The mission. Give communities that need housing a small reusable kit and an open method that turn their own soil into durable, reinforced, thermally massive walls — with a solar system that powers the build and then stays behind as the village's mini-grid.
The discipline. Every number is computed in a script, cited, or tagged [TO-VERIFY] or [TO-MEASURE];
python3 check_numbers.pyfails if this document drifts from the scripts. Every failed gate changes the plan. Killed ideas go inGRAVEYARD.mdso nobody rebuilds them. v0.1 killed one design by arithmetic, v0.2 corrected four more things, and v0.3.1 corrects v0.3's own numbers (§0). That is the method working, not the mission failing.
⚠ Not yet tested, not engineered, not code-approved. This is a research program. No wall has been built with this method. Structural earth needs a licensed engineer's review; kilns, quicklime, caustic activators, milling dust and suspended loads can injure or kill. Don't build from this document.
A line-by-line review of the scripts found these, all now fixed in the code:
- "The rebar carries 16× the lift" was wrong. The formula let the steel yield, but with earth on the compression side the earth crushes first. Counting that, the steel carries 3.0–5.7× the lift (earth at 1–2 MPa) — still plenty, not 16×. The same correction applies to the surface grid and the no-dig mid-depth bars (§10, §15).
- No safety margin on cracking. v0.3 accepted a 0.28 MPa lift stress against a 0.3 MPa crack threshold. The lift now uses the moist panel's weight and a 1.5 margin: the panel needs a modulus of rupture ≥ 0.45 MPa at lift age.
- Bed suction was missing. A panel cast in a soil pit sticks to it; at breakaway that adds 0.23–0.33 MPa, which can govern. Detailing now breaks the bed before the hoist loads up.
- Quicklime doesn't "boil off" the mix water. Its heat warms the panel ~45–65 °C; the certain gain is the 16–22% of the water it binds chemically. Unslaked lime compacted into a panel can also swell later — now a Phase A gate.
- Costs: the binder (cement or quicklime) wasn't costed, and labor used 80–150 person-days when the build schedule implies 126–222 person-days. Phase A grew from ~7 tests to 17, so its budget grew too. All totals below are regenerated.
- The hoist is sized for setting, not tilting. After the tilt the panel is lifted onto a
raised plinth, so the hoist takes its full weight. Detailing:
making-system/TILTUP_DETAILING.md.
The point is to make the wall from the ground and stand it up. v0.2 drifted into
"compress earth into forms," and put tilt-up in the graveyard on an unchecked one-liner
("earth is weak in the tension a crane lift demands"). tiltup_check.py overturns that
(moist panel at lift, ×1.5 impact):
| panel | lift stress, pick at top edge | lift stress, pick at ~0.71 of height |
|---|---|---|
| 8×4 ft × 6 in (~897 kg) | 0.85 MPa | 0.30 MPa |
| 8×4 ft × 8 in (~1,196 kg) | 0.64 MPa | 0.22 MPa |
| 8×16 ft × 6 in (~3,588 kg) | 0.85 MPa | 0.30 MPa |
Stabilized earth cracks somewhere around 0.3–1.0 MPa [TO-MEASURE]. With the pick at ~0.71 and a 1.5 margin, the panel needs a modulus of rupture ≥ 0.45 MPa at lift age to stay uncracked. If it cracks anyway, four #4 bars per 4 ft carry 3.0–5.7× the lift moment, and the earth only has to grip the bars at ~0.24 MPa. Stress depends on height and thickness, not width, so wide panels are fine with a spreader bar. The open questions are cracking strength, bed suction and bond — three cheap tests in Phase A. Precedent that earth elements can be precast and craned exists (prefabricated rammed-earth walls from Martin Rauch's workshop, e.g. the Ricola Herb Center) [TO-VERIFY]; casting flat and tilting is the new part to prove.
- The binder is fired in a batch kiln, not in the wall. v0.1 showed only ~10–15% of the
clay needs calcining (the binder). Heating a whole wall to fire 15% of it is pointless, so
the binder is calcined in a small insulated kiln, milled, and mixed with raw soil and a
one-part activator before compaction. This removes the 133–888 kWh battery, the in-wall
activator-injection problem, and in-wall calcination shrinkage cracking in one move. The
steel grid becomes plain rebar (optionally a low-temperature cure heater). The in-situ
grid-firing process stays alive as a research branch in
earth-panel/. - CSEB overstatement corrected. v0.1 said compressed stabilized earth block (CSEB) walls are non-structural without a frame. That's wrong: CSEB is used for load-bearing walls, including multi-storey buildings (the Auroville Earth Institute is the classic example). The kit's real differentiators are supply-chain independence, the bundled mini-grid, and an optional cement-free binder — not "structural where CSEB isn't."
- The binder is not where the money is. A cement-stabilized wall needs
17–22 bags of cement per house ($80–670). The calcined-clay binder saves hundreds, not thousands, per house; carbon credits on ~0.6–0.8 t CO₂ avoided are worth ~$6–39. The binder earns its place where cement is scarce, unreliable, or expensive to truck in, and as a climate story. Everywhere else, the kit should use cement stabilization. The kit supports both binders; only the kiln and mill are binder-specific. - Soil decides the binder. Calcined clay makes good binder only from kaolinite-rich clays (calcined-clay cement research points to roughly ≥40% kaolinite for performance near Portland cement) [TO-VERIFY]. Highly weathered tropical soils are often kaolinite-rich, which fits many target regions; smectite-rich soils (common in drylands) are worse. Every site gets a mineralogy test; failing sites fall back to cement.
- The reusable rig is not the big cost. Labor, roofs, the power system, and the field program dominate. Rig reuse saves ~$1k–3.7k per house on the next village.
- Dig a panel-shaped pit 6–8 in deep beside where the wall will stand. The pit is the form; its soil is the panel. No hauling. Removable edge boards keep the panel from bonding to the pit walls.
- Line it with a bond breaker (sand + plastic sheet) so the panel releases.
- Set the steel grid on chairs: rebar mesh with a welded perimeter, lift inserts at ~0.7 of the panel height and on the top edge, and connection plates for joining neighbours and the footing.
- Backfill with the pit's own soil mixed with binder and activator; vibrate and compact in lifts around the grid.
- Cover and cure. Ambient in warm climates, or run current through the grid for a warm cure (~20–41 kWh per 4×8 panel; the grid needs a low-voltage, high-current supply, ~200 A) — the heater idea survives at low temperature.
- Tilt it up with a trolley beam or gantry and a 2 t hoist (an 8×4 ft panel is ~897 kg moist; no crane needed), set it on a raised plinth, brace it, weld or bolt it to its neighbours, tie the tops with a bond beam and roof. Details: §14.
- Bulk: the pit soil, compacted around the grid. Zero firing energy. This is the structure and the thermal mass.
- Binder: 10–15% of the clay, sun-dried, fired to ~700 °C in a batch kiln by day straight off the solar array, milled fine, mixed in with a one-part activator. Or the Roman version (fired clay or husk ash + hot-mixed quicklime, §10). Fallback: 6–8% cement where the soil fails the mineralogy test or cement is cheap.
- Power: one ~20 kW array + ~50 kWh LFP battery. Build phase: runs the kiln, mill, mixer, and tools. Run phase: becomes the village mini-grid.
- Finish parts: roof, doors, windows, wiring, water catchment — dirt can't make these. In dry climates, earth vaults (the Nubian Vault technique) can replace the shipped roof.
| design | energy per house | status |
|---|---|---|
| fire the whole wall in place | ~12,600 kWh, ~28 months on 3 kW | KILLED (v0.1) |
| fire 15% binder inside the wall | ~1,900 kWh, plus big battery, injection, cracking | SUPERSEDED (v0.2) |
| batch-kiln binder, 10–15% | ~600–1,800 kWh | baseline |
- On a 20 kW array (~60 kWh/sunny day available to the kiln): 10–29 sunny days of firing per house, ~100–300 sunny days for a 10-house village. A second kiln plus ~10 kW of panels roughly halves that; panels are the cheapest lever in the kit.
- Run phase: each finished house is budgeted at ~3.1 kWh/day (lights, fans, phones, a fridge share; no air conditioning) [TO-VERIFY per site]. Thermal mass softens heat swings; it does not remove humidity. The array makes ~80 kWh on a sunny day, enough energy for ~26 houses; the battery is the real limit — 50 kWh gives 10 houses ~1.3 days of autonomy. Size the battery to the cloudiest week of the site, not the average day.
| scenario | ships | containers |
|---|---|---|
| ship everything (rig + power + 10 houses of finish parts) | ~21 t, ~74 m³ | 3 × 20 ft |
| regional sourcing (ship the rig; buy solar, battery, rebar, roofing, binder in-country) | ~2.1 t, ~10 m³ | 1 × 20 ft, room to spare |
Regional sourcing is the default: cheaper freight in a volatile 2026 market (the Drewry World Container Index hit $4,639 per 40 ft in July 2026, its highest since September 2024) [TO-VERIFY: primary source], local warranties on batteries and inverters, and no corrosive-goods paperwork for the activator. The honest slogan is "a village in one crate" — the rig ships; the commodities are bought near the site; the walls ship as zero.
| bucket | range | note |
|---|---|---|
| per house, excluding roof | $1,760–7,500 | labor (126–222 person-days at $5–15) and binder choice swing it most |
| roof per house | $900–2,000 steel · $150–500 earth vault | biggest per-house lever |
| reusable rig (once) | $8,400–32,500 | kiln, mill, mixer, tilt gear, compaction, QA kit |
| power system (village keeps it) | $14,000–39,000 | ~16–18% of the budget, becomes the mini-grid |
| field program | $12,300–39,500 | site lead, engineering, lab tests, local review |
| options, not in totals | print-and-tilt printer $3.6–10.5k; grid-cure supply $0.3–1.5k |
First 10-house village (+15% contingency):
| scenario | village total | per house (incl. power) |
|---|---|---|
| steel roofs, regional sourcing | $77k–251k | $7.7k–25.1k |
| steel roofs, ship everything | $89k–280k | $8.9k–28.0k |
| earth vaults, regional sourcing | $68k–234k | $6.8k–23.4k |
| next village (rig reused) | $67k–214k | $6.7k–21.4k |
What the number means: a permanent, reinforced house with a share of a solar mini-grid, somewhere in the high-four to mid-five figures. Whether that is good depends on one comparison only a partner can supply: what they pay today per permanent house, plus what they pay per electricity connection. [TO-VERIFY per region.] The kit wins if it matches that on cost and beats it on supply-chain independence or power.
Phase A (the gate everything hangs on): $3.2k–9.7k. Details in §8.
- Customers can't pay or pay slowly. Buyers are NGOs and governments (track-record procurement, 6–24 month cycles, budgets down about a third since 2023) or low-income households (can't pay $7–25k upfront without housing microfinance).
- Thin, lumpy margins. Integrating a $68–280k village kit at 15–25% margin is $10–70k gross per deal, a few deals a year, with heavy field support and structural liability.
- No moat. The techniques are established and the work is published openly (which makes it prior art). Fine for a charity; fatal for a venture.
- Where a business can exist: local, in-country builders who sell houses with microfinance. That is exactly who the housing-innovation accelerators fund — Habitat's ShelterTech has required applicants to be for-profit with a validated customer base.
- The public good is the open method — process, test data, QA protocol, training package, kit bill of materials. The leverage is in partners who deploy it.
- Deploy through partners, don't run field operations from Arkansas. Local NGOs and builders implement; some can become social enterprises that access for-profit innovation money using the open method. The charity stays a charity; the method still scales.
- Headwind to plan around: international humanitarian funding contracted by nearly a third since 2023, with a 20% drop in 2025 alone; the US and Germany account for nearly nine in ten dollars lost, while Gulf donors increased giving [TO-VERIFY: primary source]. A newcomer competes for a smaller pot, so lead with evidence: a measured strength number, a built demonstration, and a cost-per-house comparison from a partner.
- Start under a fiscal sponsor or as a small 501(c)(3) (the streamlined IRS 1023-EZ route is available to small organizations) [TO-VERIFY current fee and eligibility]. An independent board of at least three is what funders expect.
- Founder-volunteer until a pilot is funded. No salaries in Year 1.
- Optional, only if a commercial arm is ever wanted: SBIR was reauthorized in April 2026 through September 2031, and NSF Phase I awards now go up to $305,000 [TO-VERIFY: primary source] — but SBIR is for for-profit small businesses with a commercial market, not charities. If an existing LLC ever pursues it (e.g., for US low-carbon earth building), keep it arm's-length from the nonprofit: written conflict-of-interest policy, independent board approval of any dealings.
| stage | budget | typical sources | gate to pass |
|---|---|---|---|
| Year 1a — Phase A | $3.2k–9.7k | self-funded, crowdfunding, local Rotary/community grant | strength + mineralogy (§8) |
| Year 1b — demonstration | $10k–40k | small foundation grants, crowdfunding, in-kind | a built, tested, engineer-reviewed structure |
| Year 2 — pilot village | $68k–280k | private/corporate foundations, Rotary Global Grants, Gulf-linked donors, innovation challenges, the partner's own program budget | cost/house ≤ partner's incumbent, or clearly better on supply chain or power |
| Year 3+ — scale | via partners | partners' budgets and local social enterprises | replication by someone other than you |
Year 1b budget: demo structure $5–20k, professional engineer review $2–8k, entity and insurance $1–4k, documentation $0–2k. The demonstration should be the founder's own earth-sheltered homestead build (House #0) or a small accessory structure on the same land — check local permit exemptions for small structures first. [TO-VERIFY locally.]
What every funder will ask for, in order: the strength number, the demo, the partner letter, the cost-per-house comparison. Build them in that order.
- The wall stores no recoverable energy. "Builds and powers a village" means the shipped solar runs the build, then runs the homes, and thermal mass keeps the homes' demand low.
- Kill / pivot conditions:
- Phase A fails on both binders → the soil can't make structural walls this way; stop.
- Calcined-clay binder fails but cement works → keep going, cement-stabilized kit.
- Modulus of rupture at lift age stays below ~0.45 MPa → thicker panels (8 in needs ≥ 0.33 MPa), smaller panels, or a strongback during the lift before anything else.
- Cost per house can't approach a partner's incumbent → publish the open method only.
- No partner will deploy after a successful demo → publish, don't self-deploy abroad.
- Safety: ~700 °C kiln, caustic activator (even the one-part form needs gloves and eye protection), quicklime (exothermic, caustic), dust from milling (respirators), and suspended panels (exclusion zone, rated rigging, §14). The training package is part of the kit.
- Engineering sign-off: structural earth needs a licensed engineer's review at the demo and in-country. ASTM E2392 (earthen wall systems) is the natural starting reference.
- Partner countries: a US charity working abroad must screen partners and locations for sanctions compliance; established partner NGOs usually handle this well.
Two variables, tested separately so a failure points at its cause:
| test | question |
|---|---|
| mineralogy (XRD/TGA, 3 samples) | is the native clay kaolinite-rich enough to make binder? |
| swell test | does the soil expand badly when wet (smectite)? |
| control: commercial metakaolin + native soil | does the native soil work as the bulk? |
| native calcined clay + native soil | does the native clay work as the binder? |
| cement 6–8% + native soil | the fallback baseline — always run it |
| beam bending (modulus of rupture) at lift age | will a panel stay uncracked during the tilt? (need ≥ 0.45 MPa for 8×4×6) |
| rebar pull-out + lift-insert proof load | does the earth grip the steel (≥ ~0.24 MPa with margin), and does the insert hold 2× its load? |
| Roman binder: calcined clay + quicklime, hot-mixed | does the lime-pozzolan version reach tilt strength, and do cracked cubes self-heal when wetted? |
| hot-mix expansion | soak compacted hot-mix cubes at 7 days: any swelling cracks from late-slaking lime? |
| erosion spray + wet–dry cycling | how fast does the panel face wear, per binder? (durability, §10) |
| jar test + shrinkage box | what is the home site's sand/silt/clay split, and does it need sand blended in? |
| galvanized vs plain bar in wet earth | corrosion coupons buried in cast cubes, inspected at 6 and 12 months (doesn't hold up the gate) |
| surface-grid pull-off | cast bars half-buried on a cube face, with and without anchor legs; pull them off |
| electrokinetic box (Route B, <$200) | tub of red clay, two electrode rows, lime water at the anode, 12–24 V from a small solar panel for 2–4 weeks; measure water removed, pH and strength across the section, temperature, energy used |
| till-in-place slab (Route A) | till binder into a 4×4 ft patch of native ground, compact, cure, wire-cut, pull cores for strength |
| printed-mix control | cast cubes of a WASP-Gaia-style mix (soil + rice straw + husk + ~10% lime) beside compacted panels; compare strength and shrinkage |
| floating-shoe compaction | hand-tow a vibrating shoe over 5 cm lifts on a test bed; core it: density and strength vs the pit route |
| rice husk test burn | burn husks at 600–700 °C in the test kiln; ash colour, lime + ash cube strength vs lime + fired clay |
Fire native clay samples in a small test kiln (a 120 V model runs off an ordinary outlet or
an existing home inverter), mill, cast cubes at 10% and 15% binder plus every other mix above,
cure, and send ~80 cubes and cores to a materials lab for compressive strength. Red clays in
the southeastern US are often kaolinite-bearing — a hopeful sign for the home site, confirmed
only by the XRD. [TO-MEASURE] Protocol to be written in chemistry/ (HANDOFF task 1).
The plot tests (till-in-place patches, the shoe test bed) sit outdoors, where slope, sun and drainage
vary across the yard. phase_a_layout.py lays 16 patches out as a 4×4 grid that cancels those
gradients: four on/off factors, each on in exactly two patches of every row and column (a magic-square
layout; Yates's 1937 quasi-Latin square). The four factors are the founder's choice.
Crew of ~6, 17 panels per house: 21–37 crew working days per house (126–222 person-days: site/footing 3–5, dig + cast 4–6, tilt/brace/connect 2–4, bond beam + roof 5–10, openings/wiring/water/render 7–12). The kiln runs ahead of the crew; curing is waiting, not work.
| binder | cure before tilt | one house, elapsed | 10-house village, 2 crews |
|---|---|---|---|
| cement fallback | ~7 days | ~4–6 weeks | ~4–6 months |
| geopolymer | 7–14 days | ~5–11 weeks | ~4–10 months |
| Roman lime-pozzolan | 14–28 days | ~5–12 weeks | ~4–7 months |
House #0 will take longer (first build, learning, testing every step) — plan 3–4 months. Bigger panels (8×16 ft, five per house) cut tilt and connection work if the lift gear allows. Every panel still passes the strength gate before it's lifted (§14).
Honest lifespans (with the detailing below):
| part | expected life | note |
|---|---|---|
| stabilized earth walls | 50–100+ years [TO-VERIFY per binder] | earth lasts centuries when kept dry: rammed-earth monuments such as sections of the Great Wall, the Alhambra's walls, and Fujian tulou |
| steel grid / rebar | decades — the wildcard | earth doesn't protect steel the way concrete does unless lime or cement keeps it alkaline |
| panel joints, lime/earth render | re-seal or re-render every 5–20 years | cheap, sacrificial, planned maintenance |
| steel roof | 25–50 years | earth vaults last longer but need their render kept up |
| solar modules | 25–30+ years | ~0.5%/yr output loss |
| LFP battery, inverters | 10–15 years | the village needs a replacement fund — pay-as-you-go electricity fees are the standard mini-grid answer |
What kills earth buildings is water. Earth builders say a wall needs "good boots and a good hat": a raised, water-resistant plinth/footing so ground moisture can't wick up, and a roof overhang so rain never runs down the face. Those two details matter more than the binder.
The Roman lesson (two parts):
- Roman concrete had no rebar. Modern concrete's main killer is rusting steel; Roman concrete worked in compression (arches, vaults, thick walls) and had nothing to rust. The kit needs steel to survive the tilt, so protect it: a galvanized grid, ≥2 in of earth cover, an alkaline (lime or cement) binder, and dry walls. Zinc corrodes faster in warm water and may stop protecting steel above ~60 °C [TO-VERIFY], so keep a galvanized grid's warm cure at or below ~60 °C. Non-corroding basalt-fiber bar is an option for the body, but it can't double as the heater.
- Roman concrete is lime + pozzolan, and calcined clay is a pozzolan. The Romans even used
crushed fired brick with lime (cocciopesto). MIT's 2023 work, confirmed at a Pompeii
construction site in a December 2025 study [TO-VERIFY: primary source], showed they
hot-mixed: dry quicklime blended with the ash before water, leaving reactive lime clasts
that later dissolve into cracks and heal them. So the kit gets a third binder:
kiln-fired native clay + hot-mixed quicklime. It halves the clay you must fire (lime
supplies the rest), replaces the caustic sodium activator with lime (available almost
everywhere), and may self-heal. The costs: quicklime is its own handling hazard, lime left
unslaked in a compacted panel can swell later (§8 gate), and lime-pozzolan cures slower —
the grid's warm cure earns its keep here. Self-healing in earth panels is unproven;
Phase A tests it.
Lime heat (see
lime_heat.py): slaking quicklime releases ~0.32 kWh per kg, so a 5–7% dose puts ~13–18 kWh into each 8×4 ft panel — enough to lift it ~45–65 °C if none escaped. But it arrives as a pulse during mixing and casting, not over the days of curing, and much of it escapes first. Realistically it keeps ~4–11 kWh per panel and replaces ~10–55% of the grid's warm-cure energy (~70–190 kWh per house), roughly the first day of heating. To keep more of it: mix dry, add the water at the pit, compact while warm, cover immediately [TO-MEASURE]. The heat was paid for at the lime kiln (slaking returns ~36% of the energy it took to make the quicklime), so buy the lime rather than burning your own.
Grid material — "rebar that rusts like a shipping container" doesn't work inside a wall:
| option | embedded in damp earth | can be the heater? | verdict |
|---|---|---|---|
| weathering steel (Cor-Ten, what containers are made of) | its protective rust only forms with wet–dry cycles in open air; sealed in a damp wall it rusts like plain steel | yes | no |
| plain steel | rusts unless lime/cement keeps it alkaline and the wall stays dry | yes | only with alkaline binder |
| galvanized steel | the zinc corrodes first, on purpose, protecting the steel | yes (cure ≤ ~60 °C) | baseline |
| stainless steel | best; several times the price [TO-VERIFY] | yes | connections and splash zone |
| basalt/glass-fiber bar | can't rust | no (doesn't conduct) | body bars where warm cure isn't needed |
| protective current through the already-wired grid (impressed-current cathodic protection) | the solar system pushes a tiny current that stops corrosion; used on bridges and marine concrete | — | research option, not baseline |
Better configuration: the grid on the surface, not inside (the exoskeleton). Lay the grid on
the pit floor so it ends up half-embedded in one face of the panel. Rust then pushes the steel
away from the wall instead of cracking the earth from inside, and the steel stays visible,
inspectable, and repaintable. tiltup_check.py (surface-grid section):
| lift point | steel face | earth-only face |
|---|---|---|
| top edge | 0.85 MPa | 0.00 |
| 0.90 of height | 0.67 MPa (steel carries it, 3.0–5.6× margin) | 0.03 MPa |
| 0.71 of height | 0.30 MPa | 0.29 MPa — too close to cracking with no steel |
- Lift near the top edge (~0.90), not at 0.71. The pit floor is the face that stretches during the tilt, so the steel is exactly where it's needed; lifting high keeps the steel-free face almost unstressed. Expect hairline cracks on the steel face — the steel carries them.
- Anchors, not grip. A half-buried bar needs ~0.67 MPa of grip in the lift — too much to trust to earth. Weld short anchor legs (stainless or galvanized, 2–3 in deep) every foot or so. They are the only buried steel, small enough that stainless is affordable.
- Pick which face gets the steel by where you cast. Tilting about the footing turns the pit
floor face toward the side the panel lay on: cast inside the footprint and the steel faces the
interior; cast outside and it faces out.
- Interior face (recommended): dry indoor air, very slow rust, paintable; or bury it under a lime plaster, which keeps it alkaline and gives a reinforced-plaster skin — steel or polymer mesh bonded to adobe and masonry faces is an established earthquake retrofit.
- Exterior face: now weathering steel (Cor-Ten) actually works — it gets the wet–dry cycles its protective rust needs. Keep it off wet soil at the base, and expect rust streaks down the wall.
- Cure heat: with the grid on the pit floor, heat goes into the ground as well as the panel — lay insulation board or dry sand under the grid so the warmth goes up.
Design rule (the Roman lesson again): each panel should stand as plain earth under its own weight. Steel is for the tilt, earthquakes, and connections — insurance, not the thing holding the wall up. If the steel ever fails at year 70, the wall shouldn't.
The best wall soil is sandy with some clay — roughly 8–20% clay, the rest sand, gravel and some silt [TO-VERIFY per guide]. Pure clay shrinks and cracks; it gets sand blended in. And the binder clay doesn't have to be the wall soil: the kiln needs only 5–15% of the mass as kaolinite-rich clay, which can be hauled in while the bulk comes from the pit.
| soil | walls? | binder, best first |
|---|---|---|
| sandy / desert | yes, compacts well | cement; or lime + a pozzolan |
| sandy-clay, laterite, red clays | yes (blend if >20% clay) | Roman lime + fired clay; geopolymer; lime |
| volcanic soil / ash | yes | lime + the local ash — the original Roman recipe |
| rice-growing silts | yes, with care (erodible) | lime + rice husk ash (a strong pozzolan) |
| expansive clay (swells, "black cotton") | no — blend ≥50% sand or dig elsewhere | lime for footings and paths |
| salty / sulfate soils | test first | geopolymer; cement and lime can be attacked |
| topsoil / organic | never | strip it, dig deeper |
The analysis is data, not code. data/materials.json is the materials library (soils rules,
pozzolans, binders, kiln fuels with heating values, ash yields, firing windows, hazards).
Each site is one file in sites/ (soil test results + what's locally available). To take the
kit somewhere new, copy a site file, fill in the jar test and the local materials, and run
python3 site_profile.py sites/<new_site>.json. It returns the bulk-soil verdict, binder
recipes ranked (confirmed first, lab-dependent ones flagged), a kiln fuel plan with fuel mass
per house, hazards, and the site's TO-MEASURE list. Nine example sites ship with the repo.
Field soil kit (cheap, part of the rig): jar test (shake soil in water, read sand/silt/clay layers), shrinkage box (clay behaviour), ribbon and ball tests, plus lab mineralogy only for the clay that goes in the kiln.
Arkansas is the largest rice-growing state in the US, and husks change the kiln more than the binder:
- Husks are fuel: ~3.6–4.4 kWh per kg burned. A husk-fired kiln co-firing clay at ~650–700 °C needs roughly 130–570 kg of husks per house (~1–6 m³, a pickup load or two) instead of ~300–940 kWh of solar electricity.
- Husk ash is a pozzolan: 15–22% of the husk's weight comes out as ash, and it's reactive if the burn stays ~500–700 °C — the same window as firing the clay, so one burn does both.
- The clay becomes a bonus, not a dependency. If the XRD says the red clay isn't kaolinitic, the recipe still works: rice husk ash + hot-mixed quicklime, with the ash as the whole pozzolan (~3–7 t of husks per house, lots of surplus heat to fire clay for the next house or dry soil).
- The build stops waiting for sunshine. With a husk-fired kiln, the solar array only has to run the mill, tools, and the finished village — roughly half the array [TO-VERIFY].
- Hazards: never overburn (above ~800 °C the ash turns to crystalline silica, a lung hazard, and loses reactivity); grey or white ash is good, black means unburned carbon; respirators for ash dust; smoke control on the burner.
- Portability: rice is grown across South and Southeast Asia, West Africa, and Latin America, and about a fifth of harvested paddy by weight is husk — on the order of 150 million tonnes a year worldwide [TO-VERIFY]. "Farm waste fires the kiln and becomes the cement" travels.
Closest cousins (so nobody can say we didn't look):
- Concrete tilt-up — cast flat, tilt up, brace; widely used for North American commercial and industrial buildings [TO-VERIFY share]. The lifting, bracing, and connection engineering is mature; borrow it (§14).
- Prefabricated rammed earth (Martin Rauch / Lehm Ton Erde) — the Ricola Herb Center (2014) used ~670 factory-made earth elements of 4–6 t each, set like giant stones on mortar, made from local marl with volcanic tuff added for durability. His later unstabilized (cement-free) prefab system won a New European Bauhaus prize and claims ~50% lower cost and ~65% less production time than conventional rammed earth [TO-VERIFY].
- Rice husk ash, calcined clay, lime-pozzolan binders — each studied and used on its own.
- 3D-printed earth (WASP, Italy) — the closest thing to this whole mission, and it uses the Arkansas recipe. Gaia (2018): a site-soil mix with chopped rice straw, rice husk and 10% hydraulic lime, 40 cm walls, ~30 m² of wall printed in about 10 days for ~€900 of materials. TECLA (2021, with Mario Cucinella Architects): two synchronized arms, ~200 hours of printing, soil + water + rice husk with only ~5% binder, a double dome that is wall, roof and cladding at once. WASP pitches a "maker economy starter kit" for local self-build — overlap with this kit is real: partner or reference, don't pretend it doesn't exist. [TO-VERIFY figures]
Printing vs tilt-up, honestly:
- Printing kills formwork and most crew labor, and gives shape freedom — domes and vaults mean an earth roof, which is this kit's biggest cost and crate lever.
- But extruded earth is wet and uncompacted, so it's weaker and shrinks more (hence the straw); printing speed is limited by each layer drying; there's no easy way to reinforce across layers; and the printer is a real capital cost [TO-VERIFY price].
- Compacted, stabilized, steel-framed tilt-up panels should be denser and stronger with cheap tools [TO-MEASURE side by side]. Plausible split: tilt-up walls, printed or vaulted earth roof in dry climates; steel roof where rain is heavy (Arkansas).
What appears new (not found in a quick search; say "we could not find it," never "first"): casting flat in a pit dug from the site's own soil, tilting up with a surface-grid exoskeleton, and firing the binder with farm waste — packaged as an open kit with the power system included.
Why earth lost in the first place: not concrete's price — labor. Rammed earth is slow, hand-heavy work, and concrete won on labor and codes. Prefab and tilt-up attack exactly that.
The longevity claim, stated honestly:
- Roman concrete has lasted ~2,000 years; modern reinforced concrete is designed for roughly 50–100 and often fails sooner, mostly because the steel inside rusts, not because the cement gives out. Survivorship bias applies: we only see the Roman concrete that survived, and it is weaker than modern concrete.
- This kit borrows the three longevity moves: lime-pozzolan chemistry that keeps reacting and can self-heal; no steel buried where rust can crack it (surface grid + stainless anchors); and walls that stand in compression as plain earth.
- Allowed claim: "designed to avoid the main failure mode of reinforced concrete." Not allowed: "lasts longer than concrete" — no one can test 500 years; accelerated erosion, wet–dry, and self-healing tests plus historical analogs are the evidence. Kill-condition: if the erosion spray test shows fast face wear or cracked cubes don't heal, drop the longevity pitch.
- Start with 8×4 ft × 6 in panels. Tilting puts ~948 kg on the hoist line with impact; setting the panel onto the raised plinth takes its full weight, ~1,346 kg. A 2 t manual chain hoist on a trolley beam or rolling gantry does both. 8×16 ft panels need >5 t gear — later.
- Break the bed before lifting. Removable edge boards, sand + plastic bond breaker, jack the free edge 10–20 mm. Suction at breakaway can put 0.23–0.33 MPa on the panel [TO-MEASURE with a load cell on the first panels].
- Lift from the steel, never from the earth. Rebar hairpin inserts tied to the grid: two on the face at ~0.71 of height for tilting, two on the top edge for setting. Inserts ≥ 2× their load, rigging ≥ 5× [TO-VERIFY: OSHA 1926.704].
- Openings go between panels, not in them — a 2 ft window cut into a 4 ft panel nearly doubles the lift stress. Doors and windows are one-module gaps with short sill and lintel panels, spanned by the bond beam.
- Brace every panel at 2/3 height to a ground anchor rated ≥ 2× the brace force (3.7–5.8 kN at 40–50 m/s wind [TO-VERIFY site wind speed]) until the bond beam and roof are on.
- Connections: panel bottom welded or bolted to a steel angle on the plinth; embedded edge plates welded across each joint; a continuous bond beam (reinforced concrete, or a steel ring where cement-free) ties the tops and anchors the roof.
The pit route (§1: dig, mix, backfill onto the steel) is the baseline because every step is
established. The founder's original idea was no dig: push the steel into the ground, treat
the soil where it lies, and stand it up. v0.2–v0.3 dropped that and parked the relief cavities
and injection in earth-panel/. They come back here. Every route first scrapes off the topsoil
(organic soil never binds) and digs narrow edge trenches for the footing and to free the panel.
- Spread the binder powder (e.g. Roman lime + rice husk ash) over the panel footprint.
- Till it 6–8 in deep with a rear-tine tiller. In heavy clay, let it mellow 1–2 days (lime breaks the clods down), then till again with water.
- Compact with the vibratory plate.
- Steel: press the grid into the top surface. (Bars driven horizontally at mid-depth from the edge trenches carry only 1.1–2.0× the lift demand once the earth's crushing is counted — too thin a margin to rely on.)
- Cover and cure (grid heat or ambient).
- Cut it free underneath with a wire drawn through from the edge trenches — the way a potter cuts clay off the wheel — and tilt it up.
With the grid on the top face, lift low (~0.6 of height): the bare bottom face then sees
~0.09 MPa while the steel face takes ~0.55 MPa (nodig_check.py). Equipment added: a tiller and a
wire saw.
Pressure injection fails in clay — clay is too tight for liquid to flow. Electric current moves water and ions through clay anyway, which is how engineers have dewatered and stabilized soft clays since the 1930s. The driven steel becomes the electrodes:
- Compact the surface; put a weight on top (sandbags) to consolidate as water leaves.
- Auger the relief cavities in rows ~30 cm apart. They are the electrode wells and the feed and drain points, and they give the clay somewhere to shrink as it loses water.
- The steel that stays in the panel is the cathode (negative) — which also protects it from rust while current flows. Temporary anodes go in the other row of wells; they corrode, so they are sacrificial or pulled out afterward.
- Feed calcium into the anode wells (lime water or calcium nitrate — not calcium chloride, which releases chlorine gas at the anode and rusts steel). Optionally feed sodium silicate on the cathode side so the two meet mid-panel and gel (an electrically driven version of the two-shot silicate grouting used for underpinning) [TO-VERIFY in literature].
- Run 15–30 V DC straight from a small solar string, no inverter. At 0.5 V/cm calcium crosses the 30 cm in ~3–11 days per pass; water moves toward the cathode ~2 cm/day and is pumped or evaporated out of the cathode wells ("dry"); the current warms the soil +2 to +17 °C/day ("cure"). Three passes: ~1–5 weeks, ~5–184 kWh per panel — a wide range the bench test must narrow.
- Wire-cut underneath, tilt low (grid on top) or as the steel layout dictates.
Honest unknowns (the gates): published electrokinetic work shows big gains in soft clays, but often at foundation grade, not wall grade [TO-VERIFY]. Treatment is uneven (strongest near the cathode, acidic near the anode), gas forms at both electrodes, and the energy range is wide. Kill/pivot: if the bench box can't reach wall grade (~2 MPa compressive, ≥0.3 MPa bending), Route B becomes a footing and floor-slab stabilizer (still valuable) and walls use Route A or the pit.
Till the binder in (Route A), then run the electrodes (Route B) to push lime into the clay clods, pull water out, and warm-cure — without extra digging.
A low gantry builds each panel flat, on its steel grid, in the casting bed, by compacting thin lifts of near-dry mix — not by extruding wet mud upward — then the panel is tilted up.
- Strength: compacted, not extruded, so rammed-earth density instead of wet-mix density.
- No drying wait: compacted earth stands up immediately; flat panels don't stack height.
- Lime at the head: hot-mixed quicklime binds 16–22% of the mix water chemically and warms each lift; extra evaporation from thin uncovered lifts is [TO-MEASURE]. Gate: no late swelling from unslaked lime (§8).
- Stiffness per kg: printed ribs cut ~30% of the earth for the same stiffness, and lower the lift stress by 40%; rice husk packed between the ribs makes an insulated wall.
- Throughput: the mixer (~2 panels/hour of material) is the bottleneck, not the machine.
- Same frame, two more jobs: a tiller head makes it the no-dig machine for Route A, and a hoist trolley makes it the tilt-up gantry (§14).
- Cost: ~$3.6–10.5k DIY [TO-VERIFY], reusable. Build it hand-towed first; motorize only after the shoe proves it can reach rammed-earth density.
The founder's aim is solid rock, not compressed or printed earth. Every route we found for
turning site soil into stone is compared in rock/ROCK_OPTIONS.md: firing, melting, geopolymer,
cement, lime, sulfur and bio-cement. Melting, sulfur and whole-house firing are killed
(GRAVEYARD). The lead is printed, fired, cellular clay units, ground flat, dry-stacked without
mortar, clamped by post-tensioned rods into panels, and tilted up (making-system/BRICK_PANEL.md).
Panels come out at a third of the weight, need no cure wait, burn farm-waste fuel, and the husk
fill insulates. A polygon plan with a fired-unit dome is the long-life form. No tested system of
this kind exists; Phase R bench tests decide it, and the compacted-earth kit stays the fallback.
earth-house-kit/
README.md # start here
MISSION.md # this document (v0.3.1)
GRAVEYARD.md # killed and superseded designs, with the arithmetic
HANDOFF_TO_CLAUDE_CODE.md # task list, priority order
common.py # shared geometry, materials, beam helpers (one place to change a number)
check_numbers.py # runs every script; fails if MISSION.md disagrees with them
system_sizing.py # energy: killed / superseded / baseline designs; run phase
shipping_manifest.py # ship-everything vs regional-sourcing
costs_estimate.py # per house, rig, power, field, village, Phase A
tiltup_check.py # lift, breakaway, steel, bond, openings, hoist, bracing
build_timeline.py # crew days, kiln days, cure waits, village schedule by binder
site_profile.py # portable site analysis: soil + local materials -> recipes, kiln fuel, hazards
lime_heat.py # quicklime slaking heat per panel vs grid cure energy
nodig_check.py # no-dig routes: electrokinetic time/voltage/energy, tilt with steel on top
rock_options.py # every rock-making route compared (data/rock_routes.json)
brick_panel_check.py # printed fired units: panel, rods, kiln, throughput
thermal_check.py # printed cross-sections: 2-D heat flow, ISO 13786 time lag
form_check.py # plan shape: floor per panel, wind, self-bracing
dome_check.py # fired-unit dome: thrust, ring, minimum thickness, dry build
house_designer.py # houses/*.json -> parts list + checks (kit: data/kit_units.json, data/joints.json)
rock/ # ROCK_OPTIONS.md, RESEARCH_NOTES.md
data/materials.json # materials library (soils rules, pozzolans, binders, kiln fuels)
sites/*.json # one file per site (home site + 8 examples)
making-system/ # TILTUP_DETAILING.md, BRICK_PANEL.md, KIT_OF_PARTS.md; kiln, QA (planned)
houses/ # house designs as data
printer/ # print-and-tilt compaction printer: concept + print_check.py
earth-panel/ # sibling research branch: in-situ grid firing (not the kit baseline)
chemistry/ # (planned) Phase A protocol, lab results, binder choice per site
energy-system/ # (planned) array/battery sizing per site climate, village mini-grid
bom/ # (planned) real quotes by region (replace every [TO-VERIFY])
training/ # (planned) safety, build manual, QA for local crews
partners/ # (planned) partner criteria, MOUs, cost comparisons
grant/ # (planned) funding pipeline, applications, evidence pack
Secondary sources below are placeholders: replace each with the primary source before any external use (grant, press, public page).
- Battery prices: BloombergNEF 2025 survey (stationary packs ~$70/kWh; all packs $108/kWh), via ess-news.com/?p=7552; US installed $700–1,300/kWh (EnergySage data, via jouleio.com).
- Solar modules: US median $0.28/W Q1 2026 (Anza, via pv-magazine-india.com/?p=14706); China FOB quotes ~$0.086/W for Q1 2026 loading (pv-magazine-india.com/?p=9622).
- Freight: Drewry WCI $4,639/40 ft on 9 July 2026 (via shipuniverse.com/?p=16195) — cite Drewry.
- SBIR: reauthorized 13 April 2026 through 30 Sept 2031 (Crowell & Moring via mondaq.com); NSF Phase I $305k (via grantedai.com) — cite the statute and NSF's solicitation.
- Humanitarian funding: Global Humanitarian Assistance Report 2026 summary (alnap.org); OCHA 2026 appeal funding via developmentaid.org — cite the GHA report and OCHA FTS.
- ShelterTech eligibility (for-profit, validated customers): habitat.org 2022 call; regional transition 2026 (Columbia SIPA capstone listing).
- Roman hot mixing: Seymour et al., Science Advances (2023), MIT; the December 2025 Pompeii follow-up [TO-VERIFY citation].
- Tilt-up practice: OSHA 29 CFR 1926.704 (lift inserts and rigging); ACI 551 tilt-up guides; the Tilt-Up Concrete Association (TCA) bracing and lifting guidance [TO-VERIFY editions].
- From domain knowledge, verify before external use: calcined-clay kaolinite threshold (LC3 research, EPFL); Auroville Earth Institute load-bearing CSEB; Association la Voûte Nubienne (earth vault roofs); ASTM E2392; NZS 4299 and Peru E.080 (earth building, ring beams); zinc behaviour in warm water above ~60 °C.
v0.3.1 — the wall is cast in the ground and stood up, and v0.3's numbers are now checked by a script. First-order cost: ~$6.8k–25.1k per powered house for the first village. As a business it is weak; as an open-method charity deployed through partners it has the right shape. Everything still gates on one cheap test: Phase A.