# Modeling handoff — visible forms before hidden engineering

**A01 is ready to review, not ready to approve as a finished model.** The purpose of this pass is to make the two-vessel source composition inspectable in a coherent local scene, with editable geometry and an explicit boundary between source evidence and invention.

## 1. Review decisions must be separated

Record independent decisions for: source-anchor selection; shared staging and relative scale; ST. ELMO visible silhouette; NIGHTINGALE visible silhouette; proposed identity/name bindings; hidden geometry; and later engineering. A reviewer approving the paired blockout is not authorizing a twin-hull topology, foil mechanism, rescue conversion or production-ready model.

Do not increase detail to hide a silhouette discrepancy. Use the clay or identity view as often as the lighting view. The current rust runoff strips and reflective floor are provisional reading aids, not final look development.

## 2. Source and shared namespace

Primary source is frame 58 (zero-based), 2.416667 seconds in the 24 fps uploaded shot 004. Six frames, including the final decoded frame 96, are supplied to expose changes in view. They are not a completed camera-track solve. The filename's apparent timestamp is not asserted as a verified capture time.

| Anchor | Source feature | Relevant editable part |
|---|---|---|
| S01 | Small red forward point | `GEO_STELMO_RED_MARKER` |
| S02 | Far end of forward amber belt | `GEO_STELMO_AMBER_BELT` |
| S03 | Near forward belt / lower facet junction | `GEO_STELMO_AMBER_BELT` plus adjoining hull stations |
| S04 | Visible aft belt termination | `GEO_STELMO_AMBER_BELT` plus aft envelope |
| S05 | Upper forward side-light termination | `GEO_STELMO_STBD_UPPER_TRIM` |
| N01 | Near aft wheelhouse roof corner | `GEO_NIGHTINGALE_CABIN_ROOF` |
| N02 | Near forward wheelhouse roof corner | `GEO_NIGHTINGALE_CABIN_ROOF` |
| N03 | Aft rubbing rail / silhouette | `GEO_NIGHTINGALE_STBD_RUB_RAIL_02` |
| N04 | Far forward wheelhouse roof corner | `GEO_NIGHTINGALE_CABIN_ROOF` |

The exact source pixels, normalized coordinates, local model points, current projections, thresholds and errors are in `metadata/landmark_review.json`. The map from a manual source feature to a model point is an interpretation and can itself need revision.

## 3. First correction pass — ST. ELMO

### A. S03 near belt junction

The projected near forward belt junction is approximately 27 pixels from its chosen source target. Correct the adjacent station/facet intersection, not only a floating light tube. The lamp path should lie on the seam between the same adjoining surfaces in all views.

Lock the already close far belt endpoint and red point while testing the local near-side correction. Prefer a small local cross-section/forward-station edit over moving the entire vessel: moving the root changes the relative staging that both boats share.

Inspect the dark lower bow diagonals after each edit. Do not infer a catamaran from the frontal dark lobes alone. Keep the existing closed lower envelope as an explicitly provisional hull-volume surrogate until underwater design is separately approved.

### B. S04 aft silhouette and belt

The source aft belt termination remains left/down from the current projection. Review the vessel's apparent length, the aft narrowing, side height and belt slope together. Moving only the light endpoint could pass the point target while preserving an incorrect body silhouette.

Compare in camera match, side orthographic and plan. A correction that fits the source but produces an implausibly kinked aft hull in plan should be flagged, not buried. Model-side dimensional priors may change; source-visible contour gets priority over an earlier invented length specification.

### C. Shell and glazing boundary

The source has a dark glazing band under a pale sloping brow. The blockout simplifies this boundary aggressively; its side can read too monolithic. Resolve the windshield/side-window cut against the source before adding vents or decorative seams. Use separate mesh faces/materials, not source-image projection. Preserve the uninterrupted warm belt as the primary recognition feature.

The raised roof plates and dark inset surfaces have a small separation to avoid coplanar z-fighting. Maintain finite separation or consolidate surfaces deliberately when editing. Check the same part in the browser, an imported GLB and a Blender recreation when available.

### D. Do not add yet

Do not add hydrofoils, propulsion pods, a Kestrel docking platform, an emergency helm, human-sized cargo cells or full interior subdivisions to this approval pass. None is visible enough to constrain this source match. They are later, separate design briefs, not evidence recovered from this clip.

## 4. First correction pass — NIGHTINGALE

### A. N01 wheelhouse roof

The near aft roof corner misses its manual threshold while the forward roof anchors are closer. Test wheelhouse length, roof rake and aft-wall position before changing the entire boat's scale or heading. Keep the three-dimensional roof rectangular/constructed where possible; avoid dragging one corner independently just to satisfy a 2D residual.

Inspect the window-bay rhythm and mullion thickness in the source. The current mesh has clearly separated window planes and frames, but their arrangement is a blockout, not a traced final design. Preserve dark panes and the warm internal strip without allowing window emission to flatten every frame.

### B. Sheer, stern and forward crop

The source workboat's high rising edge and layered horizontal hull construction are more important than the provisional corruption/rust detail. Review the top rail line from aft near-contact through the cropped forward region. The bow is not fully visible; its unseen closure is a hypothesis.

The current strakes use real geometry. When editing hull stations, regenerate the strakes and rub rails so they follow the same hull; do not let them hover or penetrate after a silhouette change. Keep the primary envelope closed while allowing legitimate decorative faces to remain open.

### C. Replace the procedural corrosion rhythm

The current brown runoff strips are deliberately crude and repeat too regularly. They are not a reconstructed repair history. At look-development stage, replace them with a motivated hierarchy: source-visible broad oxidation first, then fastening runoff, edge wear, repair seams, salt/wetness and optional tiny surface detail. Do not add registration or rescue signage until the identity and conversion are approved.

Port and starboard in this blockout are not an approved asymmetric history. The unseen side should stay labeled provisional. Add distinct repairs only after deciding who maintained the vessel and what survives from each era.

### D. No rescue interior claim

NIGHTINGALE is the working proposed name. The source does not establish a medical mission. Stretchers, oxygen racks, patient sockets, cabin partitions and rescue davits have not been authored in this pass and should not appear in an asset inventory as existing geometry.

## 5. Shared stage and camera

Use independent roots for the two vessels; do not merge them into one mesh. Keep their IDs attached through each representation. The narrow source-view gap is an image-space requirement, not a proven physical clearance. The model's relative yaw and longitudinal offset are one plausible artist-fit solution.

The foreground is a wet-slipway/reflection proxy. It intentionally avoids claiming an exact flooded water level or dry support cradle. Delivered preview reflections are a mirrored-geometry approximation, not water simulation. The background includes just enough wall, portal, service strip and pipes to judge occlusion and relative scale.

The camera has a non-central principal point. Preserve its projection through the supplied viewer/config, not by using a generic GLB camera without reading its extras. A new render may be vertically misregistered simply because that off-axis shift was dropped.

Do not use the original moving shot as proof that this single-view fit holds at every time. The next motion task would require tracking source features across multiple frames, testing whether a rigid scene and moving camera explain them, and separating actual view change from any inconsistent forms. This delivery samples continuity but does not solve it.

## 6. Review gates before detail

**Geometry gate:** Both vessel identities remain legible in flat clay. No gaps appear in intended closed hulls. Named material and part boundaries match the design. Primary contours improve in source view without introducing nonsensical unseen geometry.

**Staging gate:** Both vessels share one meter/Z-up frame; individual exports remain origin-centered. No camera-facing geometry tricks are mistaken for established hidden design. The gap, relative height, hull angle and source crop are approved together.

**Correspondence gate:** Recompute the nine point projections. Preserve point definitions across iterations or explicitly version them. Meet all chosen tolerances before describing the point match as passed. Also visually review full contours; nine points are insufficient for whole-shape approval.

**Interchange gate:** Load each GLB independently, preserve metadata and axes, and test the off-axis camera. Execute the Blender builder in a real Blender installation before claiming `.blend` delivery. Import the USD in the intended USD host before declaring that route production-ready.

**Narrative gate:** Reconcile the proposed vehicle IDs and the source scene against the current authoritative atlas. Keep production representations separate from narrative entities. Approve naming, future engineering and rescue/courier roles separately from this visual blockout.

## 7. What an artist should return

Return an updated source-guided geometry revision, the same source-vs-clay comparison, updated nine-point residuals, side/front/plan views, the scene configuration, an exported review decision and the validation report. Preserve A01 as the source for comparison. State precisely which source-visible corrections changed and which unseen design decisions remain provisional.

No declaration of final asset approval should be inferred from a prettier beauty image. The next meaningful milestone is an agreed pair of silhouettes in a reproducible shared frame.