Frontend demo only — unlocks richer reasoning and removes coordinate obfuscation on synthetic data.
test / test./operator/* route. This frontend form is a demo of the UX, not real auth.
Closed-loop AI targeting for Somaliland gold operators. Every pit you dig sharpens the next recommendation within 24 hours. Offline-first tablets. Append-only audit trail. OECD Annex II-aware offtake routing.
Status: software built and operational. We're looking for a design-partner operator with a real Somaliland MEM licence to validate against. The demo location is illustrative — your actual operating area is what we'd target, defined together after discussion.
Most targeting tools deliver a static map and walk away. This one closes the loop — every assay updates the model overnight, and the next morning's recommendations reflect what you just learned in the field.
The model ranks your AOI by expected gold-per-dollar — geology, drive distance from camp, dig cost, and offtake readiness all factor in.
Foreman pulls up the next pit on the offline tablet. GPS, depth target, expected grade, and cost all pre-populated.
Tablet captures geology, photos, voice notes, and the assay when it returns from the lab. Syncs nightly via camp Starlink.
Bayesian update on surrounding cells overnight. By morning the tablet shows refreshed next-pit recommendations.
Same dashboard your geologist would see. The demo location (Mait–Erigavo basement, Sanaag) is illustrative — chosen because it sits cleanly in Somaliland-controlled territory with documented Precambrian basement, but the actual targeting runs against whatever licensed concession the operator brings. Real AOI defined together after discussion.
The Mait–Erigavo demo data (targets, pits, controls) is synthetic and clearly labelled. Everything else on this page comes from authoritative, free, citable sources. Click any reference layer in the demo map to verify against the original source.
Real mineral occurrences and prospects in Somaliland from USGS MRDS (86), USGS Africa Mineral Sites (7), and Wikidata (1). Each carries a clickable USGS deposit ID URL — verifiable by the buyer's team in seconds.
Source: mrdata.usgs.gov · public domain
Full USGS earthquake catalogue for the Horn of Africa, 1990–2026. Active-tectonics evidence — most M5+ events trace the Gulf of Aden / Afar / Somali plate boundary. Max recorded: M6.2.
Source: USGS FDSN web services · public domain
Major Somaliland settlements (Hargeisa, Berbera, Borama, Erigavo, Mait, Las Anod, etc.) with population data and Wikipedia verification links. Admin regions include the disputed Sool zone.
Source: Wikipedia (CC BY-SA), UN OCHA admin boundaries (CC BY)
Lega Dembi (Ethiopia, 3.6 Moz Au), Tulu Kapi (KEFI, 1.7 Moz), Bisha (Eritrea VMS), Sukari (Centamin, 15 Moz resource), Hassai, Block 14, Yemen basement. Used for transfer-learning the geological prior.
Source: operator NI 43-101 disclosures · public
NOAA's 228 MB global magnetic anomaly grid at 2 arc-minute resolution. Clips to any AOI for regional structural framework. Already in the data layer; the pipeline that turns it into per-cell features ships with the system.
Source: NOAA EMAG2 v3 · public domain
Curated bibliography across Somaliland geology, East African Orogen tectonics, analog deposits, ASTER alteration mapping, machine learning for prospectivity, and conflict-minerals due diligence. All findable on Google Scholar.
Categories: framework, tectonics, analogs, remote sensing, ML, OECD
Sentinel-2 alteration mapping (iron oxide, ferric/ferrous, clay) and ASTER VNIR-SWIR alteration (AlOH, FeOH, MgOH, kaolinite) are documented as runnable GEE / earthaccess pipelines in docs/sentinel2_alteration_methodology.md and docs/aster_alteration_methodology.md. They need GEE_PROJECT_ID and EARTHDATA_USERNAME credentials to run live; the operator provisions these once and the pipeline becomes part of monthly retraining. We don't pretend these layers are visible on the public dashboard — disclosing exactly which features need which credentials is part of honest due-diligence prep.
This is what makes the system different from a static map. Drag the timeline or press play to replay the first 60 days of synthetic operator data. The map shows pits as they were dug; the chart shows the operator's hit rate climbing past the baseline as the model learns from each new assay.
2026-05-20 · synthetic operator timeline (1-day = 1 frame at 0.3s)
Reading the chart: the gold line is the operator's cumulative hit rate on AI-recommended pits. The dashed grey line is the 18% baseline measured from blind-control pits — cells the model would not have recommended. The system is genuinely predictive when the gold line is consistently above the dashed line with low p-value. Honest signal, not a vanity metric.
Three tiers of access. What you see right now is the public marketing tier — coordinates are obfuscated ±1.5 km even on synthetic data, so the model is visible without leaking anything. Real operator data is only ever served behind authenticated access on a separate route.
What you're looking at now. Anyone with the URL can see it.
test / test. For prospective operators evaluating the UX before commercial commitment.
Real operator data. Per-operator tenancy. Provisioned after contract signature.
/operator/*Nine capabilities tuned to a real Somaliland operator's day — not a research pipeline.
Each assay shifts the posterior on a 1.5–3 km Gaussian kernel and re-ranks candidate pits within 10 km. Cycle time from new assay to updated recommendation: under 24 hours.
Recommendations maximise expected gold per dollar — not raw probability. Cost is a model input, calibrated as your pit log grows.
QField + custom pit-log layer. MBTiles cached locally. Nightly camp Starlink sync. Works for weeks without connectivity.
Clan / xeer / diya zones, recent security incidents, surface water (NDWI), paved roads, Berbera distance, Puntland border buffer — all baked into the ranking.
Every target carries an ORS alongside its geology score — LBMA refinery acceptance, OECD Annex II path, customs route, disclosure burden. A 90% geological target with no offtake is ranked below a 65% target that can be sold.
Photograph an outcrop on the tablet. Get back: "Looks like shear-zone iron-oxide alteration. Similar to your pits #027 and #041 (2.3 and 1.8 g/t). Recommend panning 50 m N-NE."
Operator AOI never leaves your tenancy. End-to-end encrypted sync. Audit-trail exports are per-target signed CSVs — share only what the financier needs.
Every recommendation hashed, timestamped, GPS-locked, and chain-linked. The basis of the royalty calculation and your defence in any future audit.
Monthly target lists include low-probability cells the model would not otherwise recommend. Honest validation that the system is genuinely predictive, not confirming pre-existing intuition.
Three tiers: offline tablet in the field, camp gateway with Starlink, and a remote service in UAE that handles model retraining. Designed to work for weeks with zero connectivity and resume cleanly when sync returns.
Somaliland operates without full international recognition. That makes compliance harder than in Sudan or Eritrea, not easier. The 15-item compliance gate is the actual project gate — engineered, not checklisted.
Holder named, area locked to concession/boundary.geojson, expiry tracked. Hard pipeline gate.
UK / UAE FZ / Singapore parent holds licence + contracts. Subsidiary in Hargeisa employs locally.
Entity + UBOs + directors + offtake counterparties + transport providers + key suppliers. Refreshed quarterly.
Workers, suppliers, security providers, transport. Even though Somaliland exposure is low, screening is explicit and on file.
At least one LBMA Good Delivery UAE refinery confirms acceptance of Somaliland-origin gold under OECD Annex II. Required before commercial framing.
Legal opinion on Federal Government challenge risk + mitigation. Disclosure to investors.
Sub-clan polygon controlling AOI documented, dispute history reviewed, xeer / diya protocols on file.
Operator data hosted on Cloudflare with per-operator tenancy. No Somaliland-resident infrastructure that creates a sovereignty argument for either authority.
● = system-enforced ·
● = operator-cleared ·
● = infrastructure-level ·
Full 15-item compliance checklist in your delivered docs/compliance.md.
For the geologist driving daily operations.
Targets ranked by colour (blue → purple → gold → bright gold = strongest). The #1 target pulses. Use the layer panel to toggle targets, pits, blind controls, AOI boundary.
Pop-up shows probability, expected grade range (90% conformal interval), expected cost, depth target, drive distance from camp, offtake readiness score, and the AI's one-sentence reasoning.
Use the cluster optimiser to group 4–8 pits around a single camp position. Pick the cluster that maximises expected gold-per-dollar and minimises drive time.
Selected pits sync to foreman tablets overnight. Each tablet wakes up with its assigned pits, GPS waypoints, expected depth, and notes pre-loaded.
Foreman logs each pit on completion: GPS, depth, geology dropdown, photos, voice note. Tablet is offline; sync happens at camp.
When the lab returns assay results, enter them into the matching pit. Within 24 hours the map refreshes — the cells around your last productive pit get re-ranked upward.
No magic gold detector pitches. Honest framing of where the system helps and where it doesn't.
— Measurable improvement over baseline hit rate, validated against blind controls in your data.
— Cycle time from new assay to refreshed recommendation under 48 hours (target 24).
— Every recommendation auditable end-to-end. Every cell you dig links back to the model artefact and feature stack that produced it.
— Zero confidentiality incidents — operator AOI, pit log, and assays never leave the agreed tenancy.
— Not a magic gold detector. Improves your hit rate over baseline; doesn't find gold for you.
— Learns only as fast as you dig and assay. A small-pit-throughput operator gets a slowly-improving system. Honest.
— Not a substitute for compliance. LBMA refinery acceptance is the binding constraint, not the model.
— Not a substitute for a competent geologist. Force-multiplier, not replacement.
Modest monthly service fee + a transparent royalty on production from pre-registered AI-flagged cells. The royalty mechanism is enforced by the append-only audit trail — you cannot retroactively re-flag a cell after the fact.
Annual contract, monthly billing, single AOI up to 200 km². Exact terms agreed at engagement.
Specific monthly fee and NSR percentage are calibrated to the operator's AOI size, expected throughput, and offtake setup. Discussed in the engagement call, not posted publicly until terms are real.
Pre-engineering legal-feasibility memo from UAE counsel — required before signature — billed separately at cost.
docs/compliance.md is cleared.REPURPOSE_PLAN.md and OPERATOR_VALUE_PLAN.md in the delivery.The software is built and operational. We're seeking a design-partner operator with a real Somaliland MEM licence to deploy against. Bring your licence area for discussion (or describe it), your current process, and your offtake plans. We will tell you honestly whether the system would lift your hit rate and what the binding compliance constraint looks like — including any constraints we can't help with.