 | Phase 3b (Asset Library) implemented (`docs/phase_3b_asset_grid_spec.md`): `scripts/stickman_library.gd` / `prop_library.gd`, `scripts/thumbnails/thumbnail_cache.gd` / `stickman_thumbnail.gd` / `prop_thumbnail.gd`, `scripts/asset_selector.gd` + `scenes/asset_selector.tscn`. The Stickman/Prop palette buttons now open visual selector grids (pagination 12/page, session-only selection, Browse/Refresh); `StageSpawner` registry becomes `ground/ramp/step/prop/stickman/area` (`crate`/`ball` removed) with `selected_*` session state + per-path stickman cache; `SandboxStage` owns the selector open/close flow, Esc priority, and the one-per-frame lazy thumbnail drain. Logged #17 (thumbnail cache growth / headless placeholder) and #18 (session-only selection). Verified with the new headless suite `tests/test_phase3b_library.gd` plus the updated `tests/test_phase4b1_fixes.gd`. |
| 2026-09-03 | **Selector UI bugfix round** (5 bugs, documented via `tests/test_phase3b_ui_fixes.gd` + docs in `docs/phase_3b_asset_grid_spec.md` / `README.md` / `AGENTS.md`): `AssetSelector.open()` drops its selection params (`selected_path`/`selected_id`) — no cell is pre-highlighted on open (selected stylebox removed); the selector re-centers on window resize (`size_changed` → `popup_centered()`); the stage adds a dim backdrop `_selector_dim` (`SELECTOR_DIM_ALPHA` 0.5) behind the grid; the selector's `popup_hide` routes to `_on_selector_cancelled()` (idempotency-guarded) so an outside-click un-presses the palette button; and the Direct-mode action popup opens **right of the clicked stickman** (`_world_to_screen` + 24 px) instead of at the cursor. Reviewed #17 and #18 — neither is obsolete (both concern thumbnail-cache growth/headless placeholders and session-only *persistence*, orthogonal to these UI fixes), so both remain **Open** unchanged; no duplicate rows introduced. |
| 2026-09-03 | **Rule-builder popup-anchor bugfix round** (`sandbox_stage.gd`, documented in `README.md` / `AGENTS.md`): the Phase 4 rule-builder context menus used to re-pop at the live mouse position on every re-open, so cycling "⚡ When…" → "⬅ Back to actions" → "When…" walked the menu down the screen. Now a **session anchor** records the first context menu's screen position (`_popup_anchor: Rect2i` / `_popup_anchor_set`; Direct first menu = right of the clicked stickman; rule-label edit entry = the click position) and all child popups reuse it via `_set_popup_anchor(rect)` / `_clear_popup_anchor()` / `_popup_anchor_rect()`, until cleared on confirm (`_finalize_rule`), cancel (`_cancel_rule_build`), or Direct-mode/flow exit (`_clear_director_pending`) — but **not** on `_reset_rule_builder()` (Back-to-actions reuses it). No numbered debt row described the old cursor-following behavior, so no row was flipped to Resolved and no duplicates introduced; recorded here in the Change Log only. |
| 2026-09-04 | Phase 3c (Editor Tools — Action & Rule Editing) implemented: `scripts/action_registry.gd` / `trigger_registry.gd` (extensible action/trigger templates), `scripts/action_editor.gd` / `rule_editor.gd` / `queue_panel.gd` / `rule_panel.gd` + `scenes/*.tscn` shells, `scripts/waypoint_context.gd`, plus integration in `sandbox_stage.gd` (unified `CaptureKind` target-capture, "Edit Queue…/Edit Rules…" + right-click stickman/waypoint context menus, confirmation dialog, consequence-only rule editor) and `stage_director_visuals.gd` (`hit_test_waypoint_action`, waypoint edit highlight). Logged #19 (ad-hoc popup host-visibility discipline), #20 ("Add Rule" has no source in waypoint-filtered panels), #21 (drag-reorder nearest-row-center), #22 (`--check-only --script` needs a prior `--editor` rescan), and later the same day #23 (Phase 3c confirmation dialog popped over an exclusive panel — the delete/clear-all confirm paths do not hide the `QueuePanel`/`RulePanel` first). Documented the plan-§11.2 deviation (paramless `ActionEditor` pre-fill instead of a separate ragdoll/recover confirmation dialog) in `docs/phase_3c_editor_spec.md` §2 decision 5 — a design decision, so no debt row. Static-verified via headless `--editor --quit` + per-script `--check-only`; runtime F6 flows (queue/rule panel edit-delete-reorder, waypoint context menu, visual walk edit, full vs consequence-only rule editor) still need manual verification. |
| 2026-09-11 | Phase 4c (Trigger Area Reliability) verified: `sandbox_stage.gd` `entered_area` now requires source + area instance IDs and detects visible world-AABB overlap for ANIMATED stickmen/unfrozen props, edge-triggered per area/movable pair. The prop enumeration scope bug was corrected. `tests/test_phase4c_area.gd` passed all 19 assertions, including transformed-area, leg-level overlap, wildcard/re-entry, speech consequence, stale-target, and deletion-cleanup coverage; editor verification confirmed the speech bubble fires for the configured source stickman. Logged #24 (session-only instance-ID rule references). |
| 2026-09-12 | Phase 4.5 collision bodies verified: ten explicitly synchronized `AnimatableBody2D` proxies replace the old feet-point stickman collision poll; `PropBlock` supplies world-space contact data with shape-cast CCD; stage routing preserves rig→prop rules with per-part pair debouncing; bounded quadrant-aware flinches and PLAY→DIRECT authored-pose restoration were verified. `tests/test_phase45_collision_bodies.gd` passed 223 assertions; Phase 4b (53), Phase 4c (19), and Phase 3c (96) remained green. Reconciled #1/#5 and retained #25–#27 as intentional implementation limitations. |
| 2026-09-17 | Phase 6a (Airborne Momentum, Procedural Air Pose, and Landing Absorption) implemented and user-verified. `StickmanRig` now carries signed ledge momentum with tuned exponential drag (`AIR_DRAG_X = 0.75`), swept horizontal proxy collision, additive free-fall pose with forward-before-lift arms, and height-driven landing absorption: falls below 80 px are soft; 80..900 px maps to 32..160 px torso compression and 270..520 px hand drop over 0.30 s. Independent foot contacts preserve ramps/stairs, while `LANDING_IMPACT` blocks primary queue progression and direct ragdoll remains an immediate preemption. `tests/test_phase6a_airborne_landing.gd` passed 24 assertions; `tests/test_phase4b_terrain.gd` passed 8 assertions. Logged #28 (per-proxy airborne sweep scaling). |
| 2026-09-20 | Phase 9a Prop Attach specification written and refined (`specs/phase_9a_prop_attach.md`). Approved architecture: append Attach to the existing Arrange/Reshape system; additive role-based `.prop` anchors with a fixed synthesized `origin`; and an isolated Prop Editor stickman preview that uses real arm IK while showing the full authored prop. All Sandbox/custom-prop loading, attachment, sockets, physics/collision, lifecycle, and runtime tests are explicitly deferred. Logged #36 (all gameplay consumers deferred), #37 (document-wide undo remains absent), and #38 (editor preview is not a runtime-parity guarantee). Existing #29 remains open independently because the current runtime still flattens multi-shape props. |
| 2026-09-20 | Phase 9b Prop Centering specification written (`specs/phase_9b_prop_centering.md`). Approved architecture: hide canonical origin from editable anchors while retaining it in `.prop` v1.0; draw a passive constant-screen-size `(0,0)` marker in all three modes; add an always-visible Center Prop document transaction; center the authoring camera on origin for startup/New/load/reset and preserve zoom after centering; and warn explicit saves with Center & Save / Save Anyway / Cancel when displacement exceeds `max(1024 px, 2 × bounds diagonal)`. Autosave stays silent and automatic recenter-on-edit is explicitly rejected. Logged #39 (AABB authoring center is not a future physics center of mass); #37 continues to track document-wide undo. |
| 2026-09-21 | Phase 9c Prop Editor UX specification written (`specs/phase_9c_prop_editor_ux.md`). Approved architecture: a single Front Pose / Walk Left / Walk Right preview-motion selector with dynamic Pause/Resume; manual arm IK pauses gait before taking target ownership; modal typed anchor rename; Show Other Markers controls origin plus unselected anchors; one global Snap source; theme-driven `#262C39` inspector; container-driven unclipped chrome; simplified full-figure preview with no blue border. The preview directly plays the proven in-place gait/mirror policy and remains isolated from navigation, physics, actions, and Sandbox state. Logged #40 (exclusive gait/manual hand-target ownership). |

---

_End of Log_