Gweru east two-block claim

A first read of two peg sheets supplied on 22 September 2026: a nine-peg block (A–I, 165.7 ha) and a seven-peg block (A–G, 149.8 ha) that share one edge, 315.5 ha together, about 12 km east-north-east of Gweru. This is new ground, 54 km from the nearest earlier site. The page answers one practical question: which ground deserves the first reconnaissance traverse? It is not evidence that gold is present.

What was assumed about the coordinates. The second sheet is headed 35K, the UTM zone-and-band label, so zone 35 south is taken from the sheet rather than guessed; the datum is not written and WGS84 is assumed, as on every earlier site. Both peg lists trace simple rings in the order supplied, so the footprints here are the supplied order exactly, not an approximation. The second sheet's A and G are the first sheet's E and F, which makes the two blocks adjacent along that edge. Two digits were the hardest to read on the photographs: the first sheet's C northing (7852678) and G easting (806293). Confirm those two with the holder.

Bottom line: the fused model reads both blocks above their own background, and the seven-peg block is the stronger of the two. At 10 m the seven-peg block's median sits at the 82nd percentile of the surrounding ground (median 0.377, one cell in ten above 0.5); the nine-peg block at the 73rd (median 0.336). Together that is the second-strongest whole-block ground in this study after Belingwe-West, in the same band as the Silobela ground that later assayed. The local terrain model agrees (83rd percentile, on a terrain signature that separates the 19 nearby producers from background at held-out AUC 0.89). The terrain-and-structure regional model does not (32nd), and the iron-oxide and ASTER arms are flat. This is a surface-resemblance result, not a discovery or a grade prediction.

The field decision

Start with three broad search areas, not point targets:

Seven-peg block, north-east (between pegs C and D): around 807900 E / 7853450 N, a 450 m neighbourhood. The fused model's top 10 m spot (0.69) and its sixth, the 80 m model's fifth, the regional model's first, and the two strongest lineament-density maxima all fall here. It is the tightest multi-layer convergence on the claim.
Nine-peg block, south-west corner (pegs A and I): around 805350 E / 7851280 N, a 350 m neighbourhood. The fused model's second spot at both resolutions (0.68), the terrain model's second, and the first and third ferric maxima on the claim sit here together. This is the only place where the iron-oxide layer and the model agree.
Seven-peg block, south-centre (inside the A–B–F–G corner): around 807400 E / 7852680 N, a 400 m neighbourhood. The fused model's third and fourth spots (0.67, 0.64) and the 80 m model's first and sixth, with the regional model's second about 480 m to the west.

A fourth, lower-priority traverse runs through the middle of the nine-peg block, 805950–806450 E / 7851600–7852350 N: the fused model's fifth spot, the terrain model's first and third, a ferric maximum and a lineament maximum all lie in that band without any two of them coinciding.

The numbered green areas on the map are those reconnaissance starts. Purple points are the fused model's ranked 10 m spots. Toggle the other evidence families to see whether a target persists or depends on one sensor, and tap any cell of the heatmap to see the numbers behind it.

Ranked fused-model spots

The six local maxima of the fused model at its native 10 m read, across both blocks. Four of the six are in the seven-peg block. The probability-like value is a relative model score, not the probability that a deposit exists, and the spread between first and sixth (0.69 to 0.64) is small: treat them as one tier, not a ranking to walk in order.

What each layer says

Evidence	Claim result	Field meaning
Fused model, 10 m (both blocks)	78th percentile	Median 0.353, top decile 0.47, maximum 0.69. Seven-peg block 82nd (0.377), nine-peg block 73rd (0.336). AlphaEarth-dominated surface resemblance to the ground at known producers
Fused model, 80 m	82nd percentile	Same model at the embedding's native read; median 0.367
Local terrain model	83rd percentile	Fitted on the 19 producers within 25 km of this claim; held-out AUC 0.89, spatial AUC 0.72 at an 8 km buffer. Mines here sit about 65 m higher than background (1422 vs 1357 m) on slightly steeper ground
Regional terrain + structure	32nd percentile	Median 0.241. The all-belt terrain model does not see this ground as mine-like; its signature (roughness, relief, slope) is not the local one
Lineament density	37th percentile	Not elevated on the claim as a whole; the two strongest maxima are in the seven-peg block's north-east. Dominant trend 045 to 060 degrees. Roads and field edges remain false positives
ASTER Al-OH	54th percentile	Flat. 2006 archive scene, fully exposed
ASTER Mg-OH/chlorite	52nd percentile	Flat
ASTER carbonate/Mg-OH	42nd percentile	Flat
Sentinel-2 ferric ratio	39th percentile	Not broadly iron-stained. Bare-ground coverage is excellent (97.5% of claim pixels usable in the 2025-07-24 scene), but the 19 nearby producers sit at background ferric (separability -0.04 sigma), so a ferric high does not mark gold locally

The two model reads disagree, and that is information. The fused model and the local terrain model both place the claim above the 75th percentile; the regional terrain-and-structure model places it below the median. The fused model takes 88.8% of its importance from AlphaEarth embeddings, so its read is about what the surface looks like, not about landform; the local terrain model is fitted on only 19 producers and its skill falls from 0.89 to 0.72 once nearby mines are held out spatially. Read the claim as "surface resembles producing ground, landform is locally but not regionally typical". Nothing on this page is a grade, and the 3.5 km distance to the nearest catalogued producer (108 kg recorded) is context, not evidence about this ground.

Read the surfaces

The bright ground in the seven-peg block is not one feature: it is a north-east band from the shared edge up to the C-D corner, and it stops abruptly at the block's eastern tip, where the surface east of peg E reads among the lowest in the window. The nine-peg block's north (pegs B, C, D) is likewise low. Those two dark patches are worth a look precisely because the model dislikes them: low-ranked control samples are what turn a resemblance map into a calibrated one.

What to record on the ground

At all three areas, photograph and geotag bedrock, float, quartz veining, lateritic cover, cultivated boundaries, drainage, and any workings. Record strike and dip where possible, and take control samples from low-ranked ground (the eastern tip of the seven-peg block, the north of the nine-peg block) as well as the bright zones. Those observations, not another satellite index, are the next evidence needed to calibrate this claim. If the holder can export the pegs from a GPS app as GPX or KML rather than a written sheet, the two uncertain digits resolve at the same time.

The complete provenance, reproduction commands, limitations, and all generated outputs are tracked under research/mineral-prospecting/sites/gweru-east/.