Refrhus — construction history & model ground truth

The interpretation key for Home.xml. The model encodes a house that grew in stages, so the levels are deliberately different shapes — apparent “misalignment” between levels is usually real construction history, not a tracing error. Read this before trusting (or “fixing”) any cross-level overlay.

Captured from owner knowledge on 2026-06-13, alongside the first overlay assessment (see realms/realm-siliconsaga/docs/plans/2026-06-13-sh3d-overlay-tool-design.md).

Build sequence

1. Original house — roughly 800 sqft. The current Basement is the correct shape of this first edition; the original main-floor exterior walls stood directly atop the entire basement perimeter.
2. Garage added — not original. Sits at an odd split-level height (~4 ft, roughly midway between basement floor and main floor). On the survey it is bundled with the house with no wall drawn between garage and house.
3. Two main-floor extensions:
   • Kitchen extended north (~8 ft up on the diagram), then a deck added north of that. The little wall stump by “DINING AREA” marks where the original exterior wall started, running east.
   • Primary bedroom extended south. The original exterior wall crosses where “COVERED PORCH” begins — i.e. the porch region is the southern extension.
4. Upstairs finished — awkwardly. The 2nd floor was expanded south over the extended bedroom (ending in a gable with a window, not the original sloped roof). It was not expanded north over the kitchen extension.

Levels & elevations (from Home.xml)

Heights are correct; the split-level/extension reality is already encoded in the elevations. The only thing needing work is in-plane (X/Y) frame consistency.

Why the Main and 2nd-floor north walls are ~8 ft apart

It is in-plane, not a level-height difference — the kitchen’s ~8 ft north extension:

• The basement north wall is the original footprint’s north line.
• The 2nd floor’s north wall sits essentially on that same line — measured, 4.9 in south of the basement’s, which is wall-thickness order rather than an offset to model around.
• Main extends ~7.6 ft north of it, and that extension is the whole difference: the kitchen addition, continuous with the garage and crawlspace north edge.
• So the Main and 2nd-floor north edges are not supposed to match. Measured from the model: 95.7 in / 7.98 ft / 243 cm apart at the north, while the south edges sit within 4.5 in of each other.

Basement framing — the joist datum (measured 2026-07-23)

The basement floor joists are now the most confident geometry in the house — relative spacing measured to ~1/8″. Joists run N–S, spaced along X (E–W), numbered west→east (joist-01 … joist-25), and are modeled on the basement-main-transition level (they sit atop the basement walls). Two ~3.5″ post-width main beams and one extra joist sit in the stair area (tighter spacing). Real-world caveat: some eastern joists drift (not perfectly parallel — up to ~½″ over a few feet); measure at a consistent Y. Per-bay spacing + the derived basement width: basement-joists.md. The basement is not a uniform box — zones, staggered posts/beams, and per-wall edge conditions: basement-structure.md.

What sits on top of the cinderblock (standard platform-framing perimeter):

• A flat plank laid atop the block, ~centered, shorter than a joist — the sill plate (mudsill): pressure-treated, anchor-bolted, the code-required wood-to-concrete interface. Along the E & W walls it appears to “support nothing” because the joists run parallel to those walls and bear on the N & S walls instead — so the E/W sill is just the perimeter plate (and a handy nailer; wires get stapled to it). Present on N & S walls’ eastern half too.
• An on-edge member at the outer edge of the block — the rim / band joist (a.k.a. the outermost joist), capping the floor framing flush to the block’s outer face. This is the common, expected pattern.

East vs. west wall edge conditions differ (important for the X frame):

• East wall — confident. Raw cinder is visible; a true 1″ clear gap from the block’s inner face to the first joist’s east face. Anchor the model here.
• West wall — fuzzy. Wood paneling/furring hides the block; ~8″ (estimated) from where the block likely ends to the first joist’s west face. Confirm the raw west block face on the scan visit before trusting the total basement width.

SH3D anchoring, so the model matches reality:

• A wall’s coordinate is its centerline — an 8″ cinderblock extends 4″ each side. To leave a gap to a face, place the line ½·thickness beyond it.
• Furniture X is the piece’s center — a joist face is x ± ½·width (±0.75″ for a 1.5″ joist). The joists were entered aligned to their left (west) edge, so every stored center is ~0.75″ west of true — a uniform, correctable shift (nudge all joists +0.75″ E so stored X = true center; relative spacing is fine).

Datum workflow: treat the joists as ground truth → position the E/W basement walls to the outermost joists using the real edge conditions (east 1″ + ½·wall; west ~8″ + ½·wall) → the basement width falls out of the result (no independent width measurement needed; its uncertainty lives entirely in the fuzzy west side). Then, because the E/W walls align across levels (next section), that corrected basement X frame is the anchor the Main and 2nd floors register to.

Alignment invariants (what should line up across levels)

Use these — and only these — to register levels into one plan frame:

• East & West walls align across Basement / Main / 2nd floor (the original perimeter). Exceptions: the garage, and the south (extended) part of the 2nd floor.
• Staircase roughly aligns first ↔ second (scale / exact placement may be a little off — verify, don’t assume).
• North & South walls align too — but only on the original perimeter. See the clarification below.

Do NOT expect the extensions’ north/south walls to align with anything below them — they diverge by era (kitchen +8 ft north; bedroom/2nd-floor extension south; knee-wall inset). That is real construction history, not tracing error.

But the original perimeter is one wall stack, all the way down, on every side. Where the Main floor still sits on the original footprint, its north and south walls do register to the basement block:

• The north wall of the main-floor bathroom and kids room follows the original perimeter — it is not an extension, and it should land on the basement’s north block line.
• Same for the stump near the middle of the kitchen’s west wall, which is north-aligned with the original wall, and the corresponding stump on the west wall by the main bedroom extension.
• The two extensions themselves (kitchen north, bedroom south) need their own measurements and must not be dragged onto the basement footprint.

Clarified 2026-08-23 by the owner, after the blanket reading nearly stopped a justified correction. The rule is about eras, not about compass directions.

Framed wall vs. cinder block — where the exterior face goes

Derived 2026-08-23. It had not been written down, so the upper floors drifted without anyone being able to say by how much.

A framed exterior wall’s OUTER face sits on the block’s OUTER face. It follows from the rim/band joist above: that member caps the floor framing flush to the block’s outer face, and the wall’s sheathing continues the same plane. So an exterior wall of thickness t on a level above has its centerline at block_outer_face ∓ t/2 — not on the block’s centerline, and not on its inner face.

Two consequences follow, and both are easy to get backwards:

• The wall thicknesses differ by level (8″ block, 7″ framed), so aligning centerlines across levels is wrong by 0.5″ and aligning inner faces is wrong by 1.5″. Only the outer faces coincide.
• The rule may run a siding-thickness shy of exact, since siding usually laps the foundation with a drip edge. Treat sub-inch residuals as noise, not error.

Anchor east, not west, and never the average. The east gap to raw cinder is a confident 1″; the west is a ~8″ estimate through paneling, so the basement’s own west face carries all the uncertainty in its width. Registering an upper floor onto the east face puts the known end of the datum against the thing being corrected.

Two registration seams

1. Floor-plan levels vs. each other — Basement / Main / 2nd all trace the same not-to-scale floor-plan bitmap (image 0) at independently eyeballed origins. Align via the E/W walls + stairs.
2. Survey frame vs. floor-plan frame — the Garage uses the authoritative survey (image 1) in a separate coordinate frame; it does not currently agree with the floor-plan frame. The survey is authoritative for the true exterior footprint; the floor plan is explicitly not to scale (good for interior layout only).

The parcel, from the survey

The lot as surveyed 2022-09-23. Recorded here because the survey endorsement that carried it has been removed from the repository — it was title paperwork wrapped around four useful numbers, and these are the four numbers.

A parallelogram, not a rectangle — 125 ft deep by 68.15 ft wide, matching the 125.00' and 68.15' dimensions printed on SurveyAlone.png, but with interior angles of 92°26′ and 87°34′. Opposite sides are equal and parallel; the corners are not square. Anything that squares the house to the lot boundary inherits that 2.43° error.

The frontage bearing is the one that matters beyond the footprint. The street line runs N 48°05′ W, so a front elevation square to the street faces roughly S 42° W — which is the source of the “front faces roughly southwest” claim used in the window and solar-gain reasoning. Anything that re-derives orientation should reconcile against this rather than against the floor plan, which is not to scale and carries no true north at all.

Not recorded, deliberately: the street address, deed book reference, tax lot and block, title file number and surveyor details that accompanied these bearings. None of them affect any measurement, and this repository is public.

Background-image facts (for any registration math)

Placement, taken verbatim from PlanComponent.paintBackgroundImage:

• plan = pixel · scale − origin; image top-left sits at plan (−xOrigin, −yOrigin).
• scale = scaleDistance / dist(scaleStart, scaleEnd) (cm per image-pixel).
• Plan Y is down; SH3D background images cannot be rotated (the two scale points only set a scalar scale) — any rotation must be baked into the image file.