Collected 15 September 2026. User reports are symptoms, not established causes. Preserve these alongside the Cheat Menu parity and release-deliverables checklists.
Work order: finish the native Cheat Menu port already underway, then address this queue, then perform the final 0.3.0 release checks. New reports are notes, not instructions to interrupt the active task, unless _eins0fx explicitly changes priority.
| Report | Current status | Required check |
|---|---|---|
| Equipped weapon attacks look unarmed | Fixed in the active session, confirmed by _eins0fx | Keep testing recurrence of completed-intro restrictions; preserve legitimate story locks |
| Teleporters, launch/jump points and death reset appearance | Empty-MID stall fixed; real launcher cancellation and repeated shell swaps passed | Exercise same-pawn and new-pawn transitions with an active CSS selection |
| Bloody characters return to vanilla colors or broken materials | Investigating | Trace gameplay-created material instances and preserve blood/effect parameters alongside CSS colors |
| Eredrim's diapason and Tiel's dagger float under the feet after switching shells | CSS-wide correction live locally, verified with Long Hair on both shells (16 September) | Actual attachments, native draw/stow, six swaps, Original cleanup and core reload pass. Combat abilities and other outfits still need broader coverage |
| Repeated shell hotkeys from original Cheat Menu eventually stop CSS | Colored Proxima-switch stall fixed; three colored cycles and eight earlier switches passed | Repeat gameplay-shell transitions and inspect player identity, menu lifecycle and recovery state; do not presume incompatibility |
| Seductress fabric behavior | Deferred by _eins0fx, known issue | Separate cloth/skin weighting problems from runtime pause and attachment state |
| Seductress feet float during sprint and footstep sounds are missing | Deferred by _eins0fx, known issue | Compare original and Seductress on the same gameplay shell and animation, including foot bones, notify tracks and physical material |
| Dodge produces a torn-looking trail across CSS outfits | DarkBro displacement branch corrected in the live core; _eins0fx confirms the shredded trail is gone | Preserve the native translucent overlay; see the controlled comparison below |
| HIT2 DE Scyther aims Trial Repeater into the sky | Upward default-projectile alignment reproduced on Proxima; fired trajectory not yet checked | Compare original replacement and CSS conversion with the same gameplay shell, weapon and aiming inputs; original-mod cause remains unconfirmed |
The original Cheat Menu should be disabled before testing CSSX Cheat Menu. CSS appearance compatibility with gameplay-shell changes still needs its own checks independently of that recommendation.
Local evidence before pausing this investigation: work/cssx-native/tiel-missing-bone.json, tiel-before-css.json, tiel-css-sockets.json. The blood-effect test preserved Seductress mesh/materials (blood-live-check.json). A simulated stock-mesh dynamic material delayed recovery until removal and then recovered (transition-mid-repro.json); that alone does not establish the natural portal failure's cause. The test selection on Tiel was removed and the gameplay shell returned to Genessa afterward.
15 September update: native Cheat Menu implementation is committed in 52168dd.
The empty dynamic-material recovery stall now passes three repeated live cycles,
including active-effect protection and native Inventory cleanup. See
the recovery notes and
work/cssx-native/material-recovery-live.json. This is a constituent fix, not a
claim that every reported travel/death path is verified.
JayLuk3 clarified that only Long Hair Proxima is installed. Their downloaded recording shows the floor-level accessory while the outfit remains visible. Direct cooked-mesh readback and the live Tiel comparison confirm the missing socket-parent bones. The CSS correction uses a hidden stock-mesh pose follower only for affected mesh accessories. Tiel and Eredrim pass attachment and native draw/stow checks with Long Hair. Six swaps, Original cleanup, core-reload cleanup and material/Inventory recovery checks pass. Original gameplay and CSS settings were restored. No port repack or published ZIP change is needed for this test. The separate persistent portal warning and crash remain unconfirmed. See the recording and attachment investigation.
_eins0fx requested deferring the missing attachment bones and mesh/animation-fit issues. These are known issues for 0.3.0 rather than release-blocking runtime work. No global height adjustment or attachment workaround is included in the release.
The cooked V1 and V2 reference skeletons both contain 258 bones, with identical
foot-bone transforms; neither includes Tiel's tiel_dagger bone. This does not
establish a floating-feet problem in V2. _eins0fx reports that V2 does not have the
floating/fit symptoms seen in V1. Keep cross-shell accessory attachments separate
from foot placement, cloth and movement-fit reports.
A temporary hidden follower-component probe produced a usable Tiel socket pose,
but it was not a complete gameplay fix. Its source and patch are parked under
work/cssx-native/deferred-attachment-follower.*, and the experiment was removed
from the shipping source. Probe evidence: attachment-follower-probe.json.
Eight successive Genessa/Tiel/Eredrim switches recovered each selected outfit,
kept the scythe in hand and allowed native Inventory to open and close. The five
checked attack/dodge/aim/menu restriction counts stayed zero. This exercised
retained-pawn shell changes, not death or every original Lua hotkey combination.
Evidence: work/cssx-native/repeated-shell-recovery.json.
A real Marrow Keep launcher activation and destination-selection cancellation
kept Seductress V1 applied through all 25 sampled return states. Inventory opened
afterward, with no pending recovery or ignored movement/look input. This does not
cover selecting a destination, beacon cleansing or death.
Evidence: work/cssx-native/launcher-cancel-live.json.
_eins0fx then reported a pause near the end of that animation with colored
Seductress V2. Logs showed material recovery taking 801–963 ms. A repeated
material-slot reset reproduced it at 815–842 ms: apply() discarded valid color
MIDs and regenerated dye textures even when the outfit had not changed.
The recovery path now reattaches the existing materials after a recognized stock material reset. It validates every weak reference before writing, preserves active/unrecognized game effects, verifies the resulting slots, and falls back to normal rebuilding when resources have been collected. It does not retain old worlds or change texture resolution. Three repeated runs took 0.23–0.30 ms and preserved the exact material objects. These are CSS apply timings, not whole-game frame times or a promise that all loading hitches are gone.
Evidence: work/cssx-native/color-recovery-before.json,
color-recovery-after.json. Repeatable regression check:
python3 tests/live_material_recovery_perf.py with a colored outfit and the
staged developer core. _eins0fx confirmed the natural V2 launcher animation is smooth after the fix.
The same optimization now covers a recognized stock-mesh reset on the retained
pawn. Before: 806–834 ms. After: 2.72 ms, 1.82 ms, 2.70 ms. All runs retained the same color materials.
python3 tests/live_material_recovery_perf.py --mesh covers this path.
Three additional stock/empty-MID/active-effect cycles passed with the colored V2
outfit, including Inventory cleanup and unchanged animation instance and saved
preferences. All nine portable test groups passed. New-pawn/death recovery still
requires rebuilding when the old resources no longer exist; its performance is
not established by these retained-pawn tests.
A second blood test used the currently colored Seductress V2 outfit. The native
blood-mask value changed while the skeletal mesh and all override-material
objects stayed identical. Removing only CSS_ColorBloodProbe restored the test
mask contribution. Saved preferences were unchanged. Evidence:
work/cssx-native/colored-blood-check.json. This does not cover every death,
overlay or environmental-effect path.
One directional dodge each with Seductress V1, Beaute Genessa and stock Genessa
on the same gameplay shell in Marrow Keep did not reproduce a persistent shadow
trail in the post-dodge screenshots. Component bounds changed during motion and
settled afterward. Neither fixed skeletal bounds nor parent bounds were enabled;
BoundsScale was 1, shadow-cache behavior was Auto, and capsule shadows were off.
The screenshots were taken after the animation, so a brief motion/temporal trail
still needs a capture during the reported artifact. No shadow-quality or global
bounds workaround was applied. Evidence: work/cssx-native/*-dodge-bounds.json
and *-after-dodge.jpg for those three appearances.
Epic's Virtual Shadow Maps documentation explains that deforming geometry invalidates cached shadow pages and displaced geometry needs sufficient bounds. This is a diagnostic lead, not an established cause of the reported artifact.
The Proxima aiming comparison exposed another CSS recovery blocker. Switching
from colored Seductress V2 on Genessa to stock Proxima left a CSS-created
MID_MI_Genessa_top in slot 6, beyond Proxima's six material slots. Its dye texture
parameter caused the active-effect guard to reject outfit changes indefinitely.
The material was CSS-owned, not a story restriction or another mod's effect.
detach_residual_colors() now removes only exact live MIDs in CSS's ownership
map when the game has changed that component's mesh. It neither recognizes
ownership by name nor clears unfamiliar effects. The weak dye cache remains
available for temporary stock-mesh recovery; cleanup also runs before unloading.
The regression failed before the fix with that leftover slot, then passed three
Genessa/Proxima/Genessa cycles. The tests preserved appearance preferences and
returned to the original shell. Material-only recovery remained 0.24–0.28 ms and
stock-mesh recovery 1.85–2.06 ms. Three more active/empty-effect cycles and native
Inventory cleanup passed. Run python3 tests/live_colored_shell_switch_check.py
with the documented safe-game prerequisites. Evidence:
work/cssx-native/colored-shell-switch-before.log, colored-shell-switch-after.log,
proxima-ready-check.log, and residual-color-transitions.log.
The game's sidearm name is Triarch Repeater (ID_Sidearm_MachineGun). Aiming
on Genessa with stock and HIT2 appearances kept the weapon-forward pitch close
to horizontal. The first Proxima comparison was invalid because the leftover
CSS dye material above prevented HIT2 from applying. It was repeated after that
fix, with an explicit assertion that the requested appearance had applied.
With Proxima and the same aim input, stock default projectile rotation had a pitch of 4.16 degrees; HIT2 had 19.16 degrees. Weapon-forward pitch remained about 0.54/0.99 degrees respectively, while its roll changed by about 15 degrees. This establishes an upward default-projectile alignment difference, not a verified fired trajectory or proof that every aiming mode uses that default transform. No universal gun-angle correction was applied.
The source skeletal-mesh .uexp and the converted round-trip payload in
work/convert-hit2-p-final are byte-for-byte identical: 4,804,774 bytes, SHA-256
f07b13b38ec0f4b30e33cc0947805ac12374819089d6c1d71f66b8c695c8ee47.
That check does not replace an in-game original-package comparison or verify
all post-process and firing behavior. Evidence: work/cssx-native/triarch-aim-check.json
(Genessa), triarch-KnightLady-aim-check.json (valid final Proxima run), and the
matching aim screenshots. The temporary repeater was removed, the Nail Shotgun
restored, and the gameplay shell returned to Genessa after testing.
_eins0fx supplied screenshot-2026-09-15_13-31-56.png (during dodge),
13-32-14.png and 13-32-32.png (Inventory, Harros selected) from Pictures.
The reported scope includes all CSS outfits, including Seductress V1/V2 and
ports. The motion capture shows a torn-looking surface trail, not enough
evidence to label it a shadow-cache problem. Earlier after-dodge screenshots
did not test this instant.
A separate report concerns Seductress V1/V2 becoming uniformly dark/glossy after
blood exposure. Source: ~/Uploads/ScreenRecording_09-15-2026 12-28-26_1.mp4,
from approximately 01:00. Investigate and document the authoring findings in
CSS-eins0fx-collections/seductressV2-eins0fx-CSS before changing its assets.
The previous blood check tested material identity, not visual blood coverage.
Read-only runtime inspection after the opening boss reports
CharacterId.Player.Darkform.StrongOne, CSS enabled, no recovery pending and
no selected CSS appearance for that form. This is distinct from an appearance
recovery failure: CSS stores selections per gameplay identity.
Evidence for this investigation is under work/blood-dodge-investigation/.
The white shredded geometry is a material displacement effect. During the same
frozen Harros dodge pose, removing OverlayMaterial left the white geometry in
place. Setting only DarkBro WPO Stretch Scale from 256 to 0 removed it while
retaining the normal translucent dash overlay. Setting Shell Explosion WPO Scale to 0 did not remove the artifact.
CSS now neutralizes the DarkBro stretch parameter on its owned material instances when applying/customizing an outfit. These instances share the dye material ownership and recovery cache. Original assets and unfamiliar dynamic materials are not edited. Standalone materials, including Beaute Genessa's cloth drivers, are left intact because they have no material-instance parameter API. There is no new per-dodge callback or texture rebuild.
The separate native dash overlay and shell-explosion parameter remain intact. This intentionally disables one native displacement branch on replacement material instances; it is not a general cloth, shadow-cache or bounds fix. The precise topology assumption inside the cooked shader is not established.
The read-only regression tests/live_dodge_material_check.py failed with scale
256 before the fix and passed with 0 afterward. Controlled Harros captures
compared stock, HIT2 and Seductress before/after. _eins0fx also confirms:
"The shredded-body trail is gone."
Evidence: work/blood-dodge-investigation/harros-seductress-probe*.jpg,
fixed-harros-*.jpg, parameter-before.log, parameter-after.log.
A material readback passed for one variant of each of the nine installed
outfits, including Seductress V2 Original and Beaute's standalone cloth drivers.
This is not a visual test of every variant. V2 color reset and restoring its
palette passed; saved preferences were restored unchanged. Three material-reset
recoveries took 0.26-0.35 ms and three retained-pawn stock-mesh recoveries took
58.00-59.93 ms on V2 Original, all reusing the same material objects. The latter
is a different variant from the earlier topless-harness timing and does not
establish equal mesh-reset cost. Both pass the existing 100 ms apply budget;
these are CSS apply timings, not whole-frame or loading-time guarantees.
The final native core built and staged successfully; the nine existing portable test groups passed during this investigation. The fix is staged in the active developer core. Published 0.3.0 ZIPs have not been replaced.
The authoring skin BRM red channel is solid white in both V1 and V2. A controlled V1 and V2 Original comparison changed only that channel, retaining identical green/blue channels and the same material object. The white control reproduced uniform glossy blood coverage; a diagnostic gradient restricted coverage while clean skin stayed pale. This identifies a texture-authoring problem, not a requirement to rebuild the body's geometry. Neither package was changed during that initial investigation.
The investigation, source trace, proposed mask work and copied evidence are in
CSS-eins0fx-collections/seductressV2-eins0fx-CSS/docs/blood-material-investigation.md.
Earlier blood tests only established mesh/material identity, not correct visual
coverage. Also, reflected CustomPrimitiveData is the serialized default, not
the separate internal runtime array used by the renderer. Do not cite that
property as evidence that runtime blood inputs stayed unchanged.
15 September follow-up: Seductress V1 1.0.1 now replaces only its skin BRM
with a spatial blood mask baked to its own UV atlas. All 26 other assets and
the original catalog/colors are preserved. The installed package passed native
blood strengths 0.2, 0.5, 0.8 and 1.0, clearing, edited skin color and clean
Inventory/CSS preview checks. The uniform coating is absent in those captures.
Preferences were restored exactly. Natural combat, wetness, hardening and
travel/death were not retested in this pass. V2.1 remains Claude's separate work.
See CSS-eins0fx-collections/seductress-eins0fx-CSS/docs/blood-material-fix.md
and its work/blood-fix/runtime/ evidence. No CSS/CSSX code change is required.
15 September: V1 now includes the grounded foot geometry and foot/calf weights already used by V2, seven bounded native body springs, and Chaos cloth on both skirts, shoulder cuffs and neck accessories. It preserves the 1.0.1 blood mask, material/texture assets, color controls and stable outfit ID. This is an outfit update, with no CSS or CSSX runtime changes.
The first optimized test candidate left nine loose simulation vertices after decimation and crashed during world-load restoration. The exporter now compacts all physical vertex arrays with their indices. Native Chaos neighbour checks and cooked validation reject the old data and pass the correction. The visible render mesh is byte-identical across this correction.
The corrected candidate passed world loading, outfit switching, walking, sprinting, dodging, bounded body motion, blood coverage and clearing. _eins0fx confirmed the visual motion and grounding. All four cloths are active, with 2,714 dynamic particles. Broader animation clipping and untested terrain remain possible; this does not claim every animation is perfect.
The draft number 1.1.0 was changed to 1.0.2 at _eins0fx's request. The ZIP
keeps both tested IoStore containers byte-exact; only version/validation metadata
changes. It contains exactly the matching _P.pak, _P.ucas and _P.utoc
inside CSS_Seductress_eins0fx_P/, without user state or runtime DLLs.
Release: CSS-eins0fx-collections/seductress-eins0fx-CSS/dist/releases/v1.0.2/.
ZIP SHA-256: 1620ca955fbc079a562c7c14ab4e2827c811af33848418841d58bb41aebfac4c.
Changelog: that project's docs/changelog-1.0.2.bbcode.txt, also copied to
reference/internal-docs/CSS - Seductress/08-changelog-1.0.2.bbcode.txt here.
Source notes and regression evidence remain in the V1 project's
docs/physics-proxy-crash.md and historical work/physics-1.1.0/ directory.