Research

Each of the five staged prototypes is an order of magnitude cheaper than a false start, and each carries a published kill criterion.

2026-07-11 · Kult research

The thesis to prove is narrow and falsifiable: a software pipeline plus a low-part-count, tolerance-tolerant design lets a pair of novices assemble a structurally sound, weather-tight room from machine-cut parts. Software is cheap to test; physical builds are not. So the physical proof is a ladder of five artefacts, ordered so that each rung is a cheap kill-gate for the next. The result publishes either way.

Finding

Structure does not scale-model: the square-cube law, fixed lumber dimensions and real joint tolerances void small-scale structural tests. What can be tested cheaply at small scale is the software and the mistake-proofing logic. The ladder therefore splits the proof: two cheap rungs (a printed 1:10 model, a machine-cut joint sample) test files, fit and orientation logic; three full-scale rungs in real off-the-shelf timber (a datum rig, a one-wall raise, an instrumented room module) test what only full scale can: load, tolerance absorption and novice assembly. Each rung has a named pass gate and a named kill criterion, and a rung only runs after the one below it has passed.

The five rungs

Stage 1 — 1:10 printed model of the whole house

The reference house, generated by the software and 3D-printed. Tests: the files are geometrically complete, the documented assembly order works, and a novice can follow the instructions at model scale. The cheapest possible falsification of the pipeline.
Stage 2 — mistake-proof joint sample in real timber

Machine-cut C24 joints from our files, the first time a real cutting service consumes them. Tests: file intake, cut tolerance, and whether the mistake-proofing (chamfers, hole-pattern keys, marked parts) survives real wood.
Stage 3 — datum and first-course rig

Full-scale foundation datum and first course. Tests the two error sources self-build precedents actually stumble on: a level base and a correct start.
Stage 4 — one-wall build-and-raise day

A full wall section built flat and raised with the designed mechanical raising aid. Tests the heaviest single operation, the first-article fit of a real assembly, and the instruction set at wall scale.
Stage 5 — instrumented room module at full scale

One room of the reference house, built by a novice pair from instructions alone, then load-tested incrementally. The acceptance test for the whole thesis: structure, envelope (insulation, airtightness, moisture) and assembly-hours data.

What would kill the idea

Kill criteria, standing and published in advance
Criterion	Where it is decided
The structure only stands with skilled intervention or hand-fitting	first falsifiable at Stage 4; fully at Stage 5
Cutting tolerance cannot be absorbed by any practical joint design	first data at Stage 2; decided by Stage 4's first-article fit
No cutting service accepts our files at a workable price, and the fallbacks also fail	the live supplier quote table + real orders at Stage 2/Stage 4
Labour, extrapolated to a full home, blows past H ≈ 120 + 16·A person-hours	measured at Stage 5
A rung's cost estimate inflates past ~2× before commit	stop and re-run the options analysis, any rung

The assembly gate is deliberately strict: a novice pair assembles everything under the ≤ 30 kg part-weight cap from the instructions alone, with no verbal help. An irreversible or structural mistake fails the run; a recoverable hesitation is logged as an instruction bug. Raising steps use a designed mechanical aid, because two people are below the safe headcount for a free wall raise.

What this does not show, and which risks rank highest

Nothing physical has passed yet. The ladder is the plan and its gates; rung results publish here as they land, pass or fail.
The top-ranked risk in our own register is that a software founder cannot self-review embedded engineering knowledge; carpentry ranks below it. Our validation-realism review already found real engine errors, including one that under-counted wind load. The standing mitigation is an independent timber engineer reviewing the engine's assumptions before Stage 5 runs, a review Swedish law in any case requires for real builds.
The single unverified commercial link is file intake: whether existing precut services will cut from our machine files at a workable price. Inquiries with real files are out, and the quote table decides it rather than theory.
Stage 5 proves that the house is buildable. Whether it is sellable or permitted stays open: permits, land, liability and market demand are explicitly outside the prototype's scope and tracked as separate, later risks.

Sources

Kult MVP definition — thesis, ladder, acceptance test and kill criteria (internal, founder-ratified 2026-07-06).
Kult internal research, "Prototype risks and mitigation" (2026-07-10) — the ranked risk register behind this paper.
USDA Mutual Self-Help Housing programme — the strongest precedent that novice groups complete builds (with a paid site leader): rd.usda.gov
CB Insights — why startups fail (market need ~42 %, capital ~70 % proximate): cbinsights.com