Project-M/Docs/Vault/06_Roadmap/Iteration_2026-07_CoopHades.md

title, date, tags, permalink

title	date	tags	permalink
Iteration 2026-07 — Co-op Hades Conversion (Trunk + Hub-ification + Depth + Blender)	2026-07-06	roadmap iteration coop-hades direction plan	gamevault/06-roadmap/iteration-2026-07-coophades

Iteration 2026-07 — Co-op Hades Conversion

Operator decision (2026-07-06, dev-loop iteration session): commit the Fix-the-Fight-in-Place trunk PLUS the full Co-op Hades direction — cut/mothball the tower-defense layer, base becomes a Hades-house hub, boons become mechanic-changers, kit gains a second ability, real node types, second boss, co-op revive. Depth folds in now (not after a playtest gate). Blender scope: hero props + decals. No time estimates (operator directive). Grounded by a 9-agent analysis workflow (wf_8afc6e5a-b9e, 9/9 clean) + a live end-to-end play tour with screenshots (Assets/Screenshots/iter_*.png).

Evidence base (why this direction)

• Live tour findings: half of all rooms (sub-slot X+1500) have no ground/colliders/decor; enemies interpenetrate into a blob and spawn converges on the player; HUD header reads "AT BASE" inside expedition rooms; boon offer contained a strictly-dominated pair (-25% vs -15% ability cooldown); portal beacon renders as a blown-out white egg; storage crate floats; save-persisted RunRuntime.WasAllReady=1 deadlocks the next session's launch (un-ready → re-ready is the only escape).
• Critique consensus: every reward is a flat stat mod (zero mechanic-changers/synergies); kit = 3 verbs/class with 1 distinct ability; run 2 ≡ run 1; threat is walk-out-able (boss slam escape needs 4.07 u/s vs 5.1 available; melee knockback stunlocks all trash; death = free heal + re-conscription); the boss is a 1.6× Charger with one attack. The code-side feel layer (cones, slash arcs, hit-stop, flashes) is STRONG — the gap is the asset/scene layer under it.
• Market: Windblown's chart-topping demo = 1 biome + 1 boss (loop is demo-sized already); Cult of the Lamb = the base↔run tie done right, Moonlighter = the cautionary tale; HoH2's retention lever = every run banks something; Synty-look ceiling proven by Death Scourges / RoR2 / Soulstone Survivors (dynamic-light-first, dark ambient).

Phase 0 — Stabilize (bugs found live, do first)

1. Commit the working-tree portal fix (incl. the _portalMat comment-swallow NRE repair applied 2026-07-06 — the NRE killed CombatFeedbackSystem every session and auto-paused the editor via Error Pause). Re-run 456 EditMode first.
2. WasAllReady save deadlock RETRACTED (2026-07-06): misdiagnosis caused by frozen-world polling (the execute_code main-thread-sleep hazard) — CycleDirectorSpawnSystem's restore builds RunRuntime fresh (HostSalt only, WasAllReady=0) and nothing persists it; the observed "deadlock" was a run that had already launched between MCP calls. Lesson reinforced: never sleep-poll; check isPaused first.
3. Lifecycle-aware HUD header (kill "AT BASE" in rooms; kill stale "choose your boon" after pick — the location-line switch had NO RoomExplore case, so the RoomReward text stuck through the loot window).
4. Ground the floating storage crate (root cause: GridOrigin.y=1 is the CC capsule-center plane — CellToWorld returns it; center-pivot meshes (Turret/Wall) look grounded there, base-pivot meshes (crate) float 1 u. Watch this on every future grid-placed base-pivot mesh); tame the portal-beacon HDR (2.6/3.4 bloomed to a white egg) + ground the beacon (its pos.y was the capsule plane too).

Phase 1 — Trunk (direction-agnostic physical/feel/teaching layer)

1. Rooms become places: fix the X+1500 void (ground + collider ring + decor for sub-slot 1); seed 3–6 collidable Environment-layer props per room so cover/LOS/Charger-baiting exist; readable room edge.
2. Enemy separation (server-side repulsion, MaxAlive=12 so O(n²) is fine) + soft player pushout; no standing inside the boss. Verify spawn ring staggering reads (spawn VFX).
3. Animation triage: Synty Sword/Bow Combat anim packs (shared Generic skeleton, no retarget) — death clips + corpse window, spawn, hit-react, ranged-fire, dash; creature locomotion for monsters (stop human jog on Husks).
4. Threat repair: poise byte (light hits stop stunlocking trash); boss second attack (re-enable its baked Charger lunge) + party-size HP scaling.
5. Live knobs: promote all Boss*/enemy/grace-timer knobs into the TuningConfig overlay + values-to-clipboard.
6. Teaching: RMB class ability + dash coach-marks; fix stale copy (StartingOre 50→90 etc.); run-failed banner.
7. Clarity swaps: distinct spit/pickup/harvest meshes (Synty interim), class weapon meshes on the hand bone, ground darkening + character rim/outline, point lights on projectiles (RoR2 trick).

Gate: one honest solo campaign to Victory + one 2-instance LAN lap with the overlay open.

Gate outcome (2026-07-07 operator playtest): trunk holds mechanically, but the verdict is "not cohesive, very static, visually boring — comparable games' environments are dynamic and interactable" + ability upgrades boring + the 07-07 harvest shrink/pop never showed on-screen (root cause later pinned same day: a swallowed [RuntimeInitializeOnLoadMethod] attribute — NOT the pre-flagged PTM-on-root risk, which live-verified correct). → Phase 1.5 (World Alive) inserted below; mechanic-changer boons PULLED FORWARD from Phase 3 (Phase 1.7). Explicit operator framing: make the game look better, not more "juice" — the code-side hit-feel layer stays as-is.

Slice B folded design (design-review wf_fd177263-b67, 2026-07-06 — 3 lenses done, 15 verify verdicts ALL isReal=true; synthesis died on session limit, folded by hand from the verified findings)

• B3 Dying window: server-only Dying{uint UntilTick} (TickUtil.NonZero, stamp ONCE on the 0-crossing in HealthApplyDamageSystem for EnemyTag; destroy after ~54 ticks via Temp ECB). At the mark: zero the REPLICATED AttackWindup, disable IsLunging, zero KnockbackState (else corpses keep live telegraphs). CONFIRMED: no AI pass self-health-gates today and EVERY count edge counts entities — gate the actor passes (EnemyAI/BossAI move, windup, contact) and make every count LIVING-only: RoomEnemyDirectorSystem :48/:68 (room clear + MaxAlive fit), ExpeditionObjective.Remaining, boss-summon cap, base-siege cleared checks, aggro snapshots, ProjectileDamageSystem :83-91 target snapshot + gamepad auto-aim (no Health>0 filter today — corpses become shields/aim magnets). Client: EnemyAnimationDriveSystem writes IsDead = Health.Current<=0 (param exists, never written); CombatFeedbackSystem fires the kill CRUNCH at the 0-crossing (today it fires at prune — would land ~1 s late), prune keeps only a cleanup puff; hide health bars at <=0.
• B1 separation: a pass INSIDE EnemyAISystem (sole-position-writer invariant; BossAISystem only has [UpdateAfter(EnemyAISystem)]). Pairwise among LIVING enemies, skip lunging/knocked/Dying receivers, boss immovable (others push out of its radius), displacement through the swept EnemyMoveUtil move (no wall-shoving). Player pushout radius SMALL (< melee range, ~HitRadius+0.35) or grunts can't strike + Spitter cornered-hold re-breaks.
• B2 poise: the windup-cancel is the knockback CONTINUE branch (EnemyAISystem :119-133 grunt, :234-248 charger). Threshold on the EXISTING KnockbackState.Speed channel (no per-kind data): below it, kb velocity nudges but the windup/lunge survives; at/above it, stagger as today. Make the finisher + Warrior cone stamp a higher kb Speed so the channel actually distinguishes (light 6 / heavy 9, threshold ~7).
• B4 boss lunge: BossState gains a server-only attack byte; lunge = reposition move (no unique damage — the slam stays the damage beat). Telegraph rides AttackWindup + the boss's already-baked IsLunging ghost bit; the client boss-ring override SKIPS the slam ring while IsLunging (draws the standard Charger lunge tell). The windup-elapse branch keys on the state byte (today it would deal SLAM AoE on any elapse).
• B5 party HP scale: in the RoomEnemyDirectorSystem boss branch (sole spawn site), scale Health by LIVING EXPEDITION players (RegionTag+Health>0), NOT RunParticipant (dead-respawned keep the tag at base).
• B6 run-failed banner: client-only; cache Charge + a wentInRun flag at launch; on the (in-run)->Staging transition with Charge unchanged -> "EXPEDITION FAILED" banner. NEVER key on Returning (1-tick transient, and the bank lands one tick after it). Launching->Staging aborts don't count (wentInRun gate).
• B7 fire-anim pulse: stateless absolute window derived from AbilityCooldown (the SwingActive idiom) — no cached edges (rollback/join-unsafe).
• Hygiene: bytes not enums in Bursted systems; no new [UpdateBefore/After] edges; ~54-tick window + stagger threshold + separation strength as live TuningConfig knobs. ⚠ Bursted query-set edits on an UNFOCUSED editor can leave a stale Burst binary — have the editor focused for the compile, or expect the restart cure.

Phase 1.5 — World Alive (inserted 2026-07-07 from playtest verdict; visual + dynamism, explicitly NOT juice)

Operator interactivity scope (all four selected + two write-ins): breakable clutter with drops · environmental hazards · cosmetic reactivity + ambient motion · destructible cover · collision precision · more distinct attacks/animations/effects. Art direction deliberately UNDECIDED — lighting pass first, then pick (see gate).

Stabilize first (bugs, do before the pass)

1. Node harvest feedback fix FIXED (2026-07-07 session B). Root cause was NOT the pre-flagged PTM-on-root risk — live inspection proved the Model child properly parented and root-PTM propagation exact (root PTM 0.783 → child LTW 1.723 = 2.2×0.783). The 07-07 knob edit had swallowed the [RuntimeInitializeOnLoadMethod] attribute off WorldFeelConfig.ResetDefaults → Enabled=false → the WHOLE feedback slice (shrink/pop + chips + SFX + micro-punch) silently dead. One-line restore; verified on-screen + numerically on a live server-spawned node; 455/455. Operator still eyeballs a real harvest in the next playtest.
2. Collision precision pass — SCOPED (2026-07-07 operator): environment collider FIDELITY. Collision shapes don't match the visual meshes (cover rocks etc.) — invisible walls / visual clipping. TODO: audit every Environment-layer collider (boundary rim, landmark rocks, in-room cover props) and fit them to the render mesh (tightened primitives or convex mesh colliders); verify against the baked subscene PhysicsWorld (that's where they live — classic-GameObject views lie). Re-validate the enemy stuck-fix backstop after any shape change (Depenetrate/tangent-slide was tuned on the old shapes).

Short-term wins (asset/scene/client layer)

3. Lighting/atmosphere pass (direction-agnostic): ambient way down, dynamic point lights on projectiles / portal / pickups / braziers / ability impacts, per-biome fog + color-grade tune. Build AS IF no pixel filter (it composes on top). Then the art-look decision gate: side-by-side screenshots pixel filter ON vs OFF over the lit scene → operator locks pixel-art-3D vs dark-Synty vs both. Every later visual choice hangs off this lock.
4. Ambient motion layer (client-only, zero netcode): foliage sway/rustle on walk-through, per-biome dust/embers/fog drift particles, critters that scatter, flags/banners. Biggest bang-per-effort for "not static".
5. Breakable clutter with drops: Hades-style urns/crates/Blight growths seeded per room; shatter on hit → small resource drops. Reuses the destructible-ghost + BlightClutter + pickup patterns — netcode-cheap.
6. Attack distinctness: every enemy kind + class ability gets a unique read — anim + VFX + effect shape (not more hit-stop). Builds on the shipped Blender clip pipeline + sword-combo work; creature locomotion for monsters still open from Phase 1.3.

Long-term (server-sim; adversarial design-review REQUIRED per slice)

7. Environmental hazards: exploding barrels, Blight geysers, trap tiles — damage BOTH sides so enemies can be baited into them. Real combat interplay; replication + predicted-damage surface needs the review.
8. Destructible cover: in-room cover rocks take damage and crumble (Charger smashes through, players carve LOS). Structures already have Health; re-validate the enemy stuck-fix backstop when cover can disappear mid-fight.

Gate: operator playtest says "no longer static"; node harvest feedback visibly works; art direction LOCKED.

Phase 1.7 — Boon overhaul (PULLED FORWARD from Phase 3, operator call 2026-07-07)

• Boons → ~12 mechanic-changers on existing hooks: dash damage trail, finisher AoE detonation, projectile pierce/fork/chain, ability-SWAP boons (replicated AbilityRef byte via the ClassSwapUtil pattern), knockback→pull… plus synergy tags, duplicate/dominated-offer protection. Design-review before coding (replicated ability swaps).

Gate: two consecutive runs feel different from boons alone.

Phase 2 — Hub-ification (structural; adversarial design-review REQUIRED before coding)

• Mothball, don't delete: sieges/CoreIntegrity/turret-ammo retired from the flow via baked toggles (the ScheduleEnabled pattern); code + tests keep compiling for reversibility.
• Win condition moves fully into the expedition (N boss-clears / final-boss Victory; GoalReachedSystem + RunOutcome rework; SaveData v7 additive).
• Base = hub: class/prep/meta/ready + deposit; visible growth (unlocks at Charge milestones); staging HUD → progressive disclosure (5 dense panels → contextual).
• Economy simplification (4 currencies is bureaucracy; turret-ammo Charge frees up) — scope at the design review.

Gate: returning player goes staging→launch in <60 s; hub reads as prep+growth, not friction.

Phase 3 — Depth (folded in now; design-review per netcode-heavy slice)

• Kit: second ability slot + one new active per class (reuse AbilityRef catalog + predicted-input review).
• Boons MOVED to Phase 1.7 (pulled forward 2026-07-07 — operator playtest called flat-stat upgrades boring).
• Real node types: Elite (affixed alphas via the boss spawn-stamp pattern), Shrine (combat-free choice), Survive, Cursed.
• Second boss chassis (Spitter artillery) so lap 2 differs.
• Co-op: downed/revive (ex-MC-5), duo boons, party-scaled spawns/HP; death-cost design lands here (with revive in, death = run stakes, killing the suicide-heal exploit).

Gate: two consecutive runs feel different (build + path + boss); a duo playtest yields a clutch-revive moment.

Phase 4 — Blender workstream (parallel from Phase 1; hero props + decals)

Portal gate ring · hub Core crystal centerpiece · boss bone-socket armor/spike kitbash (no new rig) + second-boss silhouette parts · distinct projectile/pickup/harvest meshes · class weapons · ground decal/telegraph quad set. Pipeline (validated + researched): UV new geometry onto the existing Synty atlas for style match (or Imphenzia palette PNG); FBX for anything skinned (glTFast Mecanim-clip issues), glb via gltfast for statics; Apply Scalings = FBX Units Scale; Key All Bones for clips; realize geo-nodes instances before export; Play-verify materials (dark-frame hazard).

Phase 5 — Dev tooling (woven through)

New DebugOps: KillRoom (death-path kill — ClearEnemies destroys without kill credit), SkipToRoom/SpawnBoss, GrantBoon, run-state line in the overlay · pause-proof screenshot rig (MCP bridge ScreenshotUtility calls EditorApplication.Step() and leaves the editor PAUSED — unpause after every capture; never Thread.Sleep-poll in execute_code, it freezes the sim) · encounter-composition buttons in DevSandbox · WasAllReady regression test.

Consolidation notes

Supersedes the DR-042 "final beat = base siege" lock (win moves fully expedition-side). DR-034 Core / EB-1/2 siege systems → mothballed, not deleted. The "present forks" ritual re-runs at each phase's design review.

2026-07-07 restructure (Phase 1 gate playtest): Phase 0 + Phase 1 SHIPPED (commits 7c1fee097…cc2e7ad95, incl. Blender clips, sword combo, pixel-art render style, enemy-stuck fix, markers, harvest feedback). Gate playtest verdict = static/boring world → Phase 1.5 World Alive inserted (stabilize: node-feedback fix + collision precision; short: lighting→art-look gate, ambient motion, breakables, attack distinctness; long: hazards + destructible cover, review-first) and boons pulled forward to Phase 1.7. Phase 2 (hub) now follows 1.7. Order: 1.5 → 1.7 → 2 → 3 → 4/5 parallel.