# Mako v04 against the user's clean SIDE view: Critique, round 2 (2026-10-20)

**Rule since round 1** (the user): *"maybe just use the side referance you dont have to obey the other wiev pictures"*.
SIDE (`reference/user_side_clean.png`) is the authority. TOP, FRONT and REAR are judged only for plausibility.

**How this was measured**
- Coordinates are sheet pixels (clean x - 1, clean y + 515). `critique/sheet_clean.png` is the sheet with its SIDE panel
  replaced by the clean one; all helpers now read it.
- Model: `previews/compare/lk_side.png` (v04).
- Pictures:
  - `critique/r2_targets_side.png`: target lines over the clean SIDE (top) and over v04 (bottom).
  - Close side-by-side crops: `p2_brow.png`, `p2_cabin.png`, `p2_jaw.png`, `p2_side_mid.png`, `p2_side_rear.png`,
    `p2_stern.png`.
  - `p2_other_views.png`: the model's TOP, FRONT and REAR.
- The silhouette was checked column by column with `critique/outline.py`.

## Where v04 stands
The **SIDE silhouette now matches** to within 1-5 px over almost the whole length:
- nose and canopy (x 45-240), collar, dorsal fin (x 440-560), big fin (x 480-600, tip within 3 px);
- rear body top and bottom (x 560-830), tail fin.

The collar band matches row for row: x 297-338 at y 740 in both.

The silhouette still differs in only four places: the clips (required), the lower front body's rear corner, the ventral
fin's leading edge, and the stern cluster.

What still makes v04 read differently from the drawing is *inside* the outline: the brow and collar area, the grey
values of the mid-body, the glass and cabin, the rear window and the stern.

## Ranked differences (most visible first)

### 1. The brow, collar top and fore clip are a pile of blocks; the sheet has one smooth orange hood (SIDE x 178-340, y 635-760)
- **Sheet** (`p2_brow.png`): the brow and the collar are **one continuous orange piece**.
  - Top edge (the silhouette): (178,702)→(200,697)→(250,684)→(275,678)→(300,672).
  - Its underside flows in one smooth concave arc into the collar's front edge:
    (178,703)→(200,700)→(240,700)→(262,706)→(280,720)→(290,740)→(297,760)→(300,780)→(293,800).
  - So the brow is a wedge that thickens from about 1 px at its tip to about 23 px at x 265 (y 680-703), with a bright
    lit top (≈ (249,201,87)).
  - The collar's back edge is one clean curve: (305,688)→(333,720)→(338,740)→(339,760)→(334,780)→(325,800).
- **v04:**
  - The brow is a 4-6 px plate that reads pale grey at its top (≈ (191,190,177) at x 220-250, y 692-700), with a black band
    under it.
  - It stops at the fore clip post (x 245-280, y 640-700).
  - Behind the post the collar top is stepped: a dark notch at x 275-295, y 685-700, and stacked faceted blocks at
    x 270-340, y 680-760.
  - In the close-up (`previews/v04_studio_clips.png`) the brow is a separate slab with a dark gap and its underside showing,
    and the collar's sides are loose flaps (see "Other views").
- **Change:**
  - Model the brow and the collar as one surface along the two curves above, with no notch, no step and no gap between them.
  - Bring the brow to its full thickness (about 20-23 px at x 260-265) and paint it the shell orange.
  - **The fore clip** (required, it stays): don't stand it on a boxy post. Grow a smooth orange crest out of the hood's top
    line: rise from about x 225, reach the clip plate at x ≈ 262, fall back to the collar top by about x 300, so the steel
    plate is the only hardware showing. The aft clip (x ≈ 350-365, about 25 px proud) gets the same low fairing from the
    shell.
  - This is the single most "wrong-looking" area left in SIDE.

### 2. The mid-body's grey values are inverted: dark cheek, black recess (SIDE x 280-520, y 700-850)
- **Sheet:** the **cheek panel and the big fin are one light blue-grey part**: cheek ≈ (138,153,164), fin ≈ (148,164,177).
  Its top edge runs (280,820)→(342,800)→(392,780)→(431,760)→(470,742).
- **v04:** the cheek renders dark slate ≈ (92,100,110). It is a different value from the fin and reads as part of the dark
  body, so the fin looks stuck on.
  - Rows 800-840: the sheet is light grey from x 343-478 (y 800) and 279-476 (y 820); v04 is dark slate from x 294-506.
- **Recess:**
  - Sheet: the recess (x 340-420, y 705-790) is a **lit mid-grey cavity** ≈ (113,124,134) with lighter inner faces.
    **Only the slot is black**: about 25 px wide at its top (x 422-447 at y 720) and tapering to (392,756) under the
    shell's hook, which comes down to (424-427,760).
  - v04: the recess is **black** from x 334 to 440 (y 720-780), with a thin 10 px slot.
- **Change:**
  - Paint the cheek as the fin's light grey from its top edge down to the fin root, so cheek and fin read as one.
  - Make the recess a mid-grey cavity with lighter inner faces; keep only the slot black, at the sheet's width.
  - Move the dark rear capsule's rounded front edge about 20 px forward: sheet outline at x ≈ 500-513 (y 780),
    480-488 (y 800), 478-481 (y 820); v04 at about 505-525.

### 3. Glass colour and the cabin (SIDE x 42-290, y 700-856)
- **Glass tint:**
  - Sheet: ≈ (143-148, 220, 245), sky-cyan.
  - v04: ≈ (116-130, 161-179, 175-201), teal-grey. It is better than v03, but it still reads as tinted petrol glass.
  - **Change:** push it toward the sheet's value in this render.
- **Cabin:**
  - Sheet: the cockpit reads through the glass as **dark navy masses** in the lower-middle and rear:
    - the seat back and headrest at x 205-245, y 738-812 (headrest top y ≈ 740);
    - a control column at x 125-150, y 785-815;
    - a console from (85,790) to (200,830);
    - the dark rear frame arc just inside the collar (x 265-285).
  - v04: the cabin reads empty, with a small stick at (185,775-800) and a low console. The light grey seat (visible in the
    hero view) disappears against the teal glass.
  - **Change:** paint the cabin interior dark navy or slate so it reads, and check that the seat back comes up to y ≈ 740
    in SIDE.
- **Frames:**
  - Remove the **grey wire hoop** across the glass ((120,738)→(95,790)→(100,830) in SIDE; two silver arches in FRONT and the
    hero).
  - Remove the **dark nose cap** (x 40-60, y 775-815). The sheet's glass is clear to the tip, inside a thin black outline.
  - The sheet's only line on the glass is a thin light inner-skin line about 8-10 px inside the outline.

### 4. The rear glass bay is too long and too bright (SIDE x 615-880, y 760-835)
- **Sheet:**
  - The bay runs **x ≈ 645-800**, y ≈ 770-832.
  - Its front end is a curved, sloped "nose": (645,828) up to (700,770).
  - The glass reads **dark** ≈ (67,83,94), with blue only in its front third; raked posts at x ≈ 700 and 745.
  - **From x 800 to 880 the side is machinery** (dark boxes, panels), not glass.
- **v04:**
  - The bay is light blue (≈ (116,161,175)) from x 615 to **875**, with a stepped lower front corner (615-650).
  - It is now the brightest thing on the rear body.
- **Change:**
  - End the glass at x ≈ 800 and give the front the curved nose.
  - Tint the glass dark, with highlights only.
  - Fill x 800-880 with stern machinery.
  - The posts are right; keep them.

### 5. The stern cluster (SIDE x 860-955, y 715-875; `p2_stern.png`)
- **Post:**
  - Sheet: a short flat vertical plate at **x 912-918, y 740-785**.
  - v04: a thin rod at x ≈ 905 rising to **y 721**, 20 px above the sheet's.
  - **Change:** lower it and make it flat.
- **Orange part:**
  - Sheet: an angled **orange lens at x 863-880, y 797-820**, on the side of the block.
  - v04: a face-on orange **ring** at (905,800).
  - **Change:** use the lens. If REAR needs the rings, turn them so they are edge-on in SIDE.
- **Tool arm:**
  - Sheet: the upper arm at **y ≈ 795-805** reaches **x 950** and ends in an open pincer (x 940-950, y 790-812).
  - v04: the tool sits at y ≈ 825 and ends at x ≈ 945, with no upper arm.
- **Lower unit:**
  - Sheet: x 875-930, y 815-**852**.
  - v04: the box bottom is at y 870, with a knob to 875.
  - **Change:** raise it to 852 and drop the knob.

### 6. Jaw details (SIDE x 120-372, y 836-990; `p2_jaw.png`)
- **Prong slit:**
  - Sheet: the jaw is a solid triangle down to y ≈ 945 (row 940: orange x 243-289). The slit between the two prongs starts
    at about **(282,948)** and is **about 6 px wide** (row 960: 262-284 | 291-305).
  - v04: the slit starts at about y 920 (x ≈ 246) and is 12-18 px wide (row 940: 241-255 | 274-286), so the prongs read as
    two rods hanging under a cut-off jaw.
  - **Change:** shorten the slit and narrow it.
- **Grooves:**
  - Sheet: two thin curved "<" strokes, 6-8 px wide (the `GROOVES` polylines are right).
  - v04: wider straight bars (e.g. x 252-266, y 851-896), plus a **third short groove at x ≈ 145-152, y 862-882** that the
    sheet doesn't have.
  - **Change:** narrow and curve the two grooves; delete the third.
- **The C's upper arm:**
  - Sheet: tapers to a point at (368,840).
  - v04: a square-ended bar at x 300-370.
- **Colour:** in this light v04's jaw is ≈ (145,71,53), brick-brown (s 0.63); the sheet's is (198,78,32) (s 0.84). The same
  under-saturation holds for the shell: (186,124,71) against (243,138,39).
  - Part of this is the comparison light: the side faces are lit weakly. Light `lk_side` from the camera side before judging
    paint, then raise the paints' saturation if they still read brown.
- **Edges:** the sill under the glass is a heavy dark slab at the front (x 140-180, y 843-865). The sheet's is a thin dark
  line at y 848-856. The front edge also gets a dark chamfer band; the sheet has one crisp edge.

### 7. Lower front body: rear corner and skid (SIDE x 330-470, y 900-960)
- **Sheet:** the bottom runs at y ≈ 953 to x ≈ 405, then **rises in a rounded corner**: (425,940)→(440,909)→(455,890) into
  the fin root (silhouette at x 440: bottom y 909).
- **v04:** the bottom stays at y 956 to x ≈ 450 (silhouette at x 440: y 956). It also carries a black **skid bar** (x 330-440,
  y 950-955) with a small post at x 438 that the sheet doesn't have.
- **Change:** round the corner up as above and remove the skid. The belly box's skid (y ≈ 905, x 515-650) is not drawn
  either: the sheet's box bottom is y ≈ 895.

### 8. Ventral fin's leading edge (SIDE x 830-880, y 845-950)
- **Sheet:** a straight leading edge from **(835,858)** to the tip (868,948) (silhouette at x 850: y 903).
- **v04:** the root starts about 15 px further aft and the edge is concave (at x 850: y 880).
- **Change:** move the root forward and straighten the edge.

### 9. Small
- **Rear body side:** the sheet has a light-grey diagonal strip from the cheek's rear top (≈ (490,780)) up and back to the
  shell edge (≈ (650,752)), and a darker recessed side panel with a rounded front (x 500-650, y 760-850). v04's rear body is
  one flat dark surface.
- **Tail fin:** the sheet's flat top spans x 848-868; v04's is narrower (≈ 850-860). Its leading-edge root is at x 835 in the
  sheet and 828 in v04.
- **Big fin:** its root panel has facet lines in the sheet (a ridge from (375,830) to (440,790)); v04's is plain. Optional.

## Other views: implausible or broken (not judged as mismatches)
- **The collar is in pieces** (hero, `v04_studio_clips.png`, TOP, FRONT, REAR).
  - Its sides are separate flat orange flaps with gaps between them and their grey or white undersides showing.
  - They stick out sideways past the body: "ears" at the collar in TOP, spikes at the shoulders in FRONT
    (≈ (1075-1100,170-190) and (1300-1335,170-190)), and the same in REAR.
  - Grey and white slivers poke out under the starboard flap (clips close-up, lower right), which looks like
    interpenetration.
  - **Change:** make the collar one closed band round the canopy's rear, with a smooth outer edge and nothing standing past
    the body. The same fix as item 1.
- **Brow plate:**
  - In the clips close-up a dark gap and the plate's underside show between it and the glass. Seat it on the canopy frame
    and close the gap.
  - In TOP it is a large orange arrowhead over most of the canopy. That is acceptable, since the fore clips at ±0.38 m need
    it, as long as it is sealed.
- **Glass:** the two silver wire arches over the dome (FRONT, hero) read as handles. Remove them, or make them flat frames
  flush with the glass (item 3).
- **REAR:** plausible (grey lobes, the stern hub with its fan duct, flush hatch outline).
- **TOP:** plausible apart from the collar ears.
- **FRONT:** the dome is round and the jaw and prongs read well; plausible apart from the collar spikes and the wires.

## Verdict
SIDE is now close in **outline**, about 90 %: silhouette within a few pixels almost everywhere, the canopy shape, the collar
band, the jaw with its C, the fin and the tail all in the right places. It is about two-thirds of the way there in
**look**. The changes that would do most, in order:
1. **One smooth orange hood:** the brow and collar as one surface along the sheet's curves, the brow about 23 px thick at
   x 265, with no notch, gap or loose flaps; the fore clip faired into it as a crest. This fixes the worst area in SIDE and
   the broken collar in every other view (item 1, "Other views").
2. **Grey values:** the cheek in the fin's light grey, the recess mid-grey with only the slot black, the dark capsule's front
   moved forward to x ≈ 480-490 (item 2).
3. **Glass and cabin:** sky-cyan glass, a dark navy cabin with the seat reading at x 205-245; no wire hoop, no nose cap
   (item 3).
4. **Rear window:** end the bay at x ≈ 800 with a dark tint and a curved front, and put machinery from 800 to 880 (item 4).
5. **Stern cluster to the sheet:** a low flat post, the orange lens, the upper pincer arm, the lower unit raised. Then the
   jaw's slit and grooves (items 5, 6).