Need help getting started? Try the player guide. These notes include older work; the game overview explains what’s ready now.

Weapon 7 conversion and periodic drain

One-way Prion melee (2026-10-22)

Normal enemies cannot melee Prion-infected enemies. Prion-infected attackers can still melee normal enemies and other infected enemies. The contact query rejects normal attackers, and the combat melee bridge also rejects a normal attacker's direct melee event against an infected enemy. Both CD contact and other enemy controllers share the combat gate.

Accepted melee hits use original damage, flinch/hit latches and death reactions; the earlier no-flinch change is reverted. Enemy targeting and ranged attacks are unchanged, as is combat against the player. The targeting fixture checks all conversion combinations, both collision paths, normal ranged retaliation, existing hit latches and lethal melee.

Healthy enemy targeting (2026-10-21)

The existing frontend nearest_converted policy now also selects targets for healthy enemies, including both bosses. They compare living players with living Prion-affected enemies using the same 3D body-center distance; other healthy enemies are excluded. Converted actors still consider every living combatant. Selection refreshes each behavior pass, without waiting for the old retarget timer. Equal distances retain active-pool order with no explicit player bias.

The selected actor drives facing, pursuit, attack checks and projectile direction. The opt-in prion_targets controller field permits an unconverted ranged actor to fire when its aim trace reaches its selected Prion target. Ordinary allies do not satisfy that new attack gate. Both boss projectile paths use the selected actor too. If a charging CE contacts Sam while pursuing a Prion target, knockback still applies to the actual player contact. The temporary controller player projection is restored after each pass. Damage, conversion health, converted behavior speed and ranged tactics retain their existing policies below.

The targeting regression covers healthy actors choosing and switching between Sam and Prion enemies, healthy ally/dead/corpse exclusion, vertical distance, multiple converted candidates, no live targets, all six ordinary ranged types, both bosses firing away from Sam toward a converted target, and the policy-off original facing behavior. No campaign playthrough is claimed.

Validation: 8,083 targeting/combat assertions over 662 frames, including 114 healthy-enemy cases and 14 ranged/boss shots. The default-off controller suites retain 27,738 ROM comparisons; all eight focused CTests and the 415 boss self-projectile checks pass. The Windows frontend and all 128 libraries rebuild.

PC Prion balance (2026-10-21)

The playable adapter enables combat.prion_balance. A successful first Prion conversion doubles the victim's current remaining health, once, before E2 allocation. Positive health saturates at32767; dead actors are not revived, and another hit on an already converted actor does not boost it again. The original conversion flag, drain period/countdown and effect-allocation failure semantics remain intact. The recovered default-off runtime preserves ROM behavior.

The subsequent user correction removes the conversion-based damage multipliers. Melee, projectiles and explosions now retain their normal damage regardless of whether either enemy is Prion-affected. Contact deals20 normally,40 for D0/D6; there is no half-melee or double-other-damage adjustment. This also applies when both enemies are converted. Original splash attenuation, difficulty scaling, player combat and boss self-projectile immunity remain intact. The current-health boost, targeting and pursuit/ranged policies are unchanged. The damage matrix checks all four conversion combinations with the policy both enabled and disabled.

Converted enemies chasing other enemies use half their previous movement coefficient. The six ranged types CF/D1/D2/D3/D4/D6 stop advancing when within 768 units and a native aim trace reaches their selected enemy. They keep facing and firing; occluded/out-of-range targets still cause pursuit. Gravity and contact queries continue while holding position, so nearby melee remains possible. Melee-only types keep approaching at the reduced speed. The existing two behavior/animation updates per tick, firing cooldowns and drain clock are unchanged. Pursuit of the player is unchanged.

test-prion-targeting adds real repeated Prion impacts, low-health doubling, saturation, dead-target and default-off checks for all ten types; an enemy/player damage matrix for contact, projectiles and explosions; and actual ranged holds and pursuit. The older special-weapon behavior-clock fixture disables this new policy to continue isolating its one-versus-two-update assertions. No campaign playthrough is required or claimed.

Validation:6887 checks,540 controller frames,270 damage cases,50 additional conversion-health cases,6 ranged holds and10 pursuit cases pass with identical MSVC and GCC ASan/UBSan/leak-check results. The existing60 enemy melee contacts and80 target selections also pass. All8 focused CTests,32316 special-weapon checks,415 boss-projectile checks and446 Gunrilla assertions pass. All128 Windows native libraries/frontend rebuild, and a hidden60-frame1920x1080 launch succeeds.

The earlier description of type 69 as freezing an enemy was incorrect. The recovered instructions implement enemy retargeting and periodic health drain. There is no demonstrated freeze/thaw transition, and no missing flag-400 freeze gate should be invented in the entity dispatcher.

The original impact routine 08010A20 accepts enemy types CD through D6 without flag 400, sets that flag at 08010C74, and writes 30 to entity +34 at 08010C78. It does not overwrite the separate halfword +36 or the target-selection timer +56. Allocation of its attached E2 effect follows these mutations; allocation exhaustion does not undo conversion.

Shared navigation 0801288C checks flag 400 at 080128D6–080128DE. If set, it subtracts ticks from +56. A null target, signed timer at or below zero, or target flag 20 causes a fresh nearest-enemy search 080036C4, with timer reset to 600. Entity +20 stores the selected target. The nearest search considers other CD–D6 bodies without flag 20, in original active-pool order; it does not exclude other converted enemies. A missing target falls back to the player. This selection happens after the existing player-sector rejection and flag-800 early exits, so it does not bypass those navigation gates.

The target affects path destinations. Phase zero still faces the player, as the original routine does. Type-CD's active controller additionally selects its alternate target for its proximity explosion. It keeps moving; it is not frozen. 08012638 still performs contact damage only when the current blocking entity has player flag 10: 20 damage for most types, 40 for D0/D6. Conversion does not introduce a universal team-based damage exclusion.

08012E1C interprets +34 as the nonzero drain period and +36 as its current countdown. It subtracts the unsigned tick byte with halfword wrapping. Only a strictly negative signed result decrements signed health by one. Surviving health reloads +36 from +34; nonpositive health sets +36=0 and body flag 1 for the controller's ordinary hit/death path. Period 30 therefore produces a 31-call cadence when ticks is one and the freshly reloaded countdown is 30. Neither expiry clears conversion flag 400 nor restores a previous target.

Read-only inspection found these calls in all ten eligible enemy controllers:

This address inspection does not claim that all those controllers are ported. The native shared navigation and drain routines were already recovered and differentially validated. The integration bug was field ownership: projectile impact wrote actor.parameter34, while the CD runtime used an unrelated controller.drain_period copy. SsaScriptActor now owns canonical parameter34 and parameter36; the CD runtime imports both at initialization and frame entry and exports both after the call, including error paths. Existing tests that set drain state now use those canonical actor fields.

test_projectile_conversion.py passes 686 ASan/UBSan assertions on real map 0: an actual weapon-7 shot converts a scripted CD, preserves an initial countdown of 17, and creates the original attached effect. The converted CD runs 49 updates with actual native visibility, nearest-enemy, path, world movement and combat. It moves in 48 updates, selects another CD, refreshes its target timer, and drains health twice with strict-negative timing. A real native damage call weakens it before the second drain triggers its ordinary death/explosion/group notification. The attached effect follows for 45 updates and is removed after the target gets flag 20. An injected animation-observer failure verifies that the countdown remains synchronized after a partial update. Corpse animation completion is controlled by the test; animation playback is not claimed.

The preexisting CD-runtime and native-enemy-query integration suites also pass after the canonical-field migration: 59,319 and 53,384 assertions respectively, both covering all 16 maps. The new result and source hashes are recorded in native-projectile-conversion-validation.json. These are native integration assertions, not additional ROM comparisons or exact compiler matches.

Requested PC behavior speed

The playable frontend enables fast_converted on entity and actor-animation runtimes. A living CD..D6 with flag400 gets two complete controller updates per world tick, including walking, facing, navigation, attack timers and periodic drain. Its actor animation likewise advances twice. Per-hit damage and original period/countdown fields retain their values; drain follows this doubled behavior clock too. Projectiles, player input, music and other actors keep their normal clock. Paused frames advance neither pass. A death/release/error in the first pass stops the second; dead actors and corpse animation/expiry use normal time.

Default runtime policies remain off, preserving the original recovery APIs. The original conversion integration still passes686 checks; entity scheduling passes85,982 and actor/corpse playback75,088. The extended special-weapon suite fires a real Prion shot at each of all ten eligible types, then verifies one versus two actual walks/retarget/drain/animation increments, unchanged unconverted actors, pause and one-tick corpse expiry. GCC sanitizers and MSVC agree.

Requested PC nearest-target behavior

The playable frontend now enables entities.nearest_converted. Before each living converted controller pass, it selects the nearest living combatant from the active pool by three-dimensional body-center distance. Candidates include players, ordinary enemies, bosses and other converted enemies. Self, dead actors, corpses, items and projectiles are excluded. Equal distances retain active-pool order, with no player preference. Selection refreshes every behavior pass, including both passes of the existing doubled behavior clock, rather than retaining the original 600-tick target preference.

The controller temporarily sees the chosen actor as its target for facing, visibility, distance checks, reactions and attacks. The real player fields are restored after the call, including errors. Converted ranged attacks launch toward the selected actor. Enemy melee contact is hostile only when the attacker has conversion flag400. Converted enemies can damage normal or converted enemies, including contacts other than their selected target; normal enemies cannot melee converted enemies. The existing combat path applies20 damage, or40 from D0/D6. Each walking update captures its actual enemy collision before a crowd sidestep can replace it; successful sliding therefore no longer suppresses melee. Contact is reset for each move and runtime call, so stale blocking pointers cannot cause a hit. Self, player bodies, corpses, released slots and nonpositive canonical target health are rejected by this additional enemy-only path. Ordinary enemy pairs remain nonhostile, and the original player-contact path is unchanged. Native floor hazards run first. With no living candidates, the target clears and the actor does not attack.

This is a requested PC policy. The recovered navigation, contact and controller routines retain their original behavior when the runtime option is off. The healthy-enemy extension above supersedes the earlier player-only policy. The existing behavior-clock fixture explicitly disables nearest selection to isolate its timing assertions.

test-prion-targeting performs actual Prion impacts on all ten eligible enemy types. The expanded melee fixture checks allowed converted-to-normal hits, blocked normal-to-converted hits and two converted enemies with crowd sliding on/off, ordinary nonhostile pairs, dead targets, disabled policy, and stale contact after both skipped movement and free walking. It also covers 80 target selections, prompt retargeting, nearest-player selection, dead/corpse rejection, vertical distance, six ranged attack types, a boss target and no-target cases. Reports: prion-targeting-validation.json and windows-prion-targeting-validation.json. The 2026-10-20 run passes 3,512 checks over514 controller frames, including60 verified melee hits, with identical GCC ASan/UBSan/leak and MSVC counters. The Windows crowd suite passes20,456 checks, and all eight focused CTests pass. All127 native Windows libraries and the playable frontend rebuild successfully. Full campaign replays were not repeated; earlier route reports describe their recorded build.