What the extraction research got wrong about the world

The project's resource research describes how 2,435 people at Richards Bay in late March 1628 would take plant food, game, fish, shellfish, firewood, thatch, timber and water off the landscape. Read as an ecologist, a geographer and a coastal engineer would read it, eight of its load-bearing claims fail. Each is set out below: what the research says, the evidence against it, the reasoning, what it changes, and the sources opened.

Checked 30 September 2026 · two independent passes, one findings-only and one that rewrote the research layer · unratified under D-121

What "the research" is. Five layers, each built on the one before.

  1. Research runs, 3 August. Twelve machine-generated reports on the site's resources. Four carry this page: herbivore biomass (out_05), wild food and fisheries (out_06), timber (out_07), construction materials (out_11). Their cited sources were never opened.
  2. Two syntheses, 19 August. The same brief given blind to two models. Each turned the four runs into one model of how the colony takes plant food, game, fish, shellfish, firewood, thatch, timber and water off the land.
  3. The reconciliation, 19 August. A third document merging the two, ruling on where they disagreed, and adding a measurement of how much of the ground around the campus is land.
  4. Two reviews, 4 September. Arithmetic, labelling and the measurement were checked. They were mostly right; one bug counted open ocean as land and was fixed.
  5. This check, 30 September. Is any of it wrong about the real world? The findings below are the answer. The reconciliation is now rewritten against them, so it is the current statement of the whole stack, now named extraction_system_2026-09-30.md; the layers beneath it are unchanged, and the corrections to each are listed at its end.

How this was done. The research runs were machine-generated from sources nobody in the project had opened. Two independent sessions read them against the ground: one listed what was wrong and cited what it opened; the other re-opened those sources, verified them, added its own, and rewrote the research layer in place. Where a figure could be checked against a source that could actually be opened, it was. Where it could not, the claim is marked unverified rather than replaced with another guess. Internal arithmetic, labelling and citation discipline had been reviewed twice already and are not what this page is about.

Water within 10 km of the campus: lake to the north, harbour and lagoon to the south-west, ocean to the east, land only to the west and north-west
The ground as it is. Lake Mzingazi 500 m north. The water to the south and south-west is the modern harbour, which in 1628 was a shallow lagoon (finding 5). The coast to the east is sand, not rock (finding 6). Land runs west and north-west, and even that is cut by the lake's outlet stream at about 2 km (finding 8). The campus has water on every side but one.
1

Game meat is lean, so the ration does not add up

What it reacts to

Layer 2, the Claude synthesis priced edible game at about 2,000 kcal per kilogram and concluded from it that a square kilometre of game is worth 2–11 times a square kilometre of plants. Canon, the famine document, your own file books 1,300 kg of edible meat a day as 2.6 million kcal and makes that the Phase-1 ration of 77–80% of full. Both rest on the same 2,000 figure; neither sourced it.

Evidence

  • USDA composition data for wild ungulate meat, raw, per 100 g: antelope 114 kcal with 2.0 g fat; water buffalo 99; deer 120; elk 111.
  • Offal, the parts canon §6.1 adds to reach its 60–65% edible yield: liver 135, heart 112, kidney 99, lung 92, tripe 85. Skin and blood are lower still.
  • Pure suet is 854 kcal/100 g. To average 2,000 kcal/kg across the edible stream, about an eighth of it would have to be pure fat. African wild ungulates in late summer do not carry that.
  • Cordain et al. surveyed 229 hunter-gatherer diets and found protein supplying 19–35% of energy, tied to a physiological ceiling on protein metabolism.

Reasoning

Weighting muscle and offal by canon's own carcass breakdown gives 1,200–1,500 kcal per edible kilogram. At that density 1,300 kg a day is 1.6–1.95 M kcal, and the Phase-1 ration is 55–66% of full, below the 60–70% floor canon itself names for heavy labour. The gap cannot be closed with more lean meat: 1,300 kg at 19% protein is already about 1.0 M kcal of protein, a third or more of all energy, at the top of the hunter-gatherer range. Holding 80% on meat alone would need 1,800–2,300 kg edible a day and push protein past that ceiling. Fat and starch close the gap, which is to say hippo, marrow and the Portuguese grain, and the grain matters more than canon's table shows.

What it moves. Phase-1 ration 55–66% on meat alone, not 80%; game against plants 1.2–8× per km², not 2–11×; grain is load-bearing, not a supplement. The fat the ration needs is hippo, elephant and marrow, which is one more reason the big animals matter more than the herd species.

Sources opened

2

The game model is inland bushveld laid over a wet sand plain that already has cattle on it

What it reacts to

Layer 1, the herbivore research run reconstructed 6,393 kg of wild herbivore per km² (range 4,800–8,000), grassland at 7,797, carried by impala, blue wildebeest and zebra across seventeen species. It fitted a rainfall curve at 1,000 mm with a soil-fertility correction and checked it against the uMkhuze nagana cull. Every layer above took the number as given: both syntheses, the reconciliation's southern-wedge arithmetic, and the map's game rings.

Evidence

  • Richards Bay rainfall is 1,221–1,228 mm a year, not 1,000, on leached dune sand.
  • Olff, Ritchie and Prins: wetter ground grows more forage of lower quality, and herbivore biomass peaks at intermediate rainfall on fertile soil.
  • Fritz and Duncan: at the same rainfall, poor soils carry about one twentieth of the large-ungulate biomass of rich ones. out_05's own low-fertility curve gives about 1,570 kg/km² at this rainfall; reaching 6,393 needs about 60% of the plain to be of medium fertility.
  • Fritz and Duncan also found pastoral and natural systems carry the same total ungulate biomass. The snapshot places cattle homesteads within a few kilometres of the campus.
  • The species ranking tracks the uMkhuze cull record, about 130 km north and inland in drier bushveld. The Eastern Shores, 55–85 km up this same dune coast, hold South Africa's largest southern reedbuck population, a species out_05 omits.

Reasoning

A dune plain with one alluvial floodplain is the poor-soil case the run itself describes, so its central figure sits too high by a factor of two or more. Cattle do not add to an intact wild assemblage; they come out of the same ceiling, and their owners hunted. And the animals are the wrong ones: the coastal grassland antelope is reedbuck, with grey duiker; red duiker, bushbuck and bushpig in forest; waterbuck, buffalo and hippo in wetland; wildebeest, zebra and impala as inland or seasonal. Re-weighting out_05's two fertility curves so that a tenth to a third of the ground is alluvium or good sand gives roughly 2,500–4,000 kg/km² before cattle. That is an estimate, labelled as one, and the honest statement is that the density is unverified. Two things this does not say. It is an average over the plain; the fertile strips along the lagoon, the lake margin and the river floodplain concentrate game, and those are exactly where the colony hunts, so the near ground may be close to the research's figure after all. And it says nothing against the big animals: elephant were abundant on this coast until the 1820s ivory trade, hippo and buffalo belong to the wetland margins, and crocodile are in every water body. What was overcounted is the inland herd species as the staple hunt, not the presence of elephant or crocodile.

What it moves. Every depletion radius, including the map's game-scarce rings, grows 1.3–1.6×, and halos form in 40–60% of the time. Canon's Phase-1 becomes roughly 70–115% of the whole landscape's sustainable offtake. Canon's springbok should become reedbuck, not the impala C-19 proposes. An elephant day or a hippo in Phase 1 stays plausible; a daily wildebeest-and-zebra harvest does not.

Sources opened

3

Water comes out of the sand under the campus and off its roofs

What it reacts to

Layer 2, both syntheses modelled water as a set of fixed intake points reached by one hazardous 460 m path to the lake, with carriage, hazard and boiling fuel as the binding terms. Layer 3, the reconciliation then priced that at 76 barrow loads a day and 119–137 kg of wood to boil, and called the fuel contradiction its most consequential single finding. The aquifer under the campus was quoted by both and then left out of the model.

Evidence

  • The campus stands on the Zululand coastal aquifer, South Africa's largest primary aquifer: unconfined, shallow, and the source rural communities draw from in hand-dug wells at up to 0.4 L/s, about 35 m³ a day from one well. The drinking floor for 2,435 people at 2.5 L is 6.1 m³.
  • The snapshot itself puts the water table close to the surface toward the lagoon and wetlands. On the map's SRTM tiles the campus sits 10–15 m above the lake, so a well there is a lined dig; at its low edges it is shallow.
  • Lake Mzingazi is fed by the same aquifer (Germishuyse and Kelbe).
  • Rain is 1,228 mm a year and no month is dry: June, the driest, averages 57 mm. On the library's 3,800 m² roof alone that is about 170 m³ a month after losses, which covers the drinking floor every day of the driest month.
  • Barath calls the aquifer extremely vulnerable to pollution.

Reasoning

Sand-filtered groundwater and roof water need no boiling if kept clean, so the hippo corridor is one option among three, not the only one. The carriage and fuel figures that follow from it are conditional on choosing the lake. The hazard that matters is the one nobody listed: 2,435 people's latrines on shallow sand will contaminate nearby wells and, in time, the lake.

What it moves. Wells or driven well-points at the campus's low edges and roof catchment into storage, with the lake corridor secondary. The binding variables become storage and keeping latrines downslope of and far from the wells. The firewood-for-boiling contradiction holds only if the settlement chooses to boil lake water.

Sources opened

4

Cooking fuel is an order of magnitude above the plan, and in year one none of it is dry

What it reacts to

Canon's Day-6 cooking plan, your own file logs about 100 kg of dried wood a day, supplied as branch offcuts from the felling project. Layers 2 and 3 kept that demand and argued only that boiling water would break it.

Evidence

  • 100 kg across 2,435 people is 0.04 kg per person per day.
  • Wessels et al.: rural South African households burn 3–4 t of wood a year at about six people each, 1.4–1.8 kg per person per day, and walk up to about 1.5 km for it.
  • FAO: fresh-felled wood is about half water; air-dry wood is 12–20%. Each green kilogram gives roughly half the heat, and air drying takes months in a climate with no dry month.

Reasoning

Big communal pots use less per head than households, but not a fortieth. A third to a half of the household rate once full rations are cooked gives 1–2 t a day. Boiling the drinking floor, if it is boiled, adds 0.4–0.9 t at a realistic 15–30% stove efficiency. Offcuts felled this week are green, so the mass needed roughly doubles until a drying stock exists.

What it moves. 1–2 t a day air-dry equivalent once the ration recovers, about double that while the wood is green. With 2,400 people the collecting itself is easy: a crew of tens, a job the colony can staff from day one. What actually binds is drying. Green wood burns at half heat, there is no dry month, so a stock cut and stacked months ahead is the thing to plan, not the headcount.

Sources opened

5

South of the campus is a harbour; in 1628 it was a calm lagoon, a metre deep and more than twice the fishery's size

What it reacts to

Layer 3, the reconciliation's own land measurement read the water south of the campus as sea: 1.5 km away on the narrowest bearing, 2.6 km due south, open ocean beyond 4 km. It called crossing it a boat-building question, and the map's conflicts C-17 and C-18 say the same. Layer 1, the wild-food run sized the old estuary at about 1,300 ha with a 60–140 t/yr fishing harvest.

Evidence

  • Read ray by ray against satellite imagery, on every bearing from about 175° to 260° the first water the dataset codes as ocean is the modern harbour: the 1.5 km pinch is an inner basin, the 2.6 km due south is the dredged entrance between the breakwaters, and the open sea on the bearings just west of south lies beyond a breakwater and sand barrier at 3.6–4.1 km.
  • Jerling 2003, with its map of the original estuary: before the 1972–76 harbour works Richards Bay was a shallow estuary of about 30 km² (3,000 ha), averaging about 1 m deep, tidal range 0.1–0.4 m, with a narrow shallow mouth at its northern end. The harbour took about 60% of it.
  • Today's harbour and remaining estuary together cover about 20 km² of water in that footprint, so on this side the modern water is smaller than the 1628 water, not larger.

Reasoning

The southern wedge still ends at water within a few kilometres, so the no-southern-hinterland conclusion survives. But the distances are harbour geometry, the 1628 shore on those bearings is unmeasured, and the lower-bound caveat holds only in the north. A lagoon a metre deep with a few decimetres of tide is a punt crossing, cheap enough to be the missing fifth way out of §8 for parties on foot, though not for the SUVs. It is also the biggest fishery in reach: seines and traps work in calm shallow water where they fail in surf, and on out_06's own per-area method the estuary harvest scales by about 2.3.

What it moves. For the south, lagoon mouth at 2–3 km, shoreline unmeasured. The lagoon as a fishing ground and crossing from the first weeks, at about 3,000 ha and roughly 135–320 t a year.

Sources opened

6

There is no rocky shore east of the campus

What it reacts to

Layer 1, the wild-food run gave two kilometres of rocky shore east of the campus holding 40–150 t of shellfish, with depletion curves for 300 and 600 collectors. Layer 1, the construction run built the project's shell-lime ceiling on the same reach. The campus snapshot, your own file is where the rock came from: it read rock platforms off satellite imagery at Two Mile Beach. Both syntheses and the reconciliation carried it forward unquestioned.

Evidence

  • Satellite imagery at zoom 17 along the beach: the shore east and south-east of Meerensee is sand under surf. The hard line is an eroding dune scarp at the back of the beach, above the tide.
  • Sink's 2001 thesis surveyed rocky shores along all 560 km of the KwaZulu-Natal coast; its site map has no rocky shore between Cape St Lucia, 35–40 km north-east, and Seula Point near Zinkwazi, about 85 km south-west. The reconciler's own search puts the nearest mussel ledges at Dawson's Rock 20–25 km up the coast and the Sokhulu ledges at 30–40 km.
  • The 12–50 t from 3 km figure is Tomalin and Kyle's study of subsistence collectors at Kosi Bay, about 210 km north, where 3 km of a 30 km shore is rock.
  • The scarp is recent: the harbour cut the coast's northward sand drift in 1976 and the beaches north of it have eroded since, so the 1628 beach was, if anything, sandier.

Reasoning

A sand beach holds no mussel bed, so every number downstream of the rocky reach has nothing under it: the 40–150 t stock, the collector depletion curves, the month-one schedule, the 8–11 t yield, the quicklime ceiling, and Claude's finding that the settlement's fourth binding constraint turns on whether the canteen keeps the middens. The east shore gives sandy-beach invertebrates and surf fish, which KZN subsistence fishers do take; the shellfish worth counting are the lagoon's, and they are unmeasured.

What it moves. No mussel bed within a day's walk. Mussel-shell lime falls to near zero and estuarine shell is unmeasured. Lime is not gone, it is far: the map already places four limestone outcrops 32–62 km north, the best of them the Uloa shell-limestone in the lower Mfolozi valley at 41 km, a surface bed locally 24 m thick. None is a day's walk. Year one builds with clay mortar; from year two a road north and a kiln at the quarry supply mortar, smelting flux and field lime together. Modern Richards Bay agrees: its cement works only grinds clinker railed in from 700 km away, and the local quarries cut aggregate.

Sources opened

7

Thatch is cut in winter after seed-fall; labour binds, not the neighbours' fires

What it reacts to

Layer 2, the Claude synthesis, carried into the reconciliation: roofing grass must be cut green in March and cured through a dry season that is the Nguni burning season, so the binding constraint is a decision made by another population. Layer 1, the construction run set demand at 260–630 t at 20–35 kg/m² and roof life at 3–7 years.

Evidence

  • CSIR's thatching guide: grass is cut from mid-winter to August once growth has stopped; in frost-free country the test is that the plumes have shed their seed; cutting earlier costs quality and next season's yield. The stubble is burnt or grazed after the cut.
  • SANS 10407: grass not cut in the growing season, free of seed when cut. A standard 175–200 mm coat is 35–50 kg/m², the standard's minimum at 175 mm being 45. It loses 20–25 mm every seven to nine years and lasts about twenty before fixings show.
  • A hand cutter with a sickle makes 50–100 bundles a day; a standard bundle of fine grass weighs 0.8–0.9 kg.

Reasoning

March grass is not roofing material; the first thatch comes in June–August. Burning follows cutting rather than competing with it, and early-winter fires are kept off by a firebreak round the stands to be cut. So the neighbours' calendar is a nuisance, not the constraint. The constraint is arithmetic: at a standard coat the roofs out_11 sizes need 455–900 t, at 40–90 kg per cutter-day that is 5,000–22,000 cutter-days inside a ten-week window, tens to hundreds of cutters. The 3–7 year life describes a thin coat.

What it moves. 455–900 t for the first roofs; a ten-week cutting window; replacement nearer 30–60 t a year than Claude's 37–210, at a roof life closer to twenty years than seven.

Sources opened

8

West needs a crossing too

What it reacts to

Layer 3, the reconciliation's land measurement counts the west and north-west sectors as unobstructed, 113 and 149 km² with no water on a straight line from the campus, and on that basis calls them the only exits with land behind them.

Evidence

  • Satellite imagery and Jerling's list of the estuary's feeder streams: a watercourse runs from the lake's south-west end to the harbour, 1.4–2 km west of the campus.
  • The water dataset resolves about 30 m; the stream is narrower than that, so it is not in the mask and the unobstructed column cannot see it.

Reasoning

Every ray from about 230° to 310° crosses the lake's outlet or the stream it joins within 2 km. Unobstructed means no mapped water on a straight line, and this water is not mapped. The only direction with no water to cross is north-west up the dune ridge.

What it moves. The west and north-west figures are upper bounds on dry-shod ground, not measurements of it. The campus has water on every side but one.

Sources opened

Smaller points

Already done. The research layer is rewritten against these findings, each corrected statement marked and its source cited where it is used; the two syntheses and four research runs beneath it are not edited, and the corrections against each are listed by file at the end of the rewritten document. The map's land measurement no longer counts open ocean as land.

Still the author's. Canon's Phase-1 ration and hunting tonnage; the springbok; the campus snapshot's rocky shore, Raphia, and route wording; and which way the colony hunts, since the land is north-west and the snapshot says south.