Sources

Sources

Every dimension the house is generated from is below, with its class and its citation. This is the project’s honesty layer, and it is given a room of its own because a smooth 3D render implies a precision the sources do not have.

The dimensions

Bibliography

• forth-1981 · ethnography

  Forth, G. L., Rindi: An Ethnographic Study of a Traditional Domain in Eastern Sumba (Martinus Nijhoff, The Hague, 1981).

• hoskins-1998 · ethnography

  Hoskins, J., Biographical Objects: How Things Tell the Stories of People’s Lives (Routledge, New York, 1998).

• waterson-1990 · ethnography

  Waterson, R., The Living House: An Anthropology of Architecture in South-East Asia (Oxford University Press, Singapore, 1990).

• depdikbud-ntt · reference

  Departemen Pendidikan dan Kebudayaan, Arsitektur Tradisional Daerah Nusa Tenggara Timur (Proyek Inventarisasi dan Dokumentasi Kebudayaan Daerah, Jakarta).

Which of these matters most

Nearly every dimension on this page is the author’s own. What the provenance bar does not say is whether that matters. Each row below is computed by rebuilding the house: one dimension is pushed 20%, and the house is measured to see how far it follows. Nothing is estimated — the figures are the difference between two runs of the generator.

1. 1menaraRise2.90 mheight of the peak
2. 2shoulderRise1.83 mheight of the peak
3. 3wallHeight1.34 mheight of the peak
4. 4coreSpanX1.24 mwidth of the eave
5. 5floorHeight1.06 mheight of the peak
6. 6banggaDepth0.68 mwidth of the eave
7. 7eaveOversail0.60 mwidth of the eave
8. 8eaveDrop0.35 mheight of the peak

If only one dimension could be surveyed, this is the one. This is sensitivity, not uncertainty: it shows what moves, not how likely the value is to be wrong. The 20% is itself the author’s choice.

Invariants

There is no measured drawing yet, so correctness rests on structural truth. These checks fail the build.

• pass
  Bilateral symmetry about The building mirrors about z = 0

  1776 points; largest mirror deviation 0.00 mm.

• pass
  Every tenon contained in its mortise

  14 pasak, takik and tumpu joints; every one engaged.

• pass
  No joint spans more than one stage

  Every joint engages two parts from the same stage or from consecutive ones.

• pass
  Stage ordering, and nothing below ground

  140 parts, every one bearing on something already standing or on the ground.

• pass
  Mesh integrity: indices, triangles, normals

  664 triangles; indices in range, none degenerate, every normal unit length.

• pass
  Four kambaniru, each named, each at its own corner

  Kambaniru uratungu, Kambaniru uma, Kambaniru padua, Kambaniru mata. The only house in this project whose posts are individuals rather than members of a rank — elsewhere the seventh post is the same thing as the eighth.

• pass
  The tower holds something, and what it holds is inside it

  A tower 14.52 m tall with the uma deta at 5.76 m inside it. On the other seven houses here the roof shelters something; this one contains something — and it is the only check in this project that tests a purpose rather than a form.

• pass
  The loft goes in before the tower that exists for it

  The loft first, and the tower built around it. Reversing them produces a byte-identical model — what changes is what the raising sequence says the building is for, which is why this needs a check rather than a comment.

• pass
  The tower stands taller than the house beneath it

  A house of 5.76 m under a tower of 14.52 m — 2.52 times. That proportion, rather than any dimension, is what makes the Sumba silhouette recognisable; and the figure that sets it is the least supported number in this pack.

• pass
  The top of the tower is flat rather than pointed

  A top of 2.21 × 2.41 m, and that is where the finials stand. A form tapering to a point would remove the place the loft occupies — so the flat peak follows from the purpose rather than from taste.

• pass
  The bangga runs all the way round, or along the two long sides

  A full circuit: a house that receives on every side.

• pass
  Thatch courses lap with no bare band, from the eave to the peak

  77 courses over 18.68 m — past the shoulder and on up the tower, unbroken where the form changes.

• pass
  Every part declares the dimensions it came from

  140 parts, every one tagged. By least-sourced input: 0 measured, 0 canon, 140 interpolated.

• skipped
  Checked against a measured drawing of a real house

  Skipped. No measured drawing has been wired in, so there is nothing to compare against. This check will not be made to pass by weakening it.

What a check is preventing

Ten green rows are not a reason to believe anything. So one check is run against a house built to break it, and what is printed below is that check’s own verdict rather than a description of it.

Shrink the ratio that sets the tower’s height and nothing collapses, nothing is zeroed; the loft is still inside the tower and the thatch still covers it. What stops being true is that the building is a container with a house at its foot — below a certain height it becomes a house with a tall roof, a different kind of object with a different reason for existing. Nobody could look at the model and say which side of the line it was on without being told what the line was for. After eight houses the shape of failure is the same as ever — a rule that cannot be carried out rather than one disobeyed — except that here what fails is not the geometry but the claim the geometry makes.

The second house is never rendered. It is built during the build, failed, and thrown away.

This bar moves as real surveys are wired in. That movement is the project's progress metric.