Geomatic LabsEST. 2026

Geospatial intelligence, built for decisions you can defend.

Federal data for any US coordinate — every value wrapped in source, vintage, and confidence.

THE SHEET IS REAL — THE US SURVEY OF RECORD, AND BADWATER BASIN CARVED FROM THE USGS 3DEP HEIGHTFIELD LOCUS SERVES. SCROLL TO DIVE · EVERY ELEVATION IS REAL

Sheet 02 — The envelope36.2297° N · 116.7672° W

Every number ships with its receipt.

Other APIs return a value. Locus returns an envelope — the value plus everything you need to stand behind it in a memo, a model, or a diligence file.

FIELD · elevationBADWATER BASIN, CA

−81.857meters

…says who? As of when? How sure are we?

{  "value": -81.857,  "unit": "meters",  "source": "USGS_3DEP",  "source_url": "https://epqs.nationalmap.gov/v1/json",  "dataset_vintage": "5/1/2025",  "fetched_at": "2026-07-04T04:49:59Z",  "ttl_seconds": 31536000,  "confidence": "high",  "nullable": false,  "null_meaning": null,}

01 · THE VALUEA number, alone.

−81.857. Every mapping API stops here — and no credit committee, regulator, or underwriter should accept it.

02 · THE SOURCEWho measured it.

A named federal instrument — USGS 3DEP — and the exact endpoint we hit. An auditor can replay the query, not take our word.

03 · THE VINTAGEWhen it was true.

Data ages. The envelope carries the survey's own date, the moment we fetched it, and how long it may be cached before it must be re-proven.

04 · THE CONFIDENCEHow hard you can lean on it.

Confidence is computed in code and only ever calibrated down. And when a field is legitimately empty, you get null with a stated meaning — never a guess.

THE STANDARD OF PROOF“An LLM said so” is not good enough.

Sheet 03 — Cited answers39.7596° N · 121.6219° W

Ask in plain language. Get an answer with footnotes.

The model plans which fields to fetch — but citations are derived in code from what was actually fetched. It cannot cite what we didn't serve.

POST /v1/ask · PARADISE, CA39.7596, −121.6219

“We're evaluating a rebuild on this parcel. What are the wildfire and flood risks?”

Wildfire is the dominant hazard here. This parcel scores in the 95th percentile nationally^[1] — a primary factor for construction planning, defensible space, and insurance.

Flood risk, by contrast, is low: FEMA Zone X, outside any Special Flood Hazard Area^[2], with riverine flood risk in the 23rd percentile^[1].

wildfire_riskflood_zoneriverine_flood_riskcoastal_flood_risk

CITATIONS — DERIVED IN CODE · CONFIDENCE high

[1] FEMA_NRIservices.arcgis.com/…/National_Risk_Index_Census_Tracts

[2] FEMA_NFHLhazards.fema.gov/…/public/NFHL/MapServer/28

Citations from code, not the model

Sources are read off the returned envelopes after the fetch. The synthesizer writes prose; it never invents a reference.

Confidence only moves down

If any input is stale, missing, or medium-confidence, the answer's confidence is calibrated down. It is never rounded up.

Bounded by the registry

The planner chooses from 137 registered fields. Outside the registry, Locus says so instead of improvising.

Sheet 04 — Screening at scale39.7392° N · 104.9903° W

From one point to a whole portfolio.

Radius queries for exposure and adjacency. Batch fetch for up to 100 sites with per-site isolation — one slow parcel never sinks the run.

POST /v1/within · eia_power_plants · 25 kmDENVER, CO

power plants within 25 km — each with name, fuel, capacity, and distance

Real response. Dots are the actual plants at their true bearings · ● gas/other ● solar ● hydro/biomass

POST /v1/batch/fetch · 4 of ≤100 sitesPER-SITE ISOLATION

Site	Suitability	Climate risk	Wildfire
Downtown Denver 39.7392, −104.9903	90	29	29
Miami Beach 25.7907, −80.1300	52	96	69
Paradise 39.7596, −121.6219	52	95	95
Lower Ninth Ward 29.9614, −90.0136	49	91	0

Composite scores 0–100 with per-input breakdown and cited sources — opinions with the work shown. Real batch response, captured 2026-07-04.

Sheet 05 — MethodADDRESS → ENVELOPE

From address to defensible answer.

Address in

Street address or bare lat/lng. Geocoding runs on the free US Census geocoder, and the resolution is echoed back for the audit trail.

"resolved_from": "1600 Pennsylvania Ave NW…"

Parallel federal fetch

One adapter per source, fanned out in parallel with per-coordinate caching. A slow agency degrades one field, never the request.

Enveloped out

Raw envelopes, transparent composite scores, or cited prose — every value carrying source, vintage, and confidence. Failures arrive labeled and retryable, never silent.

"partial_failures": [] · citations from code

Sheet 06 — Coverage indexCONUS + AK + HI

The registry at a glance.

Fields are data, not code — each one declares its source, cadence, and null semantics in a registry the planner cannot step outside of.

provenance-tagged fields

layers, terrain to demographics

federal & public datasets

metric presets for common jobs

hazards57

parcels14

utilities28

derived8

built_environment5

terrain7

land_cover7

climate4

demographics7

USGSFEMANOAAUS CENSUSEPAEIAFCCUSDAHUDFAAUSACENREL

Sheet 07 — For engineers & agentsREST · SSE · MCP

An API that shows its work.

REQUESTcurl

curl -s https://app.geomaticlabs.com/v1/fetch \
  -H "Authorization: Bearer lk_…" \
  -d '{"lat": 36.2297, "lng": -116.7672,
       "fields": ["elevation", "flood_zone"]}'

RESPONSE · ABRIDGED200

"elevation": {
  "value": -81.857, "unit": "meters",
  "source": "USGS_3DEP", "dataset_vintage": "5/1/2025",
  "confidence": "high", …
}

Streams as it lands

SSE on /v1/fetch/stream and /v1/ask/stream — fields paint the moment each federal source answers.

Agent-native

An MCP server with plus llms.txt — wire Locus into a workflow without scraping docs.

Boringly reliable on purpose

run with the network off; every adapter also carries a live-API test. Flaky federal endpoints degrade softly, never fatally.

OpenAPI, keys, limits

Interactive docs at /docs, per-user lk_ API keys, honest rate limits.