# Grouper Tender: status

Working name "Grouper". The user's blueprint: `reference/user_blueprint.jpg` ("DEEP SEA EXPLORER - SUBTUB CLASS",
DS-ST-001), *"create me 3d model of this tender from this blueprint exactly"*; then the concept render
`reference/user_concept.jpg` (*"Check this picture as a referance"*). Blueprint rules shape; concept rules the look.

## Done
- 2026-10-22, Designer: zooms with rulers (`build/gr_zoom.py` -> `reference/z4_*.png`, concept zooms `z_concept_c_*.png`);
  silhouettes and drawn lines of every view (`build/gr_trace.py`, `gr_views.py` -> `reference/mask_*.png`, `lines_*.png`,
  `build/outline_tables.json`); every number in `build/gr_measure.py` (scale 0.0171 m/px, origin, clamp axis, hull loft
  tables, every part, cabin, moving parts, materials, the concept's strips, traces and lamps); check pictures
  `reference/trace_*.png` (`build/gr_tracecheck.py`); the design `design/DESIGN.md`.
- 02:43-03:01 two copies of the Designer wrote here (the Administrator's mistake); reconciled at 03:05-03:20: one set of
  scripts (gr_masks.py removed, gr_views.py kept), masks and tables regenerated, `gr_measure.py` revised for the concept.

## Rebuild the readings
`python build/gr_views.py && python build/gr_sil_tables.py && python build/gr_tracecheck.py` (Python 3.14, PIL, numpy).

## Next
- Coder: build headless from `design/DESIGN.md` and `build/gr_measure.py` (the Mako's and Spearfish's pipeline copied
  into `build/` with a `gr_` prefix): hull loft, parts, collar as the Kazah clip, cabin, rig, checks (moon pool, carousel,
  clamp with the antenna raised and folded, motion, view), surfaces, UCX, FBX, side-by-side against the sheet and the
  concept.
- Not measured yet: moon pool and carousel (the Coder's checks).

## Coder, 2026-10-22: geometry, rig, collision (done)
`Tender_Grouper.blend`, built headless in 4 s: `blender -b --factory-startup --python build/run_build_headless.py` (rebuild on
the saved file: `blender -b Tender_Grouper.blend --python build/run_build_headless.py`). 7.45 x 2.70 m, keel z -1.73 to the
collar's lip 1.375 (antenna tip 1.83); origin 9 cm behind the clamp axis (0, -0.09).
- Objects: `SM_Grouper_Hull` (skin, collar, trims, every static fitting; 83.7k tris, quads + 2 tris), `SM_Grouper_Glass`
  (2.1k), `SM_Grouper_Cabin` (11.9k), moving: `SM_Grouper_Hatch`, `SM_Grouper_Door_L` / `_R`, `SM_Grouper_Rotor`,
  `SM_Grouper_Antenna` (origins on their axes). 0 n-gons, 0 zero-area faces. 12 sockets (SOCKET_Clamp, _Seat_L/R_Eye, ...).
- Rig on `Grouper_Root` (simple expressions, all valid): `hatch_open_deg` 0-100, `door_l_deg` / `door_r_deg` 0-95,
  `rotor_turns`, `antenna_fold_deg` 0-70 (fold it for docking), `lights` (for the surfaces pass).
- Collision: 29 convex `UCX_SM_Grouper_Hull_NN` (2.8k tris, convex within 0.5 mm), "GROUPER | Collision (UCX)" hidden.
- Materials (flat looks, names for the surfaces pass): M_Grouper_Gunmetal, _Band, _FrameDark, _Acrylic, _Strip (groove
  floors, the collar's ring, the duct's ring), _LampWhite, _ScannerLens, _DuctGlow, _Recess, _Steel, cabin _Cabin, _CabinFloor,
  _Seat, _Screen, _CabinDark (M_Grouper_Trace and _Marking exist, unused: traces and markings are texture work). Seam fields
  on the hull vertices: `seam_chine`, `seam_belly`. No UVs yet (surfaces pass).
- Checks (`sh build/run_checks.sh`, results `build/checks/*.json`): motion 0 hits (hatch 4.7 mm, doors 3.1 mm, antenna 6 mm,
  rotor 4.9 mm least gaps); view: both eyes straight ahead through the front glass, 51 % of the cone through glass; cabin 0
  crossings, 0 within 5 mm (least 11.8 mm); arm head (the Tiger's 20-pose dump): 0 hits, claws 5 mm from the lip, hatch 30 mm,
  antenna 80 mm raised / 319 mm folded; clash audit: 113 part-to-part contacts, intended joints; folds: 26 quads > 90 deg left
  (10 at each pod's nose tip, 14 inside the front glass trim's lower corners, 2 at the stern).
- Previews: `previews/wb_*.png`, `previews/topo_*.png`, `previews/compare/sheet_side|top|front|back.png`.
- Tools: `gr_preview.py`, `gr_compare_sheet.py`, `gr_section.py` + `gr_section_plot.py` (true sections), `gr_shot.py`,
  `gr_fold_audit.py`, `gr_clash_audit.py`, `gr_dump.py`.

## Next
- Surfaces pass (UVs, baked masks, the strips' and traces' emissive masks, markings "Subtub 1", insignia, seal), the concept
  camera fit and `previews/compare/concept_pair.png`, ship checks (moon pool, carousel, 86-pose clamp in the arm and ship
  files, BOARD-to-POOL), FBX + layout JSON, studio renders.
- Known: the junction ahead of the door where the band's front, the frame's corner, the pod's nose and the cheek meet
  (x 836-850) has wavy column flow and the 10 folded quads; the pods' tail cones bend in toward the hull (BACK draws them
  concentric) so they never hang free.

## Coder, 2026-10-22 (finishing step): fit checks against the ship and the arm (done)
All at the current scale (S 0.0171, unchanged), results in `build/checks/`:
- `gr_dump_tris.py` -> `grouper_tris.npz` / `grouper_points.npy` (docked: antenna folded 70 deg) and `_raised` (as drawn).
- `gr_sweep_check.py` (ship file, read-only): moon pool at BOARD **1.455 m** to the deck well's lip (limit 1 m; Kazah 1.206);
  carousel table: side 0.834 m, roof 1.218 m (antenna raised: 0.783 m). `sweep_and_moonpool.json`.
- `gr_carousel_bvh.py`: 360 placements, 0 crossings, no ship face within 0.6 m (both antenna poses). `carousel_bvh*.json`.
- `gr_clamp_check.py`: arm file 86 poses 0 intersections, least gap 5 mm (claws to the collar's lip r 0.552); ship file
  3 Tender places 0 intersections, 5 mm. Raised antenna: 73 mm from the ore hose (folded: 465 mm). `clamp_*.json`.
- `gr_descent_check.py`: BOARD to POOL 9 steps, 0 crossings, least ship gap 1.337 m, arm 7.3 mm (claws). `descent.json`.

## Coder, 2026-10-22 (finishing step): surfaces, export, pictures, comparison sheets (done)
- **Geometry added** (rebuild, `gr_doors.py`, `gr_clamp.py`): a door stop ring inside each doorway's sleeve 8 mm behind the leaf
  and a seat ring under the hatch's rim, so no line of sight runs through the 8 mm door gaps or the 6.7 mm hatch gap into the
  cabin (the side render saw through both doors; the cabin check now finds 0 cabin vertices visible from outside, was 8).
  Motion unchanged (0 hits), clash audit: intended joints only. Hull 86,148 tris.
- **Surfaces** (`gr_decals_draw.py` then `gr_surfaces.py`, about 2.5 min): UVMap atlas (exterior 4096, cabin 2048) + UV_Tile;
  baked Masks / BevelNormal and the exterior **Emissive** map (strips, traces, duct glow, lamps, scanner cells); the concept's
  materials on the user's pack sets; mesh decals `SM_Grouper_Decals` ("Subtub 1", generic striped insignia, generic seal with a
  cyan circuit emblem, UI panels, the small stern insignia); 9 light drivers (`lights * k`, simple, valid).
- **Export** (`gr_export.py`): `fbx/SM_Grouper_*.fbx` (10 files incl. `Grouper_Assembled.fbx`, hull with 29 UCX), checked by
  re-import (movers at their pivots); `Grouper_layout.json`.
- **Pictures** (`gr_render.py -- v01 128 studio water concept lookalike`, `gr_preview.py -- wire tag=v01`, `gr_pairs.py`):
  `previews/v01_studio_*` (8), `v01_water_nose|stern`, `v01_topo_*` (8), `compare/concept_pair.png` (camera fitted by
  `gr_concept_fit.py`, 35.8 mm, rms 9.0 px main landmarks), `compare/lookalike.png`, `compare/cmp_<view>.png` (IoU SIDE 0.938,
  TOP 0.948, FRONT 0.959, BACK 0.951, BOTTOM 0.864), `compare/side_inkover.png`.
- `README.md`, `build/run_all.sh`, `build/run_checks.sh` (full chain).

## Next (open)
- Topology: 26 folded quads (pod nose tips, front glass trim corners, stern), the wavy junction at SIDE x 836-850; the pods'
  tail cones bend in (BACK draws them concentric). No LODs; hull 86k tris; hull atlas fill 38 %.
- The user's word on the look and the additions; Unreal import.

## Critique, round 1 (2026-10-22): verdict FIX
`critique/round1.md` (pictures and scripts in `critique/round1/`; nothing in the model, scripts or .blend changed). Outline 88 %
(IoU SIDE 0.935, 0.976 without the drum-mount box; TOP 0.945, FRONT 0.959, BACK 0.951), look 62 %. Seven MAJOR items:
1 the drum's mount is a wire cage, not the concept's solid bracket + bent tube (SIDE loses 0.75 m2 at x 845-925); 2 topology:
dense 161 x 144 grid, 86k tris, wavy cheek flow, 26 folded quads, no LODs; 3 BACK: pods' tail cones are lumps, not ringed
cones; 4 circuit traces are broken pixel dashes (draw them as vector lines); 5 light strips 3-5x too wide (concept 2-3 px vs
model 8-14 px from the same camera); 6 windows dark, the concept's are full of lit cyan screens; 7 SECTION A-A seats are boxes.
Twelve MINOR items (gloss, blotchy frames, porthole see-through, lamp beside the drum, drum ribs, stepped duct, collar pockets'
finish, BACK plate, door latches/hinges, hidden faces, texel density, scale note to the user).

## Coder, critique round 1 (2026-10-22, in progress)
Backup of the state before this round: `versions/v01_before_critique_r1/` (.blend, build/, fbx/, textures/, layout JSON,
previews/compare/, README, STATUS; 101 MB).
Done so far (geometry, built headless, `sh build/gr_quick.sh TAG` = build + fold audit + wireframes `previews/TAG_topo_*`):
- M2 re-loft: 72 columns (was 144), 103 stations (was 161); free columns smoothed along the hull (`smoothed_arcs`), station
  pairs kept consistent; groove walls 8 mm, the frame band's front end and lower edge 30 mm chamfers; folds 0 (was 26).
  Hull 46.8k tris (was 86.1k; the skin grid 15.6k, was 49k). Hidden faces pruned (fittings >15 mm under the actual skin, grid
  faces inside the pod pieces and under the collar).
- M3: the pod is a straight nacelle (axis +-1.017, z -1.382); its nose and its tail cone are revolves (`gr_fittings.py`);
  the tail steps in at BACK's rings r 15/11/8 px, end cap r 5 px; the grid's lobe sinks only inside those pieces.
- M1: solid bracket block 0.30 m wide under the nose (bolts), the bent tubes as drawn (8x zoom), sled rails, hangers.
- m5 drum grooves, m6 stepped duct (slot follows the shroud now), m7 pocket chamfers (8 deg) dark, rim r 0.553, m8 slim skeg
  root, m9 oval latches and flush hinges (axis 4 mm out), m3 smoked porthole acrylic (M_Grouper_AcrylicDark).
- M7 bucket seats, console column with the two-pad control box; M6 screen banks (dashboard 3 per pilot + middle, under the
  side windows), overhead screens twice the area, ceiling glow strips (M_Grouper_CabinGlow).
- TOP's pod vents and raised capsule built (the README said they existed; they did not).
- Checks so far: motion 0 hits (doors 3 mm, hatch 4.1 mm, antenna 6 mm, rotor 4.9 mm); view 52.3 % through glass; cabin 0
  crossings, least 15 mm, 0 seen from outside.
Next: surfaces (strips as thin lines, vector traces, satin gunmetal, frames' set, screens cyan), LODs + export, ship checks,
renders and comparison sheets, README.
- (later the same session) Surfaces redone: the hull alone on the 4096 Exterior atlas, its skin one arc-length island written
  by the build (UV_Grid, v per station: no folds), 300 px/m (was 239); doors, hatch, glass, rotor, antenna on a new 2048
  "Parts" atlas (M_Grouper_Part* copies); strips = a 28 mm line in each groove floor (strip_v/strip_w vertex fields), strength 3;
  traces = vector polylines (`gr_decals_draw.py` TRACES), 10 mm lines, 20 mm dots; satin gunmetal (coat 0.06 at 0.32); the
  frames' StainedSteel tiled 0.30 with its contrast flattened; screens on a lit cyan ground, no amber, strength 5; cabin glow 3;
  porthole smoked; decal sheet 2048 x 2048 (pad 0.26 m: the coarser hull faces). Bake margin 2 px, pack margin 4.5 px.
- m4: a lamp beside the drum (first on the cradle's side frame; moved to the drum face's lower rim, r 0.505, 26 deg below the
  side, because from the concept camera the drum hid it); landmark SOCKET_DrumLamp_L fits within 6.3 px.
- Export with LODs (`gr_export.py`): hull 47.4k / LOD1 23.7k / LOD2 9.5k; glass, doors, hatch, rotor too.
- `run_all.sh` now always builds a new file; `gr_fold_audit.py` writes `checks/folds.json` (over 20/25/90 deg, skin alone).

## Coder, critique round 1: DONE (2026-10-22)
Final pipeline `sh build/run_all.sh r1 128` + `build/gr_lods_preview.py`, all on a new file. Every item of `critique/round1.md`
worked; README.md has the round's table ("Round 1") and the current numbers.
- Checks: moon pool 1.457 m; carousel side 0.836 / roof 1.21 m (raised 0.789), triangle sweep 0 crossings; clamp 86 poses + 3
  ship places 0 hits, least 5 mm; BOARD to POOL 0 crossings (ship 1.301 m, arm 7.2 mm); motion 0 hits (hatch 4.1, doors 3.0,
  antenna 6, rotor 4.9 mm); cabin 0 crossings, least 9.1 mm, 0 seen from outside; view 52.3 % through glass; drivers simple.
- Mesh: hull 47,322 triangles, quads only (was 86,148), folds 0 (was 26), quads over 25 deg 254 (was 432), over 20 deg 349;
  no slivers; LOD1 23,661 / LOD2 9,450. Exterior about 55k (the critique's target was about 35k: not met).
- IoU: SIDE 0.942, TOP 0.950, FRONT 0.954, BACK 0.953, BOTTOM 0.868. Concept camera rms 9.0 px main landmarks, 16.2 all.
- UV: hull 304 px/m (4096, alone), doors 510, hatch 586 (Parts 2048), cabin 246.
- Current pictures: `previews/r1_*`, `previews/lods_hull.png`, `previews/compare/*` (concept_pair, lookalike, cmp_*,
  side_inkover, lk_*). `previews/v01_*`, `wb_*`, `topo_*` are the first delivery's (stale).

## Next (open)
- Twists over 20 deg (349) are not under the critique's limit; the exterior is about 55k triangles, not about 35k.
- The front window reads darker than the side window from the concept's camera (its screens face the pilots).
- The scale note for the user: the sheet's largest label says 15,500 mm; the model is 7.45 m (the collar sets the scale).
- The Critique's round 2; the user's word; Unreal import.

## Critique, round 2 (2026-10-22): verdict FIX
`critique/round2.md` (pictures, scripts, JSON in `critique/round2/`; nothing in the model, scripts or .blend changed; Blender
headless read-only). Outline 89 % (IoU SIDE 0.939, 0.976 without the drum-mount box; TOP 0.947, FRONT 0.954, BACK 0.953,
BOTTOM 0.866, 0.908 at best shift), look 75 %. Motion re-run by the Critique: 0 hits. Three MAJOR items:
1 (new) the claw pockets are torn: the skin's pruned opening under the collar (r up to 0.63) is open between the Kazah wall
(r 0.535) and the skin in the pockets; the two aft pockets (45/135 deg) are see-through holes (336/337 rays reach the far flank
and the cabin; `collar_rays.json`); the collar ring's edge is a sawtooth; 2 the drum mount is still a square-tube scaffold:
the drawn bracket plate (top y 400 x 903-921, front x 921 to y 450, aft diagonal (903,401)-(888,440)) is open, SIDE misses
2,302 px in the mount box; 3 from the concept camera the front window is black (nothing lit where the sight lands: port seat
front, port wall, rear bulkhead) and the screens clip to flat cyan (strength 5). Eight MINOR: flag/seal decals share 4 faces
per side (bite in the flag), drum band at y -3.131 wound inward + window reveals seen from behind (Unreal see-through),
102 quads twisted > 45 deg (pod fillet, cheek), drum reads as a tyre, drum lamp a 3 cm stud, pod vents short and tile-like,
two empty UI rectangles on the trace panel, stair-stepped light-line edges.
Next: Coder, round 2 fixes; then re-run the ray tests (`critique/round2/cr2_holes.py`, `cr2_collar_rays.py`), clamp, moon pool.

## Coder, critique round 2 (2026-10-22, in progress)
Backup of the state before this round: `versions/v02_before_critique_r2/` (.blend, build/, fbx/, textures/, layout JSON,
previews/compare/, README, STATUS; 92 MB). Geometry done so far (built headless, new file each time):
- M1 collar (`gr_clamp.py`, `gr_finish.py`): the skin is no longer pruned under the collar (its opening is the grid's clean
  r 0.512 ring); every collar section ends in one skirt at r 0.695 fitted to the faceted skin (+5.5, +1.2, -3 mm); the pockets
  have a floor plate from the Kazah wall (r 0.535, crisp corner) to the skirt; pockets sampled every 2.25 deg, chamfers 1 deg.
  Pocket rays (`build/gr_collar_rays.py`, copy of the Critique's): 1-2 grazing hits per pocket (was 336/337 aft).
- M2 drum mount (`gr_fittings.py`): solid convex block 0.34 m wide (SIDE outline (876,397) (913.5,396) (916.3,401) (916.3,450)
  (906,450) (889,440.5)), bolts; two round stubs per side to hubs at (905,407) and the clevis (912,445); round bent tube
  (0.10 m) between them; cheek bracket plate + rails into the bend; hangers, square U tube, cross bars, flange clamps gone.
  SIDE IoU 0.957 (was 0.939; `build/gr_iou.py`, copy of the Critique's cr_iou.py).
- m4 drum: straight body, 0.06 m edges, four raised rings 6 mm proud, 8 mm wide. m5 drum lamp: housing r 0.064, lens 10 cm,
  axis r 0.53 at -25 deg (concept camera ~8 px from (703,358)). m6 pod vents: recessed slots 769.7-797 and 808-830.2 (ends on
  the lip rings). m2: window flanges 0.16 m, lining fitted to trims without flanges, flanges not shortened at convex corners,
  pod lip rings capped (closed), porthole acrylic a slab, loft caps oriented by ray parity (seat fronts). Back-face rays
  (`build/gr_holes.py`): ~330 of 699k (was 1,874; drum 2, was 1,020).
- m3: collar chamfer morph eased, window trims sampled at 6 deg; quads twisted > 45 deg split into triangles (94 quads ->
  188 tris, the critique's fallback; the pod fillet grid itself was not re-pinned). Over 25 deg: 166 (was 254).
- M3 cabin (`gr_cabin.py`): 4 bulkhead screens, wall screen panels between window and door and aft of the door (both sides),
  a 3-screen top row on the dashboard's front edge under z -0.07; view check unchanged (52.3 %).
- Surfaces edits made, not yet run: screens peak 0.52 x 1.8, cabin glow 9, clear acrylic rough 0.02 (no dirt map), strip
  lines inset 3 mm from face edges with 5 mm soft edges, emissive bake at 2x + box filter; decals: separate lifts where
  markings overlap, quads split on the ridge diagonal; traces replace the two empty UI rectangles (m7).
Next: full pipeline (`sh build/run_all.sh r2 128`), close-ups, all checks, README.
- (later) Window flanges now follow the lining 16 mm on the cabin side (projected along the opening's normal), 0.16 m long
  (0.10 under the side windows, 0.05 on the front glass's lower half, never past the cabin's front end wall); the lining's
  opening is fitted to the trims with round 1's 5 cm flanges; each pane a closed 4 mm slab (`glass_slab`, a single refracting
  surface hid the cabin from the concept camera); ledge screens 8 cm lower. Cabin check 0 crossings, view 52.3 % both seats,
  back-face rays ~305. Full pipeline `sh build/run_all.sh r2 128` started (log `build/checks/run_all_r2.log`).

## Coder, critique round 2: DONE (2026-10-22)
Final pipeline `sh build/run_all.sh r2 128` (7.5 min, a new file) + `build/gr_lods_preview.py`; every item of
`critique/round2.md` worked; README.md has the round's table ("Round 2"), the checks table and the Open list.
- Checks: moon pool 1.457 m; carousel side 0.836 / roof 1.204 m (raised 0.786), triangle sweep 0 crossings (both poses); clamp
  86 poses + 3 ship places 0 hits, least 5 mm (hose 73 mm raised); BOARD to POOL 0 crossings (ship 1.248 m, arm 7.1 mm);
  motion 0 hits (hatch 4.1, doors 3.0, antenna 6, rotor 4.9 mm); cabin 0 crossings, least 2.1 mm (flanges under the lining),
  0 seen from outside; view 52.3 % both seats; drivers simple and valid.
- Round 2 tests: pocket rays 1/3/0/1 (were 336/337/4/8); see-through 315 of 699k rays (was 1,874; drum 2, was 1,020);
  decals 0 coincident, 0 quads (498 tris), 0 missing; quads > 45 deg 0 (105 split into 210 tris), > 25 deg 181, > 20 deg 286.
- IoU SIDE 0.957 (was 0.939), TOP 0.948, FRONT 0.954, BACK 0.953, BOTTOM 0.867. Concept camera rms 9.3 px main, 16.3 all;
  the drum lamp 8.3 px.
- Mesh: hull 52,168 tris (25,979 quads + 210 tris), glass 1,944 (slabs), cabin 10,914; LOD1 26,083 / LOD2 10,430.
- Current pictures: `previews/r2_*` (studio, water, topo, cu close-ups incl. `r2_cu_pockets_4.png`), `previews/lods_hull.png`,
  `previews/compare/*` (concept_pair, lookalike, cmp_*, side_inkover, lk_*, iou_diff_*). `r1_*`, `v01_*`, `wip_*` are stale.
- Not done: the pod fillet / cheek grid was not re-pinned (the split is the fallback); see README "Open".
Next: the Critique's round 3 or the user's word.

## Coder, round 3 (2026-10-22, in progress): topology and smooth lines, after the user's "fix the topology and lines needs to be smooth" and "remove those flags"
Backup of round 2: `versions/v03_before_topology_fix/` (made by the Administrator). The flags were taken out at source by the
Administrator (gr_measure FLAGS = False; the decal sheet skips the insignia tiles). Work so far (dev builds into the
scratchpad, never into Tender_Grouper.blend):
- New skin generator (`build/gr_skin.py`, `gr_skin_layout.py`, `gr_skin_mesh.py`, `gr_skin_relax.py`, `gr_skin_build.py`, run
  by `gr_hull.py`): the design lines fitted to the sheet's strokes as fair curves (`build/gr_fair.py` -> `build/fair_curves.json`,
  rms 0.16-0.34 px), planar stations placed by the lines' and silhouettes' turning (107), 96 columns pinned to the grooves,
  FRAME_BOT, the door and the pod (18 below the lower groove, was 12: the nacelle was faceted), a structured nose cap with four
  poles, hole rings round the door, side and front glass, the collar's square warped onto its circle, relaxation with the
  stations kept in order; levels offset in the stations' planes (the stern's boat-tail folded otherwise); the dense surface's
  normals smoothed (zebra steps).
- Trims sampled by their own 3D turning (4 deg, 6 deg in round corners) after a Taubin pass: door and side-glass trims pass.
- Decals: each marking a conforming quad grid (25 mm cells, 1.5 mm out), no staircase (gr_surfaces.py build_decals).
- `build/gr_topo_check.py` -> `build/checks/topology.json`, `lines.json`; `gr_preview.py -- zebra` (striped matcap).
Open while working: the nose's cheek and chin junction, twisted quads at the pod's ends, the grooves' hooks at the stern and
the lower groove's plunge, the hull's triangle count (about 73k, target about 60k).