Alice wind crossing: static trajectory diagnosis (2026-11-03)

The retained attempt establishes entry into the steam/basin and a later underwater state, but cannot establish where Alice first touched the wind or whether she rose and then fell. The authentic speed -1 trigger is a thermal, with reversed, reduced gravity while its flag is active. It is not a fixed upward launch impulse. Launch position, lateral alignment and forward travel remain the first variables to observe. There is no reproducer of an emulator acceleration defect.

This single-task investigation used original files and retained evidence only. No guest, browser, campaign, restart, boat operation, guest write or production change occurred. Coordinator board entries at12:54:17/12:54:29 confirmed the cap reread, observed one live predecessor worker and authorized this static successor. Evidence and executable recovery recipe are in scratch/runs/20261103-alice-wind-trigger-trajectory/ (not committed). The predecessor publication is 967773795fbcdbb96ca4bd896a6de9ae9816f411.

What the captures actually observe

I personally reviewed approach_wind_gap.png, ordinary_wind_jump.png and wind_observe.png, and read their receipts and the original visual-review ledger. At30418 Alice is on/over the near wooden approach, facing the visible steam and far-side structures. At30698 the steam largely obscures her and the camera is below the upper crossing. At30938 the view is underwater, beneath stone geometry. The last image cannot uniquely identify which overhang or which side of the jet she is under. Camera coordinates are not player coordinates.

Commands Retained observation
189/190 down D+W;191 step80;192/193 release;194 step80 Approach image at30418, after diagonal approach and neutral execution
198/199 down W+Space;200 step180 No intervening launch/contact/apex image
201/202 release both;203 step100 First jump image at30698, 100 steps after release
207 step240, no new movement keys Underwater image at30938, 340 neutral steps after release

These are controller step counts, not measured guest simulation seconds. The receipts' frozen tick setting is200ms, but that does not prove one simulation frame per controller credit. No duration, distance, speed or jump-apex calculation here converts credits to guest seconds. The late images neither prove nor exclude a transient lift or a missed landing during the unobserved interval.

Exact original geometry and its limits

Fresh parsing of FAKK42 (12-byte header,20 lump descriptors) reproduces all368 entities and every retained plane, side, brush, model and shader record exactly. Model4 comprises brush843, six axial planes: its translated bounds are its actual box, not merely a conservative model AABB:

x[-5192,-4984], y[5020,5132], z[-320,96], origin (-5088,5076,-112), trigger_accelerate, speed -1, spawnflags 8.

South means decreasing y. Adjacent original emitters are (-5093.44,5052.36,-291.64) and (-5102.40,5057.60,-281.60); their visible steam is a landmark, not a precise outline of the contact volume.

Original brush Exact or explicitly sampled geometry
219, near plank x[-5092,-5056], y[5152,5280]; top z=-120 at y5280, rising to z=-88 at the south nose y5152
218, far plank x[-5092,-5056], y[4828,4892]; north tip y4892 has top z=-104
403, basin water x[-5440,-4608], y[4736,5376], z[-384,-320]; shader textures/liquid/tower2, contents0x20000020
415, basin floor x[-5456,-4592], y[4720,5392], z[-400,-384]
616, south rock shelf At x=-5074: z[-175.7117,-140.4551] at y4960; z[-212.0925,-119.8832] at y4928

The plank centerline is x=-5074,18 units from either side. The previously suggested trigger-center x=-5088 is only4 units inside the plank's west edge. This is a useful ordinary alignment correction, not a measured Alice position. Near plank nose to trigger north face is20 world units. The two wooden tips are 260 units apart; the south rock can provide a nearer landing than the far plank. The trigger ends at y5020,128 units before that far wooden tip. These dimensions explain why staying over the jet and choosing the exit direction matter; they do not prove a required jump range or a collision-free swept path.

The water surface and thermal bottom both lie at z=-320. Below that level, the player's origin can be outside the wind even while a body hull still overlaps its boundary. Basin water extends well north, south and east of the thermal. An underwater frame therefore fits falling out of its sides, through its bottom, or under the south shelf after passing the jet. It cannot distinguish them. The scene's matching steam/water/stone makes this basin the supported visual identification; there is no recovered exact player x/y/z, heading, launch height, contact history or entry trajectory. Assigning those from the screenshots would be an estimate, not an original-file fact.

vertical-brush-samples.json records point sections of all840 world brushes, including solid/playerclip and water. It excludes mesh/mover collision and player hull expansion, and makes no claim to reconstruct the underwater camera pose.

Original thermal implementation

Addresses below are preferred original fgamex86.dll VAs (base0x10000000), not runtime addresses or proposed trace points. Full original disassembly and decoded constants are retained in this run.

These facts make a late/low or off-center entry, a lateral exit, or travelling beyond the jet before enough height was gained all plausible. They do not select one cause. The images also leave open ordinary collision/ledge contact and a brief successful lift followed by a fall. Holding Space is the authentic +moveup binding; the original readme documents it as jump/swim up. It does not document an extra glide key required here. An unobserved key requirement is not an evidence-backed diagnosis.

Bounded ordinary next attempt, only after separate authorization

Use the ordinary opening and first gnome route already retained; do not restore a synthetic position or call scripts. At the crossing, align with the middle of the narrow rising plank, face its southward axis toward the steam/landing, and approach its visible nose with W in2/4/8-step increments, releasing and capturing immediately after each increment. Do not insert an80-step neutral wait before inspecting footing. The plank nose, rather than the early screenshot's unknown point along the plank, is the intended launch landmark.

For one initial jump, press W+Space and capture after2 steps while held. Review before each additional2/4/8-step increment. Keep this first contact/lift observation to at most32 held credits; after release capture immediately and after2/4/8 neutral credits, with at most64 total credits from initial press. These are controller bounds only. Stop upon submersion, death, loss of recognizable footing or failure to locate Alice; no blind continuation or automatic retry. If an early image clearly shows ascent in the jet but insufficient forward clearance, a separately bounded, reviewed continuation can target the south rock/plank. Do not coast240 neutral credits without observation.

The question this resolves is: does Alice leave the plank airborne and rise while still centered in the steam, or does she fall/exit laterally before lift is visible? Observe that before selecting a new heading or hold length. No exact numerical heading is recovered from the old turn inputs. No engine implementation work is justified unless a separate scoped reproducer establishes faulty behavior.

After a reviewed south landing, the original route turns east along the landing near y4760, crosses the second dialogue boundary x[-4776,-4760], y[4576,4936], z[-136,144], then follows the timber passage toward the knife altar (-3978,4326,-168) (Item_WeaponPickup_Knife, altar_knife.tik, knife_cat). Let all natural dialogue/pickup cinematics finish. This is static availability, not a proven traversed route, knife ownership or combat success.

Identities, checks and recovery

Original file/member SHA256
demo/pak0.pk3 14ce74568458078f51f5b4f42a65c5a5888c01c487bbf8c064f541039a306556
maps/d_gvillage.bsp 4d2435f126831467d52f23722823b5486b6e9ce2f9149bd2f5e06cca94457855
maps/d_gvillage.scr c9f5c008e4ea35f1ead0e83bbdfcdc86801bcb7bfa0779481748e5430cf4738e
maps/cinematics/gvillage_rabbit.scr 00554179a0786cbc1e2e15f1c399e6f766513c7fba5f4957cda582a52a218c63
default.cfg fa3a595ec886b92e96a4a572ac51a93516b462c9d48dfa35258f453001150e8e
demo/fgamex86.dll 42838c2a96520c4b8b42e710d7d5bcf3a7a7f2a71b141a9f75bff01ee28f3d4c
demo/cgamex86.dll 2fc6c957d8f554ff296271c9c50ec6aa3b5f72531061bf8a23af8d219e6d8c24

All6054 predecessor archive payloads were freshly checked against the full manifest; selected captures, commands, ledger and receipts were copied from verified archive bytes into this run. The three local predecessor receipts differ only by one added trailing newline; archive copies match the published hashes exactly. No predecessor artifact was overwritten. Original parsed geometry equals the retained arrays, and the file sizes/hashes and constants are retained as JSON. No production code changed, so no build or game test was run for this document.

RECOVERY.md, full-manifest.json, checks.json and predecessor-verification.json in this run retain reproduction/check instructions and local artifacts. No boat is assigned to this worker. Recorded predecessor bx_3uvpk4nf is STOPPED, original expiry2026-11-03T13:59:05.938Z, snapshot retained, shutdown owner alice-wind-knife-native; its state was read from released evidence only. Runtime lifecycle ownership remains with the coordinator; no boxes were touched.