The 1628 site

The teleported segment show on map

inferred Roughly oval, not a clean geometric shape: about 400 m on the long axis north–south and 216 m on the short axis, about 6.79 ha.

inferred Library and plaza at the north end, dormitories at the south, everything else between them along the long axis. About 2,400 people were in the segment at the moment of displacement.

inferred Library to dorms is about 400 m — five minutes walking, three at a run.

inferred The Biology Department's agricultural research fields were partially caught at the northern edge of the oval, nearest the lake and forest side.

The ground under the campus — high, dry, sand, and three water sources

computed On the map's SRTM tiles the oval sits at 18–24 m above sea level on the back of a stabilised dune, sloping about 6 m from the library end down to the dormitories. North of the library the ground falls 20 m in 400 m to the lake at 5 m; east a dune ridge rises to 31–34 m before the beach; west and south-west the ground falls to 5–8 m within 1.5 km toward the lagoon.

inferred The campus is the only high dry ground within a kilometre, above any flood. Below the arrival surface is loose dune sand with no bedrock for hundreds of metres: anything heavy settles, and settles unevenly where the ground it brought is thin.

checked Water is in the sand, on the roofs and in the lake, in that order of ease. The Zululand Coastal Aquifer is the largest primary aquifer in South Africa, shallow and unconfined, drawn by rural wells at up to 0.4 L/s; it is also extremely vulnerable to pollution, so the latrines of 2,435 people on shallow sand go downslope and well away from wells and the lake — on this terrain, south-west. Rain is 1,228 mm a year with no month under 57 mm. The lake is 20 m below the library and 525 m away: lake water comes uphill, 76 barrow loads a day at 80 L for the drinking floor alone, and needs boiling.

not established What the displacement cut took and left — which foundations, basements and mains arrived — is a story decision not yet made. The proposal under discussion is one irregular surface, 5–15 m down, that nobody will explain and the engineers will spend months mapping.

Not established. The depth and shape of the cut surface; well depth and yield at the campus; roof area and storage.

In dispute: C-16 C-21

The boundary is asymmetric

inferred Hard forest edge north and east; open grassland south. The campus does not sit within continuous forest — it occupies a boundary zone.

inferred North: riparian forest along the drainage channel, too wet for fire and naturally closed-canopy. East: dune forest on stabilised sand ridges, fire-resistant. South and south-west: flat, well-drained sandy ground, fire-maintained grassland burned annually or biennially by Nguni pastoralists and kept open by elephant browsing.

inferred On the forest-edge sides, elephant corridors and browse lines create a more open parkland structure than modern post-elephant forest, with a visible browse line at 4–5 m.

inferred The colony has two fundamentally different outdoor environments — oppressive enclosed forest north and east, open grassland south — and the contrast shapes route preference, gathering patterns and expansion direction.

inferred Without fire this entire zone would be closed-canopy forest. The grasslands are maintained, not climatic.

The layered lake shore, 1628 show on map

inferred In 1628 the lake shore is sealed behind a layered barrier: open water → papyrus beds up to 4 m tall and Phragmites reed beds at 2–3 m → swamp forest → dune forest on sandy ridges.

inferred Hippo trails are the only access through the reed barrier to open water — well-worn muddy paths carved by pods moving nightly between water and grazing.

not established The band depths drawn here are proportional, not surveyed. Canon gives the order and the vegetation heights; it gives no widths.

Lake Mzingazi — the modern impounded shore show on map

mapped SRTM Water Body Data class BH080. This is a reservoir contour: an earthen berm or weir was installed around 1942–43 to increase storage, with further raising and concrete works in the mid-century and the 1980s.

measured On a WRC bathymetric dataset the lake covers 10.57 km² at the present 3.03 m sill, 9.41 km² at 2.0 m — the elevation that report used as one possible representation of the natural outlet — and 8.26 km² at 1.0 m. About 11% and 22% less.

mapped A lower natural stage widens the papyrus, reed and swamp belt between the plaza and open water, which is what canon already describes as being there.

not established The published chronology of the weir raisings is inconsistent between three sources and the increments should not be added together and subtracted from today's level.

computed Measured on this map: the nearest modern impounded shore is 514 m on a bearing of 000° from the arrival coordinate, and 314 m on 000° from a library plaza reconstructed at the north end of the 400 m oval. Canon's 460 m falls between those two, so it is consistent with a plaza a short way north of the arrival coordinate and inconsistent with one 200 m north of it.

Not established. The 1628 lake shoreline. No mapped contour, no pre-impoundment survey on a reliable datum, no local palaeo-shoreline study at the necessary resolution.

In dispute: C-06 C-16

The drinking route — the seasonal drainage channel show on map

inferred Canon does not route drinking water to the lake shore. It routes it to the seasonal drainage channel through the riparian forest — a small stream and the natural route to the water, reliable in March after the wettest months of January and February, less reliable in winter, the driest season, though no month is dry (June–August 57–65 mm a month).

inferred The lake proper was probably a viable low-salinity source. The lower Mzingazi outlet and the Richards Bay lagoon were not reliably fresh: they are the parts most likely to be tidal and brackish, especially in dry conditions. That distinction is new — the project did not previously carry it.

not established Whether Mzingazi was literally the nearest potable source from any particular point cannot be established: it would need the exact landing position, the 1628 lake and ocean shorelines, and the positions of minor streams, dune-slack pools and groundwater seeps.

inferred "Freshwater" here addresses salinity, not pathogens.

In dispute: C-07 C-16

Coastline (modern) show on map

mapped SRTM Water Body Data, 1 arc-second (~30 m), class BA040 (ocean). The landward edge of this polygon is the coastline drawn here.

computed Ocean occupies about 46% of the circle — roughly 3,590 km² of 7,854 km². Any area budget spread over the full circle overstates every terrestrial class by close to a factor of two.

computed Ocean-excluded envelope: 4,240–4,280 km² (424,000–428,000 ha), about 54%, computed by two auditors from GSHHG, SRTM Water Body Data and ESA WorldCover, which agree within about 1%.

Not established. The 1628 ocean shoreline. No Richards Bay-specific dated shoreline reconstruction fixes the beach in 1628; transferring a late-Holocene sea-level curve from a distant estuary would create false precision. The 1628 envelope. 424,000–428,000 ha is the modern mapped value used as the working denominator.

In dispute: C-04

Lakes and inland water (modern) show on map

mapped SRTM Water Body Data class BH080. 39 water bodies fall in the three tiles.

mapped out_04 names the principal freshwater bodies as Mzingazi, Cubhu, Nsezi and Eteza, with modern surfaces of about 1,100 ha, 464 ha and 260 ha and a seasonal Eteza, and puts the 1628 freshwater-lake class at 25 km².

mapped The dataset carries no names. Only Mzingazi is named on this map, because canon independently states its position relative to the campus. Matching the other names to polygons is not attempted.

Not established. Which polygon is Cubhu, Nsezi or Eteza.

In dispute: C-06

Rivers (modern, double-line) show on map

mapped SRTM Water Body Data class BH140. Only rivers wide enough to map as two banks appear; the Mhlathuze floodplain itself is not a mapped class here.

mapped The Mhlathuze catchment is 4,209 km² and delivers water and sediment from well beyond the circle, so local floodplain extent depends on rainfall upstream of anything shown.

Not established. Mhlathuze flood recurrence, which is what decides whether levee cropping is viable at all.

Elevation at the campus (modern, 90 m resampled to 25 m) show on map

mapped SRTM3 resampled to a 25 m grid — finer than the source posts, so this is the 90 m surface interpolated onto a finer frame, not 25 m data.

inferred Canon puts the site on a stabilised dune about 15–24 m above sea level and roughly 15 m above the lake, gently sloping, the library end a few metres higher than the dormitories, on sandy coastal soils over older dune formations, free-draining and low in fertility without amendment, with the water table close to the surface toward the lagoon and wetlands.

inferred Bream Hill is level with the campus or a few metres above it, 10–20 m above the ground falling away to its west and south-west, and the first observation post; the dune ground east of the campus rises higher, to about 25–40 m.

Elevation (modern, 90 m resampled to 250 m) show on map

mapped SRTM3, 3 arc-second, void-filled, resampled to a 250 m grid in this map's plane. Voids are drawn as gaps and never interpolated.

mapped No elevation surface appears anywhere in the project's research record. The arable analysis turns on terrain position — convex interfluves, upper footslopes and old river terraces above the floodplain — and was made without one.

inferred The site stands on a stabilised dune about 15–24 m above sea level, gently sloping; the Indian Ocean Coastal Belt biome spans 20–450 m.

Not established. 1628 terrain. Parent material, texture, depth and drainage carry back four centuries; topsoil chemistry and channel positions do not.

Bream Hill show on map

derived by the map A low rise about 20–28 m above sea level: level with the campus or a few metres above it, and 10–20 m above the ground falling away west and south-west toward the lagoon. Probably grassland-topped and wind-swept.

derived by the map Offers views over the forest canopy toward the lagoon and possibly the ocean. Named for a modern street; unnamed in 1628.

oNgoye scarp forest massif show on map

inferred An elongated, roughly coast-parallel scarp-forest massif about 10 km long and several kilometres wide, on a ridge 300–500 m above sea level.

inferred Closed core about 39 km² of the circle's 65 km² scarp-forest estimate; the surviving oNgoye Forest is about 3,900 ha.

derived by the map The most readily mapped forest block in the circle, and the highest-quality sawtimber reserve — not the routine charcoal source.

mapped The vendored elevation surface peaks at 471 m at 37.8 km on 255°, inside this sector and inside the 300–500 m the source states. That is the one place a described feature and a measured dataset can be checked against each other on this map.

1 · South — open grassland (primary exit) show on map

inferred The easiest route out. No cutting required, good visibility from any slight elevation, direct access to grazing wildlife.

inferred In March the grass is 1–1.5 m tall and green — pushing through, but no machete work. Scattered trees provide landmarks; wildebeest, zebra and southern reedbuck visible at a distance (impala unverified this far south).

inferred This becomes the main hunting, expansion and patrol direction. The grassland routes cost nothing to maintain.

checked Checked against the world: open grassland 1.5–2.6 km deep on bearings 135°–225°, then water. On 135°–165° the sea beach; on 170°–225° what is today the harbour and in 1628 was the Richards Bay lagoon, about 3,000 ha and a metre deep, its narrow mouth at this end. There is no hinterland on foot to the south: the south and south-east sectors hold 3.78 km² of land in total, 2.93 km² of it with no water on a straight line, and the figure does not grow out to 30 km (Jerling 2003; land_reach.py).

2 · North — Mzingazi Lake, ~460 m show on map

inferred Through riparian forest. The seasonal drainage channel provides a partial natural corridor with slightly more open canopy along the stream bed.

inferred Hippo trails offer pre-cleared paths through the densest sections and through the papyrus and reed barrier at the shore — the only access to open water.

inferred The daily water route. Short but never safe: crocodiles, hippos, snakes and leopards all use the same corridor. Even a well-maintained 4 m track means 460 m of path with forest pressing in on both sides, and subtropical vegetation closes a cleared path within weeks.

inferred 460 m is a map measurement from the library plaza to the lake. Canon does not treat the modern water's edge as the 1628 water's edge — it describes a shore set back behind papyrus, reed and swamp belts.

checked Checked against the world: the lake is one of three water sources, not the only one. The campus stands on the Zululand Coastal Aquifer, shallow and unconfined, which rural wells draw at up to 0.4 L/s — about 35 m³ a day from one well, six times the drinking floor — and 1,228 mm of rain a year with no dry month makes the roofs a source. The lake is 20 m below the library; water from it comes uphill (Barath 2015; Weather SA 1961–90).

6 · North-east — the dune corridor, the one dry exit show on map

inferred Not among the snapshot's five routes. Measured from the arrival coordinate, north-east is the only bearing with no water inside 20 km: the dune corridor between the lake's eastern arm and the sea.

checked Checked against the world: every other way out crosses water within about 3 km — the lake north at 525 m, the lake's outlet stream west and north-west at 1.5–2 km (too narrow for the map's water dataset to show), the lagoon south and south-west, the sea east. Dune forest, so it must be cut; but it is the only route that needs neither a ford nor a raft.

inferred Which crossing canon uses, and from when, is a ruling for the author.

4 · East — to the sand beach, ~3 km show on map

inferred Following the natural dune ridgeline. Dense dune forest, but the ridge sits slightly higher, producing thinner undergrowth in places; elephant corridors may provide partial clearings.

inferred The most likely route Captain Wu's patrol used toward the coast. Ends at a sand beach under surf: no tide pools or rock shellfish within a day's walk, and surf makes beaching boats hard.

inferred Zero visibility past the tree line. Anything could be 2 m away and invisible.

checked Checked against the world: the hard line in satellite imagery at Two Mile Beach is an eroding dune scarp above the tide, and recent. A survey of all 560 km of KwaZulu-Natal rocky shore has no site between Cape St Lucia, 35–40 km north-east, and Zinkwazi, 85 km south-west; the nearest mussel ledges are Dawson's Rock at 20–25 km and Sokhulu at 30–40 km. The sea is 3.0 km due east and 1.9 km south-east (Sink 2001; Esri World Imagery).

5 · South-west — lagoon and river, 1.5–2 km show on map

inferred The most difficult route. Starts as grassland, then descends into wetland — progressively worse terrain, grassland to dense riparian forest to mangrove, ground soft, wet and insect-heavy.

inferred Useful for reaching the open lagoon shore. The lagoon is salt or brackish and not drinkable; the Mhlathuze, fresh upstream, enters it through a swampy delta on its far western side, its modern lower course 11–12 km west-south-west.

checked Checked against the world: this route meets the lagoon, not the sea. Before the 1972–76 harbour works Richards Bay was a shallow estuary of about 30 km², averaging about 1 m deep, with a tidal range of 0.1–0.4 m: a punt crossing and the largest fishery within reach, not a boat-building problem. The harbour took about 60% of it. The 1628 shore on these bearings is unmapped; the modern harbour's shore is engineered (Jerling 2003).

3 · Bream Hill, ~500 m SW show on map

inferred A low rise level with the campus or a few metres above it, 10–20 m above the ground falling away to its west and south-west; probably grassland-topped, reached via grassland and scrub from the south side of campus.

inferred The first observation post, with views over the forest canopy toward the lagoon and possibly the ocean.

inferred Named for a modern street; unnamed in 1628.

checked Checked against the map's elevation tiles: Bream Hill is 24 m above sea level, the same height as the library end of the campus.

Captain Wu's patrol had the directions wrong

inferred The Day-0 patrol report says "Lake Alpha ~3 km E (large), Lake Beta ~2 km W (small)". Mzingazi is north at about 460 m, its eastern arm reaching north-east to 1.5–3 km; the lagoon system is south and south-west at 1.5–2.6 km; to the west lies only the lake's small outlet stream. Due east is forest, then the coast 3 km out.

inferred Consistent with a rushed patrol in unfamiliar terrain, no calibrated compass, disoriented by the displacement and by forest that obscures the sun. The findings are accurate; the directions attached to them are not.

inferred This is the in-world version of what this map does at the regional scale: the bearings are the part that goes wrong first.

How much of the disc is ground — 55% at 5 km, 53% at 20 km

computed Every areal model in this corpus computes on a full disc, and none of them is keyed to where the people stand. Measured from the arrival coordinate against the water dataset this map already vendors, the disc is part sea.

computed The walk penalty is the ratio between the radius a disc model reports for a worked area and the radius that actually encloses that area. It is 1.06 at 1 km and settles near 1.37 beyond 7 km, because past about 4 km the water on the seaward bearings is open ocean and the share stops falling.

computed By 12 km the seaward half of the landscape is gone: E 7%, SE 3%, S 4% against N 92%, W 94%, NW 98% inland. Every long-range activity is forced into the same inland arc, whatever it is extracting — and that is the arc the arable answer points at.

not established "Unobstructed" counts only land with no water between it and the campus along the straight line. It is a lower bound, because a party can walk around an inlet; the land column is an upper bound, because it counts ground across the estuary. Nothing in the record says where between them the truth sits.

computed Modern shoreline. The lake is impounded and the lagoon dredged, so both non-ocean waters are larger now than in 1628: the land share is a lower bound and the walk penalty an upper one.

Not established. The 1628 shoreline. This is the modern one, and no dated reconstruction exists.

In dispute: C-17 C-18

Game-scarce radius — whole settlement, 1,328 kg/day show on map

derived by the map The radius grows as the square root of time: it doubles when cumulative hunting time or demand quadruples.

derived by the map The model treats 70% of total biomass as accessible to ordinary hunting and defines a visibly depleted zone once 25% of that has been removed, then multiplies the biological radius by 1.25 for avoidance, irregular habitat, route concentration and incomplete search.

derived by the map Impala, warthog, wildebeest and zebra on nearby open ground provide most early kills and fall fastest. Grey duiker, bushbuck, red duiker and bushpig persist closer to camp because cover reduces encounter probability. Buffalo, elephant and giraffe pass through a depleted zone but learn to avoid it. Hippo depletion is discontinuous — one accessible pod can be heavily reduced with no circular expansion at all.

not established 1,328 kg/day is not a hunting rate the story reaches until the Day-23 breakthrough. It is 0.30 kg of edible meat per person per day for 2,435 people, divided by a 55% edible fraction — a sustainable-consumption rate run backwards. Canon's Days 1–22 are sporadic: one reedbuck, one francolin, a few hares, 80 kg of crocodile, then nothing. This curve does not describe canon's first month and should not be read against it.

derived by the map The 1.25 multiplier is an operational allowance, not a biological one. It enlarges the enclosed area by 56%, so the ring drawn here must not also be read as the area from which biomass has actually been removed.

In dispute: C-17

The same curve on the land that is there show on map

derived by the map Same model, same parameters, same swept areas. The only change is that the area is placed on the land that is measurably there rather than on a disc.

computed The correction is small close in and settles beyond about 7 km, because past 4 km the water on the seaward bearings is open ocean and the land share stops falling.

not established It is still an upper bound on reachable ground. Beyond about 5 km a large part of what this counts as land lies across the estuary and the lagoon, and nothing in the record says whether the settlement could cross them.

Not established. Traverse speed by vegetation class, and the working-day length — without them no radius in this corpus is a duration, and none can be tested against a day trip.

In dispute: C-17

Game-scarce radius — one party at 150 kg/day show on map

derived by the map About nine such party-equivalents are needed to supply the settlement's assumed meat demand. Because they would overlap around the same paths, water crossings and grasslands, their combined footprint resembles the settlement-level curve rather than nine separate circles.

not established out_05 never says how many people a party is, so this cannot be converted into labour and cannot be checked against the labour projection's 10 to 25 hunters.

Not established. The size of a hunting party.

The 19 km sustainable catchment show on map

derived by the map Carefully rotated, this could theoretically supply 0.30 kg edible meat per person per day indefinitely in an average climatic period. It would expand after drought and contract in unusually productive years.

derived by the map Harvesting solely from the nearest few kilometres would not be sustainable even though the broader 50 km landscape holds ample biomass.

not established A 19 km radius is a 38 km round trip before any hunting or any carrying. Nothing in the record provides for an overnight hunting camp — no camp, no crew, no supply, no guard.

The same catchment on the land that is there show on map

derived by the map The catchment out_05 calls sustainable is a rotation over ground, so the area is the quantity that matters and the radius is only its summary. Putting the area on measured land is therefore the more meaningful of the two corrections on this layer.

computed The habitat this area is drawn from is out_05's 4,450 km² of "effective herbivore habitat", which is larger than the ocean-excluded envelope it is defined as a subset of. If the total is right the density is about 4.5% higher than stated, which moves this radius by roughly 2%.

In dispute: C-02

The gathering radius — 2 to 3 km show on map

derived by the map If all 400 gatherers remain within two or three kilometres of one camp, use the lower half of every yield range from the outset.

derived by the map First 1–3 days approximately the tabled rates; by days 4–7 commonly 0.4–0.9 kg and 260–600 kcal per active gatherer; thereafter either move camp, extend the radius substantially, accept lower returns, or lean on fishing, shellfish, hunting and stored food.

not established These are scenario ranges built on an assumed knowledge penalty, not measurements. No experiment validates the discount, and instruction is a model input rather than a discovered result.

computed Measured on this map: a 2.5 km disc about the arrival coordinate is 69% ground. The scenario behind this ring spreads 400 gatherers over the disc, so it spreads them over about 13.6 km² rather than the 19.6 km² a disc model assumes.

Not established. Any measured novice gathering rate, learning curve or depletion curve. The number of active gatherers depicted in Days 4–9 of the draft.

In dispute: C-14

Wild herbivore biomass

inferred 6,393 kg/km², plausible range 4,800–8,000 — about 57.7 animals/km², range 45–75.

inferred Across the 4,450 km² effective habitat: about 257,000 animals and 28,450 tonnes of standing live biomass.

inferred That is the same order as good modern reserves, as canon also puts it — comparable to recovering Gorongosa at about 5,938 kg/km² at similar rainfall, below historically overstocked Hluhluwe–iMfolozi, whose >8,000 kg/km² was half white rhino, fenced and managed.

checked Checked against the world, 2026-09-30: this density is unverified and on the high side of out_05's own curves. It was fitted at 1,000 mm on a soil-fertility mixture; the coast gets 1,228 mm on leached dune sand, the low-fertility case out_05 itself describes, whose curve gives about 1,570 kg/km². Cattle homesteads a few kilometres away draw on the same forage ceiling. A plain this wet and poor likely carries 2,500–4,000 kg/km² before cattle, more on the floodplain and lagoon margin where the colony hunts, and every depletion radius scales with the inverse square root of the density (Fritz & Duncan 1994; Olff et al. 2002; Weather SA).

checked Checked: the species list is uMkhuze's, a savanna reserve 130 km north and inland. It omits the southern reedbuck, the antelope of this coast's tall grass and wetland, at up to 35/km² on the Eastern Shores 55–85 km up the coast; impala here is unverified; springbok, which the Chinese drafts hunt, is extralimital in KwaZulu-Natal. Elephant, hippo, buffalo and crocodile belong to the lagoon margins and swamp forest and are not in question (du Plessis & Peel 2016; Anderson et al. 2016).

measured C-1 was checked against the literature and upheld — but out_05's own back-calculation is wrong twice. The nagana campaign ran March 1943 to February 1950, about seven years rather than the 9–11 it assumed, and it covered the uMkhuze Reserve plus adjoining Crown Lands and unoccupied farms rather than the reserve alone. The two errors push in opposite directions. Do not quote the 54–59 animals/km² derivation.

Not established. A habitat map to key the densities to.

In dispute: C-02 C-19

Density by habitat — the form needed to paint a map

inferred Floodplain and lagoon margin carry the highest biomass at 9,428 kg/km²; grassland 7,797; forest edge and open woodland 5,904; wetland 5,118; closed forest 2,922.

inferred These are the figures to apply to mapped land-cover areas. There are no mapped land-cover areas.

inferred out_05's mosaic — 35% grassland, 30% forest edge and open woodland, 15% closed forest, 10% wetland, 10% floodplain — is its own assumption and is not out_04's class breakdown.

In dispute: C-13 C-19

Sustainable offtake and the residual cattle capacity

inferred Biomass-weighted sustainable annual fraction 6.7%: 5.71 animals and 431 kg live mass per km² per year.

checked Checked: edible game is lean. Raw wild-ungulate muscle is 99–120 kcal per 100 g and offal 85–135; only fat is dense. A whole-carcass edible stream is about 1,100–1,500 kcal/kg, not the 2,000 canon's famine document and one synthesis use. Canon's Phase-1 peak of 1,300 kg edible a day is therefore 1.4–2.0 M kcal, 55–66% of rations rather than 80%, and already about a million kcal of protein — the top of what hunter-gatherer diets reach. Fat (hippo, elephant, marrow), lagoon fish and the Portuguese grain carry the rest (USDA FoodData Central; Cordain et al. 2000).

inferred At 55% edible yield and spread over 2,435 people that is about 1.19 kg edible wild-animal product per person per day — an ecological production ceiling, not a camp supply. It assumes access to the entire landscape, efficient recovery, no spoilage, no transport cost and perfect rotation.

inferred Forage capacity is 0.14–0.18 LSU/ha, but the wild herbivores already consume 0.09–0.11 LSU/ha grass-equivalent. Residual for cattle is about 0.06 LSU/ha — 16–17 ha per LSU, roughly 13,000–40,000 LSU across the whole circle, shared with an existing dense pastoralist population.

measured That residual stands on forage competition alone. out_08's argument that nagana compounds it lost its main support when C-2 was rejected: the published tsetse southern limit is the Umfolozi, which is what canon already says.

In dispute: C-02 C-20

First-year cropped area — two plans, one document

inferred Phase-1 plan: about 30 ha in Year 1 — 10–15 ha maize blocks, 3–5 sorghum and millet, 3–5 groundnut, 2–3 sweet potato, 1–2 cowpea seed block, 0.5–1 taro or rice at the lagoon margin, 2–3 intensive gardens. Year 2 about 50–60 ha.

inferred Planting waves in the same conversation: about 1 ha at Day 35, 35 ha at Day 70, 40 ha at Day 85 — about 76 ha in crop by Day 85.

inferred A factor of 2.5, both stated as the plan. At the document's own 100–150 person-days per hectare, 76 ha needs 7,600–11,400 PD against the 9,000 PD capacity the same document assumes, and demands more labour than the Year-2 plan while landing inside the famine window. 30 ha sits comfortably inside capacity.

inferred Land is not the constraint. The most expansive plan is 0.019% of the land area and under 1% of what the existing population already farms. Labour is — and it collides with the 250–450 full-time-equivalent textile workers the fibre finding needs out of the same 2,435 people.

In dispute: C-12

The good ground: 12–35 km west, 235°–310° show on map

inferred On convex interfluves, upper footslopes and old river terraces above the floodplain — from the Felixton–Empangeni corridor through the rolling country west and north-west of eNseleni.

inferred The four runs place it at: 12–30 km W/WNW/WSW; 15–30 km W/WSW; 12–25 km W–WSW; and 255°–310° at 16–35 km. The arc drawn here is the union.

derived by the map Only runs C and D derived their own quantities. Runs A and B were given the project's figures in their prompts and echoed them, so agreement among four is not corroboration by four.

inferred Field selection wants broad convex interfluves and upper or lower midslopes below 5–8%, deep red, yellow-brown or structured loams, and old Mhlathuze terraces standing clearly above active channels. Expect to survey roughly twice the area you intend to open.

not established A ranking conflict survives inside the agreement: one run puts the higher Mhlathuze alluvium and low colluvial benches first and the rolling upland second; the other three put the upland first or fuse the two. Which surface wins is not established at circle scale.

not established Fields 12–35 km inland are a day's walk each way from the campus, which implies satellite settlement rather than one nucleated site. Nobody has costed it, and the audit records that the logistics conclusion is plausible rather than established.

not established "West of the site" is ambiguous between the coordinate the research was commissioned against and the arrival coordinate, which are 4.61 km apart. This band is drawn from the commissioning coordinate. Measured from the campus instead, the same band's near edge is about 16 km away on a due-west bearing rather than 12, so the day's walk is a third longer than the figure suggests. The source does not say which point it means.

Not established. The polygon. Closing it needs the ARC Land Type memoir for sheet 2830 Richards Bay (~823 pp, 1986/88), a licensed document — and one auditor holds that even the memoir would not be sufficient, because association mapping cannot locate individual soil forms. Soil-form hectares anywhere in the circle. Untreated 1628 soil chemistry. Parent material, texture, depth and drainage carry back four centuries; topsoil pH, phosphorus and organic carbon do not. 1628 cultivated extent. Farming communities had occupied this coast for centuries; their field positions cannot be reconstructed at this scale. Whether labour binds before land. None of the four runs contains a labour-capacity model.

In dispute: C-11

The 71.0 ha augered parcel — the only measured ground in the circle show on map

measured Desktop land-type mapping predicted an average Class III. Close-spaced augering found 37.5 ha — 52.8% — of wet Class V Katspruit/Westleigh ground, and the assessment concluded ordinary agriculture was not feasible there.

measured Minor elevation differences separate cultivable sand ridges from non-arable wetland over a few hundred metres, and the difference is invisible without an auger.

computed A correction inside that source: its detailed table gives Namib/Class III as 27.6 ha, and only that value closes the parcel — 27.6 + 3.3 + 37.5 + 2.6 = 71.0 ha. Its executive summary says 19.2 ha, which leaves 8.4 ha unaccounted for. Use 27.6; 19.2 is a transcription error.

not established This makes the coastal plain a poor and risky first choice. It does not establish that every coastal parcel is worse than every inland one — all four runs retain some elevated Namib/Fernwood rises as secondary opportunity.

The safe crop package

inferred Sorghum, millets, cowpea and other African pulses, and cucurbits are common to all four runs as the agronomic core.

inferred Maize should not be treated as securely present here in the early 1600s, and the caution extends to sweet potato, cassava and peanut. Two of four runs argue this explicitly; one does not conduct the analysis, and one recommends groundnut and sweet potato without the caveat. The historical-safety argument rests on two runs, not on a consensus.

inferred Under low input, sorghum or millet intercropped with cowpea or Bambara groundnut, rotating with grazed or wooded fallow, outperforms continuous cereal — the best terrace and footslope ground degrades a full suitability class within a few seasons of continuous unreturned cropping.

Phosphorus — the constraint that does not go away

inferred Legumes and manure supply nitrogen. Nothing available in 1628 adds phosphorus — ash, bone, fish waste and settlement refuse redistribute a finite local stock.

inferred Sustainable cropped area is therefore tied to livestock numbers and returnable biomass rather than to arable hectares, and cultivation tends to contract around homesteads and cattle enclosures where manure concentrates.

inferred All four runs treat phosphorus as a serious long-term constraint; they do not unanimously rank it first.

Rainfall — a gradient, not a number

measured SAWS station 0305167 W, 3.74 km from the circle centre, 1922–1975: 1,221.1 mm/yr, SD 373.4, CV 0.31. A separate design-rainfall listing for the same station gives 1,248–1,255 mm/yr.

computed WorldClim 2.1 at a cell 1.30 km away: 1,140 mm/yr. CHIRPS bracketing cells: 961–1,039 mm/yr.

mapped Regional interpolation across the Richards Bay coastal aquifer shows a steep east–west gradient: above ~1,000 mm near the coast, falling sharply inland, possibly below 800 mm within ~20 km. The cropping belt sits 12–35 km inland, on the dry side of that fall.

not established No mean for the circle or for the cropping belt has been established, and a coastal gauge is the wrong instrument for an inland belt.

not established Numerical 1628 rainfall cannot be recovered. The modern gradient is usable as a scenario, with high interannual variability.

Not established. A belt mean for the 12–35 km cropping arc. Numerical 1628 rainfall. The point mean annual evaporation — the 1,902 mm figure is a broad regional summary, and a local assessment instead gives ~1,675 mm reference PET and ~1,300 mm mean annual evaporation, possibly under different definitions.

In dispute: C-11

Lime cannot be budgeted from pH

measured Inland upland soils near Empangeni measure pH 6.21–6.50 in water and 5.04–5.61 in KCl — the same soils, two scales, differing by 0.89–1.21 units. Much of the runs' apparent disagreement is a reporting-medium artifact.

measured But that experiment cannot establish untreated 1628 soil reaction: its arable plots had been cultivated for about 38 years with irrigation, fertiliser and about 2 t lime/ha/yr, and even its "undisturbed" grassland had prior cultivation, grazing and burning.

measured A separate cultivated field measured pH(KCl) 4.8 with only 1% acid saturation — a low KCl number with essentially no lime requirement.

not established A blanket 2–6 t/ha is unsupported, and so is a blanket zero. The answer is selective, test-driven liming.

Not established. Untreated belt pH in 1628. Whether the shell-lime supply of 1.0–1.6 t/yr suffices — undeterminable until treated hectares and measured requirements are known.

In dispute: C-10

Shell lime from the local shore — a ceiling built on a shore that is sand

inferred At a conservative renewable yield of 5.5 t whole mussels per km of shore per year, the local 1.5–2 km reach gives 8–11 t whole mussels/yr and only 1.0–1.6 t usable quicklime/yr.

inferred One tonne of quicklime at 1:3 makes 3–4 m³ of mortar, so a sustainable shell-lime season supports 4–6 m³ of mortar — not a masonry town. Shell lime is a special-purpose material for hospitals, kitchens, latrines, water installations and whitewash.

checked Checked against the world, 2026-09-30: the 1.5–2 km reach is a sand beach under surf, and the nearest mussel ledges are 20–25 km up the coast. The 8–11 t and the 1.0–1.6 t of quicklime have no shore under them; the 12–50 t from 3 km they were scaled from is the Kosi Bay fishery, 210 km north. What is near is lagoon and beach shell, unmeasured. The limestone the map places north — Umkwelane at 32 km, the Uloa coquina at 41 km, Somkhele at 52 km, Mfelezi at 62 km — is the lime source from year two, by a road and a kiln at the quarry; year one builds with clay mortar (Sink 2001; Tomalin & Kyle 1998; Botha 2018).

not established Whether the Uloa coquina is a genuinely different feedstock or the same carbonate in another physical form depends on a project definition of "shell" that has not been examined. Uloa coquina is itself lithified shell.

In dispute: C-10

Uloa Bridge — shell limestone, coquina, calcarenite show on map

measured The most promising carbonate system inside 60 km, and heterogeneous. Shell-rich "Pecten Bed" about 4–5 m; measured composite section for the whole formation about 6.8 m.

measured Fossiliferous material was excavated at Sapolwana for use beneath a cane railway, and the classic section was cut by road and rail works — coherent excavatable bodies exist.

not established No reserve tonnage, no deposit-scale chemical grade. Much of the formation is deeply weathered and decalcified. Bed thicknesses are not reserves.

measured One petrographic point count at Warner's Drain reported about 45% micrite. That is one specimen under a microscope and the run says explicitly it must not be used to calculate lime yield.

computed The kiln belongs at the quarry: burning drives off CO₂, and hauling raw rock 32–52 km would waste animal capacity. 100 units of pure CaCO₃ give about 56 units of CaO.

In dispute: C-10

Umkwelane Hill — Cretaceous impure limestone show on map

mapped Approximate point 28.49°S, 32.145°E. The auditor recomputed 32.04 km at 016.9° true on a WGS-84 inverse geodesic and confirms the run's figure.

measured Bluish-grey, abundantly fossiliferous, overlain by earthy calcareous sandstone — described only as "impure limestone".

not established No reserve tonnage. No CaCO₃, CaO, MgCO₃, silica or clay assay is published.

inferred A genuine possible kiln feed, but only by selectively taking the most limestone-rich beds. It cannot presently be classified as high-calcium, dolomitic or naturally hydraulic lime feed.

In dispute: C-10

Somkhele, Enseleni Creek — Lower Karoo impure limestone show on map

mapped Reconstructed point about 28.296°S, 32.066°E; the auditor recomputed 52.19 km at 001.7° true.

not established The survey calls it impure limestone and states that the percentage of impurities is large. No reserve tonnage, thickness or lateral extent.

inferred A fallback experimental kiln feed only. Too distant and too poorly characterised to choose ahead of Uloa or Umkwelane.

In dispute: C-10

Mfelezi Station limestone — the nearest quantified limestone show on map

measured About 5 Mt at 62% CaCO₃ / 37% SiO₂ / 1% MgCO₃ — poor lime feed with a large inert fraction, and explicitly not natural-cement rock without test burns.

mapped The nearest quantified real limestone to the site, and outside the 50 km circle.

In dispute: C-10

Arenaceous limestone under the lagoon — real, and not hand-workable

measured Bores immediately north of the lagoon passed through about 27 m of sand before reaching fossiliferous arenaceous limestone. No exact bore coordinate, limestone thickness, reserve or assay is supplied.

measured Twenty-seven metres of unconsolidated saturated sand overburden, groundwater control and unsupported excavation rule it out as a hand-worked lime quarry.

Cape St Lucia — Isipingo aeolianite and beachrock show on map

measured A historical account describes fractured calcareous sandstone extending for roughly a mile below the lighthouse, but gives no thickness or volume.

inferred Poor lime feed — carbonate-cemented quartz sand or conglomerate, not clean limestone. Coastal position also makes animal haulage awkward.

St Lucia estuary mouth — Isipingo Formation show on map

not established Aeolianite, calcarenite, calcified beach conglomerate and beachrock, intermittently exposed. No reserve, no assay.

The forest estate

inferred 7.46 million m³ merchantable standing over 56,500 ha; 4.45 million m³ operationally accessible. Area-weighted mean 132 m³/ha.

inferred "Merchantable" means sound bole and major straight-stem volume about 20 cm diameter and larger. It does not mean sawlog quality: much of the dune, swamp and elephant-worked stock would serve only for poles, split work or fuel.

inferred The accessibility factors are not biological measurements. They represent what a pre-road settlement could repeatedly reach without spending more labour and animal power on transport than on felling and conversion.

inferred The absolute corner range is 2.9–16.3 million m³, too wide to be useful; a credible correlated range is about 4–13 million m³.

Not established. Stand locations. The forest estate is an area budget; nothing in the record places a stand.

In dispute: C-03

The constraint is increment, not inventory

inferred Sustainable annual cut about 49,170 m³/yr from 34,225 ha of accessible forest, leaving roughly 30% of estimated accessible increment uncut. Defensible planning range 30,000–90,000 m³/yr.

inferred Against a scenario charcoal demand starting at 1,500 t/yr and growing 5% annually, demand crosses the sustainable cut within decades. Kiln efficiency alone cannot reconcile a growing iron industry with indigenous-forest growth: dedicated plantations must begin in 1628.

inferred Sixth of the seven binding constraints — and it rests on the 56,500 ha figure, which out_04's classes do not support.

inferred Mimusops and Sideroxylon coppice poorly. Never rely on coppice to replace old-growth boat and construction timber. Scarp forest is the highest-quality sawtimber reserve and should not be the routine charcoal source.

In dispute: C-03

Ninians Quarry — gneissic granite show on map

inferred The best durable masonry source: squared and coursed rubble, plinths, foundations, retaining walls, steps and coarse ashlar. Joint-bounded blocks can be won with plugs, wedges and hammers.

measured Historically recognised as admirable building stone even roughly dressed. Its cost is labour, not durability. Fine dressing is extremely laborious.

not established No public reserve tonnage, block yield or compressive-strength data located.

not established This replaces the distant dolerite as the nearest known good hard rock — geologically only. No route survey, haul capacity, quarry labour, tool wear, political access or seasonal passability appears anywhere in the evidence, so no conclusion about first- or second-year masonry affordability follows.

In dispute: C-08 C-09

Richards Bay Quarry — dolerite show on map

inferred Excellent for strong rubble masonry, footings, revetment and retaining work. Poor for regular dressed ashlar: fresh dolerite is hard, compact and slow to shape with hand tools.

not established The best documented road-metal source in the radius. No public reserve tonnage or quarry-specific laboratory results found.

In dispute: C-09

Reserve No. 9 dolerite — the former nearest quantified hard rock show on map

measured About 17.9 M m³ under only 0.3–1.0 m of overburden. Excellent for rubble, aggregate and road metal; poor for dressed ashlar.

computed This was the compilation's nearest quantified hard-rock resource. Ninians at about 18 km reduces that straight-line distance by roughly 49–64%.

In dispute: C-09

Diabase west of Umkwelane Hill show on map

inferred A plausible local hard-stone source for a lime-working settlement: rubble, kiln foundations, broken road stone. An untested prospect, not the equivalent of the proved Richards Bay quarry.

Mtuba Crushers — basaltic andesite show on map

measured In a CSIR comparative investigation, sample D7 showed significant fracturing and disintegration during ethylene-glycol treatment and ranked 12th of 12. Assessed as unsuitable for high-standard road base.

inferred One sample is not a whole quarry — but it warns against assuming the nearest hard rock is the right hard rock.

Natal Group sandstone — Eshowe-side western fringe show on map

measured Commonly thin-bedded, false-bedded and liable to split after weathering; compact thick beds occur locally but are uncommon.

not established Only a prospect until a particular bed survives wetting and exposure tests.

Zircon in the dune sand — the one promising local refractory

inferred 0.8% of the dune sand, about 128 t per hectare-metre in situ — the only route to high-performance glass and metal-contact refractories, but requiring electric fusion.

inferred Dune sand also carries ilmenite, which alters refractoriness and mould–metal reactions in foundry use. Ilmenite itself remains unsmeltable pre-industrially.

Hlabisa — kaolinitic material show on map

not established Patches in weathered feldspathic rock, with no published reserve, analysis or demonstrated performance.

inferred No proved refractory-grade fireclay, magnesite, dolomite, graphite or chromite is documented within 100 km. Refractories are the second binding constraint and gate everything past the bloomery.

Bull's Run–Eshowe — kaolinitic material show on map

not established Patches in weathered feldspathic rock; no published reserve, analysis or demonstrated performance.

Karkloof bauxite — refractory or ceramic feed only show on map

inferred 39–40% Al₂O₃, yielding refractory or ceramic feed and never metal, because primary aluminium needs Hall–Héroult electrolysis.

inferred Aluminium changes the most of the minor metals — but only as salvage. The campus stock is a finite inherited capital stock and should be allocated by alloy and function, not melted indiscriminately.

Port Durnford cliffs — bog iron show on map

inferred Bog iron at the Port Durnford cliffs, ferricrete on the coastal terrace, dune magnetite as a blend. Nothing in the 2026-08 wave disturbs the existing metallurgy canon on iron.

measured Port Durnford is not a limestone source. The formation there is chiefly carbonaceous mudrock, lignite/peat and sand; the old ~50 ft calcareous sandstone description referred to a section farther north in the St Lucia district.

Umfuli–Mthonjaneni pegmatite — tin show on map

measured The find of the metals wave, and the only one of the eight structural metals that is unambiguously extract. Cassiterite reportedly assaying 6.65% Sn, with a modern minimum "proven" reserve of 36,000 t — about 2,394 t contained tin.

inferred Workable by selective mining, hand cobbing, stamps, buddles and charcoal reduction of SnO₂: genuinely seventeenth-century technology. A 20-worker pilot at 100 t ore/yr yields about 4 t tin. Output 4–45 t Sn/yr.

Sebenzani — copper and nickel show on map

inferred 0.6% Cu low-grade sulfide. Output 0.2–5 t/yr. Salvage and trade now, extract later.

inferred Copper, not iron, is the binding metal. The settlement can make structural iron, tools and machine frames without copper; what it cannot replace is copper's combination of conductivity, corrosion resistance, ductility and solderability.

inferred The first copper "mine" is the campus itself — windings, transformers, cable, refrigeration tubing, brass fittings.

Sithilo Complex — chromium show on map

inferred 9–50% Cr₂O₃ pods; 50–500 t concentrate, zero alloy chromium. Refractory use only — chromite is not improvisable.

Mambulu Complex — vanadium show on map

inferred 22 Mt at 0.56% V₂O₅; effectively zero alloy vanadium. Substitute; do not prioritise.

Mvoti scheelite workings — tungsten show on map

measured 1.76% WO₃ locally, worked in the Second World War. Produces only a concentrate until hydrogen reduction and powder metallurgy exist.

inferred Tungsten matters first as salvaged tooling: carbide inserts, end mills, HSS cutters and TIG electrodes on the campus are strategic consumables.

Mfongosi vein — lead and silver show on map

inferred A 1.7 m argentiferous galena vein; 2–70 t/yr. Prospect, then extract selectively.

inferred Lead chambers need 300–800 kg of lead for a 10–20 m³ chamber, which the settlement does not have. Without it, glass and stoneware cap acid production around 10–30 t/yr.

St James–Dania–Nondweni — zinc show on map

inferred Sphalerite to 15% Zn over 5 m. Zero until retorts exist; then 5–60 t/yr.

Dwaalhoek wad and Dipka pyrolusite — manganese show on map

inferred 50–500 t ore, metal negligible. Small oxide quantities only.

KwaMachi — nearest securely documented exposed pyrite show on map

not established Sulfur is the gate on everything past lime, potash and fermentation, and it is not local. No grade or reserve is reported here.

inferred The Tugela Terrane 65–150 km W is the first serious exploration target, honestly priced: 50–70% chance of finding visibly pyritic material, under 15% chance of proving a deposit supporting more than 100 t of acid per year.

Barberton–Malalane stibnite — antimony show on map

inferred The only defensible purpose-mined specialty metal, conditional on finding a hand-sortable vein: 1–5 t/yr. Its urgency is reduced by the good tin prospect.

The settlement core, 0–5 km show on map

inferred A flat, sandy coastal plain within a deliberately open or semi-open part of the mosaic. North is Mzingazi with open water, reed and sedge margin, swamp forest, moist grassland and higher dune or lowland forest. West and southwest the ground descends toward channels and wet flats connected to the Mhlathuze lagoon. Eastward: lowland grassland or thicket, stabilised dune forest, younger dune vegetation, beach, ocean.

inferred The site itself was more likely burned grassland or an enlarged anthropogenic clearing with nearby woody islands than closed mature forest, given the dense pastoral population and the proximity to water, grazing and travel routes.

inferred The dune-forest boundary was not a straight north–south line: forest tongues followed older stabilised dune ridges inland, grassland and wetland tongues entered through dune slacks, and paths, cattle routes and settlement clearings broke it.

The northern and north-eastern coastal plain, 5–50 km show on map

inferred A broad sandy plain of fire-maintained grassland interleaved with lowland forest, thicket, palm-dotted wooded grassland, swamp forest and innumerable wet depressions. The coastal dune-forest ribbon continues parallel to the shore.

inferred Forest increased on older sheltered dunes and in wet firebreaks; grassland increased on seasonally wet sands, around settlement clusters and where repeatedly burned.

inferred Lake Eteza lies roughly 30–35 km north and forms a node of seasonal open water, swamp vegetation and coastal grassland. It supplies the nearest long Holocene pollen record relevant to the reconstruction.

The north-western and western interior, 10–40 km show on map

inferred The plain becomes more dissected and soils locally improve relative to the extremely leached coastal sands. Settlement and cultivation are consequently denser on well-drained interfluves and valley margins — a finer-grained mosaic of cultivated clearings, fallow grassland, grazed woodland and riparian forest.

inferred Lake Nsezi and the Nseleni–Mposa river system form a conspicuous wetland corridor, with reed beds and flood-prone ground extending beyond the open-water lake.

inferred The western edge of the circle approaches the transition from flat coastal-belt vegetation to more strongly rolling terrain.

The west-south-western scarp sector, 25–45 km show on map

inferred The most readily mapped forest block: oNgoye. An elongated, roughly coast-parallel scarp-forest massif around 10 km long and several kilometres wide, centred on a ridge 300–500 m above sea level, closed core about 39 km².

inferred Fire-prone grassland surrounds much of the ridge; recurrent edge burning prevented forest descending indefinitely onto the gentler slopes around it.

The Mhlathuze sector, 5–40 km show on map

inferred The largest continuous floodplain complex in the circle. The river approaches from the west and spreads across a low-gradient alluvial plain before entering the back-barrier lagoon: a broad belt of seasonally inundated grassland threaded by active and abandoned channels.

inferred Reed bed and swamp forest occupied the wettest backswamps; riparian forest followed elevated channel margins; grazed and burned floodplain grassland occupied surfaces that dried in winter.

inferred Toward the mouth, salinity and tides replace river flooding as the control. The natural sequence is river channel and freshwater marsh, brackish reed and sedge, shallow lagoon and mudflat, a relatively small mangrove fringe, sand barrier, dune vegetation, beach. It should not resemble the modern deepwater harbour.

The southern coastal sector, 25–50 km show on map

inferred The dune-forest ribbon remains conspicuous; inland the cover shifts toward the rolling KwaZulu-Natal Coastal Belt with more fertile substrates, denser cultivation and a tighter grassland–thicket–forest mosaic.

inferred The Mlalazi estuary at the circle's southern margin contributed lagoon water, reeds, floodplain and dune forest — and essentially no mangrove in 1628. Its modern mangrove stand colonised only after twentieth-century manipulation of the mouth.

The eastern sector show on map

inferred Beyond the beach, virtually the whole eastern semicircle is open ocean. The terrestrial coast occupies only a thin chord near the western edge of that semicircle. This geometric fact dominates the full-circle percentages.

Why the modern landscape is not a proxy

inferred 650–950 km² of the terrestrial portion is now eucalyptus and pine plantation; 350–600 km² sugarcane; 180–280 km² urban and industrial; 20–35 km² is post-1970 harbour geometry.

inferred Mangrove today is about 760 ha against roughly 80 ha in 1975 — modern mangrove is largely an artifact of harbour engineering, and the Mlalazi mangroves did not exist before twentieth-century mouth manipulation.

inferred When reassigning plantation polygons, 55–70% reverts to grassland or wooded grassland, not to forest. Deleting plantation and calling it forest would overestimate natural forest; calling all of it grassland would also be wrong.

inferred This is why the map carries no satellite basemap: the modern surface is a documented distortion of the 1628 one, and showing it would import all of that silently.

The unallocated remainder

computed The thirteen classes were apportioned against a non-ocean envelope of 4,090 km². The corrected ocean-excluded envelope is 4,240–4,280 km².

computed Roughly 150–190 km² is therefore unallocated to any class.

computed The compilation declined to redistribute it, and said where it most plausibly sits: in estuarine and lagoon water and the low-lying wet classes, which is precisely where different shoreline definitions disagree.

In dispute: C-01

Tsetse — the southern limit, 30–40 km north show on map

measured De Beer et al. (2016), the current review of tsetse distribution in north-eastern KwaZulu-Natal, puts the southernmost limit of tsetse distribution in South Africa at the southernmost extent of the Umfolozi River. Modern surveys pushed it slightly further — G. brevipalpis at Ocilwane, 5 km south of the Umfolozi — not to the Mhlathuze.

measured out_08 argued the historical boundary was at the Mhlathuze, 1.5–2 km from the site, making animal trypanosomiasis a live constraint on cattle. That challenge is rejected. Underneath it is a real 1926 government report on nagana at the Mhlathuze settlements — but it is cited in the horsefly literature: nagana occurred there by mechanical transmission, without a resident tsetse population. "The disease was recorded here" had been read as "the vector lived here."

measured Three qualifications keep this from being a clean win. Mechanical transmission is a real pathway needing only an infected reservoir within reach of biting flies, though no experimental evidence exists for it here. The boundary is soft and under-surveyed — the Ocilwane record exists because somebody finally surveyed an area not previously surveyed. And the modern distribution is post-eradication: G. pallidipes was wiped out of Zululand by a 1945–52 DDT campaign, and before it Harris collected 2 million flies in a month from 1,000 traps in 1932.

measured The pre-1950s fly population north of the line was therefore far denser than any modern figure implies. That raises what 1628 looked like north of the line; it does not move the line.

In dispute: C-15

Malaria exposure and its timing

inferred Expected settlement entomological inoculation rate 5–25 infectious bites per person per year, central 10–15. Seventh of the seven binding constraints, in a wholly non-immune population.

measured The site was malarial in 1628, and its modern non-malarial status is an artifact of control, not ecology. The risk boundary was pulled north from Richards Bay to about Mtubatuba–Mfolozi in 2013; DDT reintroduction after 1999–2000 produced a 91% incidence decline and apparently eliminated An. funestus from KwaZulu-Natal, while surveys still find large vector breeding populations locally.

measured In the pre-chemical era the 1931 season saw malaria as far south as Port St Johns, about 400 km south, and the 1932 epidemic killed about 22,132 of 985,000 at risk.

inferred Late March is not a safe end-of-season arrival: first P. falciparum cases could appear in the first or second week of April 1628, with a conspicuous cluster by mid-to-late April, and the first full transmission season following in November 1628 – March 1629.

Not established. Vector breeding-site locations. The habitats are water types, not mapped polygons.

An. arabiensis — warm shallow temporary freshwater

inferred Breeds in rain pools, hoofprints and ruts. Feeds indoors and out.

inferred Keyed to a water type, not to a mapped extent. Every hoofprint in a game landscape is habitat, which is why no zone is drawn.

An. funestus — permanent vegetated water

inferred More strongly human-feeding and indoor-biting; potentially the most dangerous sustained vector.

inferred Its habitat is the lake margin and the reed and swamp belt the campus route north crosses — permanent, vegetated, and exactly where the water is.

An. merus — brackish and estuarine pools

inferred Outdoor-biting, breeding in brackish and estuarine pools — the lagoon margin and the Mzingazi outlet.

not established The compilation's argument that "an estuarine 1628 Mzingazi is better merus habitat than today's freshwater lake" rests on a reconstruction the 2026-08-16 wave rejected: the lake was probably already freshwater in 1628, and the twentieth-century works made a reservoir out of a lake rather than a lake out of an estuary. What survives is narrower — the outlet and the lagoon were tidal and variably brackish, which is merus habitat without needing the lake to be.

In dispute: C-07

Schistosomiasis — focal, in quiet warm vegetated freshwater

inferred Risk sits in quiet warm vegetated freshwater margins, backwaters and tributaries — not the wave-exposed shore and not the fully brackish lagoon.

inferred That is the same water the drinking route crosses.